Skip to content

Hypergraph is more powerful. Transform unstructured text into structured knowledge with LLMs. Graphs, hypergraphs, and spatio-temporal extractions — with one command.

feat: 添加通用知识提取模板并重构模块结构

yifanfeng971d39fd71fe38
diff --git a/hyperextract/templates/en/__init__.py b/hyperextract/templates/en/__init__.py
index 10dac70..5f14636 100644
--- a/hyperextract/templates/en/__init__.py
+++ b/hyperextract/templates/en/__init__.py
@@ -1,4 +1,35 @@
-# from .general import KnowledgeGraph, ConceptMap, LogicGraph
+from .general import (
+ # Knowledge Graph
+ KnowledgeGraph,
+ # Entity Registry
+ EntityRegistry,
+ # Keyword List
+ KeywordList,
+ # Encyclopedia Item
+ EncyclopediaItem,
+ # Concept Hierarchy
+ ConceptHierarchy,
+ # Cross Reference Network
+ CrossReferenceNet,
+ # Personal Profile
+ PersonalProfile,
+ # Social Network
+ SocialNetwork,
+ # Life Event Timeline
+ LifeEventTimeline,
+ # Regulation Profile
+ RegulationProfile,
+ # Clause List
+ ClauseList,
+ # Penalty Registry
+ PenaltyRegistry,
+ # Operational Procedure
+ OperationalProcedure,
+ # Penalty Mapping
+ PenaltyMapping,
+ # Compliance Logic
+ ComplianceLogic,
+)
from .agriculture import CropCycleGraph, LivestockGraph, AgriPestControl
from .biology import TaxonomyGraph, BiologicalNetwork, MetabolicHypergraph
from .finance import (
@@ -31,18 +62,37 @@ from .medicine import DrugInteractionGraph, ClinicalTreatmentTimeline, Diagnosti
from .news import BreakingEventGraph, TopicTimeline, OpinionSet
__all__ = [
+ # general
+ ## Knowledge Graph
"KnowledgeGraph",
- "ConceptMap",
- "LogicGraph",
- "CropCycleGraph",
- "LivestockGraph",
- "AgriPestControl",
- "TaxonomyGraph",
- "BiologicalNetwork",
- "MetabolicHypergraph",
- "MolecularStructureGraph",
- "ChemicalReactionHyper",
- "LabProtocolTemporal",
+ ## Entity Registry
+ "EntityRegistry",
+ ## Keyword List
+ "KeywordList",
+ ## Encyclopedia Item
+ "EncyclopediaItem",
+ ## Concept Hierarchy
+ "ConceptHierarchy",
+ ## Cross Reference Network
+ "CrossReferenceNet",
+ ## Personal Profile
+ "PersonalProfile",
+ ## Social Network
+ "SocialNetwork",
+ ## Life Event Timeline
+ "LifeEventTimeline",
+ ## Regulation Profile
+ "RegulationProfile",
+ ## Clause List
+ "ClauseList",
+ ## Penalty Registry
+ "PenaltyRegistry",
+ ## Operational Procedure
+ "OperationalProcedure",
+ ## Penalty Mapping
+ "PenaltyMapping",
+ ## Compliance Logic
+ "ComplianceLogic",
# Finance
"RiskAssessmentGraph",
"MarketSentimentGraph",
@@ -81,4 +131,15 @@ __all__ = [
"BreakingEventGraph",
"TopicTimeline",
"OpinionSet",
+ # Agriculture
+ "CropCycleGraph",
+ "LivestockGraph",
+ "AgriPestControl",
+ # Biology
+ "TaxonomyGraph",
+ "BiologicalNetwork",
+ "MetabolicHypergraph",
+ "MolecularStructureGraph",
+ "ChemicalReactionHyper",
+ "LabProtocolTemporal",
]
diff --git a/hyperextract/templates/en/general/__init__.py b/hyperextract/templates/en/general/__init__.py
index e69de29..d7304a5 100644
--- a/hyperextract/templates/en/general/__init__.py
+++ b/hyperextract/templates/en/general/__init__.py
@@ -0,0 +1,48 @@
+from .knowledge_graph import KnowledgeGraph
+from .entity_registry import EntityRegistry
+from .keyword_list import KeywordList
+from .encyclopedia_item import EncyclopediaItem
+from .concept_hierarchy import ConceptHierarchy
+from .cross_reference_net import CrossReferenceNet
+from .personal_profile import PersonalProfile
+from .social_network import SocialNetwork
+from .life_event_timeline import LifeEventTimeline
+from .regulation_profile import RegulationProfile
+from .clause_list import ClauseList
+from .penalty_registry import PenaltyRegistry
+from .operational_procedure import OperationalProcedure
+from .penalty_mapping import PenaltyMapping
+from .compliance_logic import ComplianceLogic
+
+__all__ = [
+ # Knowledge Graph
+ "KnowledgeGraph",
+ # Entity Registry
+ "EntityRegistry",
+ # Keyword List
+ "KeywordList",
+ # Encyclopedia Item
+ "EncyclopediaItem",
+ # Concept Hierarchy
+ "ConceptHierarchy",
+ # Cross Reference Network
+ "CrossReferenceNet",
+ # Personal Profile
+ "PersonalProfile",
+ # Social Network
+ "SocialNetwork",
+ # Life Event Timeline
+ "LifeEventTimeline",
+ # Regulation Profile
+ "RegulationProfile",
+ # Clause List
+ "ClauseList",
+ # Penalty Registry
+ "PenaltyRegistry",
+ # Operational Procedure
+ "OperationalProcedure",
+ # Penalty Mapping
+ "PenaltyMapping",
+ # Compliance Logic
+ "ComplianceLogic"
+]
diff --git a/hyperextract/templates/en/general/clause_list.py b/hyperextract/templates/en/general/clause_list.py
new file mode 100644
index 0000000..6178318
--- /dev/null
+++ b/hyperextract/templates/en/general/clause_list.py
@@ -0,0 +1,104 @@
+"""Clause List - Break down regulations into atomic clauses for quick indexing.
+
+Suitable for clause retrieval, full-text comparison, etc.
+"""
+
+from typing import Any
+from pydantic import BaseModel, Field
+from langchain_core.language_models import BaseChatModel
+from langchain_core.embeddings import Embeddings
+from hyperextract.types import AutoList
+
+
+class Clause(BaseModel):
+ """Regulation clause"""
+ clauseId: str = Field(description="Clause ID, e.g., Article 1, Section 2.1")
+ title: str = Field(description="Clause title/subject", default="")
+ content: str = Field(description="Clause content")
+ category: str = Field(description="Clause category: GeneralProvisions, RightsAndObligations, Procedures, Penalties, SupplementaryProvisions, Other", default="Other")
+
+
+_PROMPT = """You are a professional regulation clause breakdown expert. Please break down the text into atomic clauses.
+
+### Extraction Rules
+1. Break down the regulation into independent clauses according to its natural structure
+2. Assign an ID to each clause (e.g., Article 1, Section 2.1, etc.)
+3. Extract the content of each clause
+4. Add an appropriate title or subject to the clause (if available)
+5. Assign a category to each clause: GeneralProvisions, RightsAndObligations, Procedures, Penalties, SupplementaryProvisions, Other
+
+### Constraints
+- Keep clause content consistent with the original text
+- Do not omit any clauses
+- Maintain the integrity of clauses
+
+### Source text:
+"""
+
+
+class ClauseList(AutoList[Clause]):
+ """
+ Applicable documents: Company internal management systems, administrative regulations, operation manuals, compliance guidelines
+
+ Function introduction:
+ Break down regulations into atomic clauses for quick indexing. Suitable for clause retrieval, full-text comparison, etc.
+
+ Example:
+ >>> template = ClauseList(llm_client=llm, embedder=embedder)
+ >>> template.feed_text("Universe First Slacker Company Employee Attendance Management System...")
+ >>> for clause in template:
+ ... print(f"{clause.clauseId}: {clause.content}")
+ """
+
+ def __init__(
+ self,
+ llm_client: BaseChatModel,
+ embedder: Embeddings,
+ *,
+ max_workers: int = 10,
+ verbose: bool = False,
+ **kwargs: Any,
+ ):
+ """
+ Initialize clause list template.
+
+ Args:
+ llm_client: LLM client for knowledge extraction
+ embedder: Embedding model for semantic search
+ max_workers: Maximum number of worker threads, default: 10
+ verbose: Whether to output detailed logs, default: False
+ **kwargs: Other technical parameters, passed to base class
+ """
+ super().__init__(
+ item_schema=Clause,
+ llm_client=llm_client,
+ embedder=embedder,
+ prompt=_PROMPT,
+ max_workers=max_workers,
+ verbose=verbose,
+ **kwargs,
+ )
+
+ def show(
+ self,
+ *,
+ top_k_for_search: int = 3,
+ top_k_for_chat: int = 3,
+ ):
+ """
+ Display clause list.
+
+ Args:
+ top_k_for_search: Number of clauses to return for semantic search, default: 3
+ top_k_for_chat: Number of clauses to use for chat, default: 3
+ """
+ def itemLabelExtractor(item: Clause) -> str:
+ if item.title:
+ return f"{item.clauseId}: {item.title}"
+ return f"{item.clauseId}"
+
+ super().show(
+ item_label_extractor=itemLabelExtractor,
+ top_k_for_search=top_k_for_search,
+ top_k_for_chat=top_k_for_chat,
+ )
diff --git a/hyperextract/templates/en/general/compliance_logic.py b/hyperextract/templates/en/general/compliance_logic.py
new file mode 100644
index 0000000..057cb6d
--- /dev/null
+++ b/hyperextract/templates/en/general/compliance_logic.py
@@ -0,0 +1,156 @@
+"""Compliance Logic - Model complex logic of "under what conditions, which subjects must/may not perform which actions".
+
+Suitable for compliance auditing, risk identification, etc.
+"""
+
+from typing import Any, List
+from pydantic import BaseModel, Field
+from langchain_core.language_models import BaseChatModel
+from langchain_core.embeddings import Embeddings
+from hyperextract.types import AutoHypergraph
+
+
+class ComplianceElement(BaseModel):
+ """Compliance Element - Node"""
+ name: str = Field(description="Element name")
+ type: str = Field(description="Element type: Subject, Action, Condition, Other")
+ description: str = Field(description="Brief description", default="")
+
+
+class ComplianceRule(BaseModel):
+ """Compliance Rule - Hyperedge"""
+ rule_type: str = Field(description="Rule type: Must, MustNot, May, Other")
+ action: str = Field(description="Required or prohibited action content")
+ condition: str = Field(description="Applicable conditions/prerequisites", default="")
+ penalty: str = Field(description="Violation consequences/penalties", default="")
+ source_clause: str = Field(description="Source clause of the rule", default="")
+ notes: str = Field(description="Additional notes", default="")
+ participants: List[str] = Field(description="List of participating element names")
+
+
+_PROMPT = """You are a professional compliance analysis expert. Please extract compliance elements and rules from the text, modeling the logic of "under what conditions, which subjects must/may not perform which actions" to build a compliance behavior hypergraph.
+
+### Node Extraction Rules
+1. Extract elements involved in rules as nodes
+2. Assign type to each element: Subject, Action, Condition, Other
+3. Add brief description for each element
+
+### Hyperedge Extraction Rules
+1. Only create compliance rule hyperedges from extracted elements
+2. Assign rule type: Must, MustNot, May, Other
+3. Fill in required or prohibited action content (required)
+4. Extract applicable conditions/prerequisites (if available)
+5. Extract violation consequences/penalties (if available)
+6. Extract source clause (if available)
+7. Add additional notes (if available)
+8. List all participating element names (required)
+
+### Constraints
+- Ensure rule logic accuracy
+- Maintain objectivity and accuracy, do not add information not in the text
+- Each hyperedge must contain rule_type, action, and participants
+- Try to fill other fields, use empty string for missing fields
+
+### Source text:
+"""
+
+_NODE_PROMPT = """You are a professional compliance element recognition expert. Please extract all compliance-related elements as nodes from the text.
+
+### Extraction Rules
+1. Extract elements involved in rules
+2. Assign type to each element: Subject, Action, Condition, Other
+3. Add brief description for each element
+
+### Source text:
+"""
+
+_EDGE_PROMPT = """You are a professional compliance rule extraction expert. Please extract compliance rule hyperedges from the given node (element) list.
+
+### Constraints
+1. Only extract rules (hyperedges) from the known element list below
+2. Do not create unlisted elements
+3. Each rule must contain rule_type, action, and participants
+4. Try to fill other fields, use empty string for missing fields
+
+"""
+
+
+class ComplianceLogic(AutoHypergraph[ComplianceElement, ComplianceRule]):
+ """
+ Applicable documents: Compliance guidelines, administrative regulations, company management systems
+
+ Function introduction:
+ Model complex logic of "under what conditions, which subjects must/may not perform which actions". Suitable for compliance auditing, risk identification, etc.
+
+ Example:
+ >>> template = ComplianceLogic(llm_client=llm, embedder=embedder)
+ >>> template.feed_text("Universe First Slacker Company Compliance Manual...")
+ >>> template.show()
+ """
+
+ def __init__(
+ self,
+ llm_client: BaseChatModel,
+ embedder: Embeddings,
+ *,
+ extraction_mode: str = "two_stage",
+ max_workers: int = 10,
+ verbose: bool = False,
+ **kwargs: Any,
+ ):
+ """
+ Initialize compliance logic hypergraph template.
+
+ Args:
+ llm_client: LLM client for knowledge extraction
+ embedder: Embedding model for semantic search
+ extraction_mode: Extraction mode, either "one_stage" (extract nodes and edges simultaneously)
+ or "two_stage" (extract nodes first, then edges), default: "two_stage"
+ max_workers: Maximum number of worker threads, default: 10
+ verbose: Whether to output detailed logs, default: False
+ **kwargs: Other technical parameters, passed to base class
+ """
+ super().__init__(
+ node_schema=ComplianceElement,
+ edge_schema=ComplianceRule,
+ node_key_extractor=lambda x: x.name,
+ edge_key_extractor=lambda x: f"{sorted(x.participants)}_{x.rule_type}_{x.action}",
+ nodes_in_edge_extractor=lambda x: tuple(x.participants),
+ llm_client=llm_client,
+ embedder=embedder,
+ extraction_mode=extraction_mode,
+ max_workers=max_workers,
+ verbose=verbose,
+ prompt=_PROMPT,
+ prompt_for_node_extraction=_NODE_PROMPT,
+ prompt_for_edge_extraction=_EDGE_PROMPT,
+ **kwargs,
+ )
+
+ def show(
+ self,
+ *,
+ top_k_for_search: int = 3,
+ top_k_for_chat: int = 3,
+ ):
+ """
+ Display compliance logic hypergraph.
+
+ Args:
+ top_k_for_search: Number of nodes/edges to return for semantic search, default: 3
+ top_k_for_chat: Number of nodes/edges to use for chat, default: 3
+ """
+ def node_label_extractor(node: ComplianceElement) -> str:
+ return f"{node.name} ({node.type})"
+
+ def edge_label_extractor(edge: ComplianceRule) -> str:
+ return f"[{edge.rule_type}] {edge.action}"
+
+ super().show(
+ node_label_extractor=node_label_extractor,
+ edge_label_extractor=edge_label_extractor,
+ top_k_nodes_for_search=top_k_for_search,
+ top_k_edges_for_search=top_k_for_search,
+ top_k_nodes_for_chat=top_k_for_chat,
+ top_k_edges_for_chat=top_k_for_chat,
+ )
diff --git a/hyperextract/templates/en/general/concept_hierarchy.py b/hyperextract/templates/en/general/concept_hierarchy.py
new file mode 100644
index 0000000..dfefacc
--- /dev/null
+++ b/hyperextract/templates/en/general/concept_hierarchy.py
@@ -0,0 +1,156 @@
+"""Concept Hierarchy - Build subclass relationships (Subclass-Of) or composition relationships (Part-Of).
+
+Suitable for scientific disciplines, textbook knowledge points, etc.
+"""
+
+from typing import Any, Optional
+from pydantic import BaseModel, Field
+from langchain_core.language_models import BaseChatModel
+from langchain_core.embeddings import Embeddings
+from hyperextract.types import AutoGraph
+
+
+class ConceptNode(BaseModel):
+ """Concept node"""
+ name: str = Field(description="Concept name")
+ category: str = Field(description="Concept type: CoreConcept, Subconcept, Attribute, Instance, Other")
+ description: str = Field(description="Brief concept description", default="")
+
+
+class HierarchyRelation(BaseModel):
+ """Hierarchy relationship edge"""
+ source: str = Field(description="Parent concept/whole concept")
+ target: str = Field(description="Subconcept/part concept")
+ relationType: str = Field(description="Relationship type: Subclass-Of (taxonomy), Part-Of (composition), Instance-Of (instance), Other")
+ details: str = Field(description="Detailed relationship explanation", default="")
+
+
+_PROMPT = """You are a professional concept structure analysis expert. Please extract concepts and their hierarchical relationships from the text to build a concept hierarchy graph.
+
+### Node Extraction Rules
+1. Extract all concepts: CoreConcept, Subconcept, Attribute, Instance, etc.
+2. Assign a type to each concept: CoreConcept, Subconcept, Attribute, Instance, Other
+3. Add a brief description for each concept
+
+### Edge Extraction Rules
+1. Only create edges from extracted concepts
+2. Relationship types include:
+ - Subclass-Of: Taxonomy relationship (e.g., "Dog" is a subclass of "Animal")
+ - Part-Of: Composition relationship (e.g., "Tire" is part of "Car")
+ - Instance-Of: Instance relationship (e.g., "Zu Chongzhi" is an instance of "Mathematician")
+ - Other: Other relationships
+3. Each edge must connect extracted nodes
+
+### Constraints
+- Maintain clear hierarchical relationships, avoid circular dependencies
+- Do not create concepts or relationships not mentioned in the text
+- Maintain objectivity and accuracy, do not add information not in the text
+
+### Source text:
+"""
+
+_NODE_PROMPT = """You are a professional concept recognition expert. Please extract all concepts as nodes from the text.
+
+### Extraction Rules
+1. Extract all concepts: CoreConcept, Subconcept, Attribute, Instance, etc.
+2. Assign a type to each concept: CoreConcept, Subconcept, Attribute, Instance, Other
+3. Add a brief description for each concept
+
+### Source text:
+"""
+
+_EDGE_PROMPT = """You are a professional hierarchical relationship extraction expert. Please extract hierarchical relationships between concepts (nodes) from the given concept list.
+
+### Relationship Type Explanation
+- Subclass-Of: Taxonomy relationship (subclass and superclass)
+- Part-Of: Composition relationship (part and whole)
+- Instance-Of: Instance relationship (instance and class)
+- Other: Other relationships
+
+### Constraints
+1. Only extract edges from the known concept list below
+2. Do not create unlisted concepts
+3. Maintain clear hierarchical relationships, avoid circular dependencies
+
+"""
+
+
+class ConceptHierarchy(AutoGraph[ConceptNode, HierarchyRelation]):
+ """
+ Applicable documents: Science textbooks, knowledge base documents, classification system descriptions
+
+ Function introduction:
+ Build subclass relationships (Subclass-Of) or composition relationships (Part-Of). Suitable for scientific disciplines, textbook knowledge points, etc.
+
+ Example:
+ >>> template = ConceptHierarchy(llm_client=llm, embedder=embedder)
+ >>> template.feed_text("Machine learning is a branch of artificial intelligence...")
+ >>> template.show()
+ """
+
+ def __init__(
+ self,
+ llm_client: BaseChatModel,
+ embedder: Embeddings,
+ *,
+ extraction_mode: str = "two_stage",
+ max_workers: int = 10,
+ verbose: bool = False,
+ **kwargs: Any,
+ ):
+ """
+ Initialize concept hierarchy graph template.
+
+ Args:
+ llm_client: LLM client for knowledge extraction
+ embedder: Embedding model for semantic search
+ extraction_mode: Extraction mode, either "one_stage" (extract nodes and edges simultaneously)
+ or "two_stage" (extract nodes first, then edges), default: "two_stage"
+ max_workers: Maximum number of worker threads, default: 10
+ verbose: Whether to output detailed logs, default: False
+ **kwargs: Other technical parameters, passed to base class
+ """
+ super().__init__(
+ node_schema=ConceptNode,
+ edge_schema=HierarchyRelation,
+ node_key_extractor=lambda x: x.name,
+ edge_key_extractor=lambda x: f"{x.source}|{x.relationType}|{x.target}",
+ nodes_in_edge_extractor=lambda x: (x.source, x.target),
+ llm_client=llm_client,
+ embedder=embedder,
+ extraction_mode=extraction_mode,
+ max_workers=max_workers,
+ verbose=verbose,
+ prompt=_PROMPT,
+ prompt_for_node_extraction=_NODE_PROMPT,
+ prompt_for_edge_extraction=_EDGE_PROMPT,
+ **kwargs,
+ )
+
+ def show(
+ self,
+ *,
+ top_k_for_search: int = 3,
+ top_k_for_chat: int = 3,
+ ):
+ """
+ Display concept hierarchy graph.
+
+ Args:
+ top_k_for_search: Number of nodes/edges to return for semantic search, default: 3
+ top_k_for_chat: Number of nodes/edges to use for chat, default: 3
+ """
+ def node_label_extractor(node: ConceptNode) -> str:
+ return f"{node.name} ({node.category})"
+
+ def edge_label_extractor(edge: HierarchyRelation) -> str:
+ return edge.relationType
+
+ super().show(
+ node_label_extractor=node_label_extractor,
+ edge_label_extractor=edge_label_extractor,
+ top_k_nodes_for_search=top_k_for_search,
+ top_k_edges_for_search=top_k_for_search,
+ top_k_nodes_for_chat=top_k_for_chat,
+ top_k_edges_for_chat=top_k_for_chat,
+ )
diff --git a/hyperextract/templates/en/general/cross_reference_net.py b/hyperextract/templates/en/general/cross_reference_net.py
new file mode 100644
index 0000000..8986a3c
--- /dev/null
+++ b/hyperextract/templates/en/general/cross_reference_net.py
@@ -0,0 +1,157 @@
+"""Cross Reference Network - Map hyperlinks or mutual reference relationships between concepts.
+
+Suitable for encyclopedia internal links, cross-entry association graphs, etc.
+"""
+
+from typing import Any, Optional
+from pydantic import BaseModel, Field
+from langchain_core.language_models import BaseChatModel
+from langchain_core.embeddings import Embeddings
+from hyperextract.types import AutoGraph
+
+
+class ReferenceNode(BaseModel):
+ """Reference node"""
+ name: str = Field(description="Concept/entry name")
+ type: str = Field(description="Node type: Entry, Section, Concept, Person, Location, Other")
+ description: str = Field(description="Brief description", default="")
+
+
+class ReferenceRelation(BaseModel):
+ """Reference relationship edge"""
+ source: str = Field(description="Source entry/concept")
+ target: str = Field(description="Referenced entry/concept")
+ relationType: str = Field(description="Relationship type: Hyperlink, Reference, SeeAlso, Related, Other")
+ context: str = Field(description="Reference context description", default="")
+
+
+_PROMPT = """You are a professional knowledge network analysis expert. Please extract concepts/entries and their mutual reference relationships from the text to build a cross-reference network.
+
+### Node Extraction Rules
+1. Extract all mentioned concepts or entries
+2. Assign a type to each node: Entry, Section, Concept, Person, Location, Other
+3. Add a brief description for each node
+
+### Edge Extraction Rules
+1. Only create edges from extracted nodes
+2. Relationship types include:
+ - Hyperlink: Hyperlink relationship
+ - Reference: Reference relationship
+ - SeeAlso: See also relationship
+ - Related: Related relationship
+ - Other: Other relationships
+3. Record reference context description
+
+### Constraints
+- Each edge must connect extracted nodes
+- Do not create nodes or relationships not mentioned in the text
+- Maintain objectivity and accuracy, do not add information not in the text
+
+### Source text:
+"""
+
+_NODE_PROMPT = """You are a professional entry recognition expert. Please extract all concepts or entries as nodes from the text.
+
+### Extraction Rules
+1. Extract all mentioned concepts or entries
+2. Assign a type to each node: Entry, Section, Concept, Person, Location, Other
+3. Add a brief description for each node
+
+### Source text:
+"""
+
+_EDGE_PROMPT = """You are a professional reference relationship extraction expert. Please extract mutual reference relationships between nodes from the given node list.
+
+### Relationship Type Explanation
+- Hyperlink: Hyperlink relationship
+- Reference: Reference relationship
+- SeeAlso: See also relationship
+- Related: Related relationship
+- Other: Other relationships
+
+### Constraints
+1. Only extract edges from the known node list below
+2. Do not create unlisted nodes
+
+"""
+
+
+class CrossReferenceNet(AutoGraph[ReferenceNode, ReferenceRelation]):
+ """
+ Applicable documents: Wikipedia entries, knowledge base documents, web content with internal links
+
+ Function introduction:
+ Map hyperlinks or mutual reference relationships between concepts. Suitable for encyclopedia internal links, cross-entry association graphs, etc.
+
+ Example:
+ >>> template = CrossReferenceNet(llm_client=llm, embedder=embedder)
+ >>> template.feed_text("Machine Learning is a branch of Artificial Intelligence...")
+ >>> template.show()
+ """
+
+ def __init__(
+ self,
+ llm_client: BaseChatModel,
+ embedder: Embeddings,
+ *,
+ extraction_mode: str = "two_stage",
+ max_workers: int = 10,
+ verbose: bool = False,
+ **kwargs: Any,
+ ):
+ """
+ Initialize cross-reference network template.
+
+ Args:
+ llm_client: LLM client for knowledge extraction
+ embedder: Embedding model for semantic search
+ extraction_mode: Extraction mode, either "one_stage" (extract nodes and edges simultaneously)
+ or "two_stage" (extract nodes first, then edges), default: "two_stage"
+ max_workers: Maximum number of worker threads, default: 10
+ verbose: Whether to output detailed logs, default: False
+ **kwargs: Other technical parameters, passed to base class
+ """
+ super().__init__(
+ node_schema=ReferenceNode,
+ edge_schema=ReferenceRelation,
+ node_key_extractor=lambda x: x.name,
+ edge_key_extractor=lambda x: f"{x.source}|{x.relationType}|{x.target}",
+ nodes_in_edge_extractor=lambda x: (x.source, x.target),
+ llm_client=llm_client,
+ embedder=embedder,
+ extraction_mode=extraction_mode,
+ max_workers=max_workers,
+ verbose=verbose,
+ prompt=_PROMPT,
+ prompt_for_node_extraction=_NODE_PROMPT,
+ prompt_for_edge_extraction=_EDGE_PROMPT,
+ **kwargs,
+ )
+
+ def show(
+ self,
+ *,
+ top_k_for_search: int = 3,
+ top_k_for_chat: int = 3,
+ ):
+ """
+ Display cross-reference network.
+
+ Args:
+ top_k_for_search: Number of nodes/edges to return for semantic search, default: 3
+ top_k_for_chat: Number of nodes/edges to use for chat, default: 3
+ """
+ def node_label_extractor(node: ReferenceNode) -> str:
+ return f"{node.name} ({node.type})"
+
+ def edge_label_extractor(edge: ReferenceRelation) -> str:
+ return edge.relationType
+
+ super().show(
+ node_label_extractor=node_label_extractor,
+ edge_label_extractor=edge_label_extractor,
+ top_k_nodes_for_search=top_k_for_search,
+ top_k_edges_for_search=top_k_for_search,
+ top_k_nodes_for_chat=top_k_for_chat,
+ top_k_edges_for_chat=top_k_for_chat,
+ )
diff --git a/hyperextract/templates/en/general/encyclopedia_item.py b/hyperextract/templates/en/general/encyclopedia_item.py
new file mode 100644
index 0000000..ff684ee
--- /dev/null
+++ b/hyperextract/templates/en/general/encyclopedia_item.py
@@ -0,0 +1,106 @@
+"""Encyclopedia Item - Extract structured attribute information for a single subject (similar to an Infobox).
+
+Suitable for Wikipedia, professional dictionary entries, etc.
+"""
+
+from typing import Any, Optional, List
+from pydantic import BaseModel, Field
+from langchain_core.language_models import BaseChatModel
+from langchain_core.embeddings import Embeddings
+from hyperextract.types import AutoModel
+
+
+class EncyclopediaInfo(BaseModel):
+ """Encyclopedia item infobox"""
+ title: str = Field(description="Entry title")
+ entityType: str = Field(description="Entity type: Person, Location, Organization, Event, Concept, Item, Other")
+ alternativeNames: List[str] = Field(description="Aliases, common names, foreign names", default_factory=list)
+ category: List[str] = Field(description="Category tags", default_factory=list)
+ summary: str = Field(description="Abstract/introduction", default="")
+ attributes: List[str] = Field(description="Key attribute list, each element in format 'AttributeName: AttributeValue'", default_factory=list)
+ keyEvents: List[str] = Field(description="Important events/milestones", default_factory=list)
+ relatedConcepts: List[str] = Field(description="Related concepts/entries", default_factory=list)
+
+
+_PROMPT = """You are a professional encyclopedia editor. Please extract structured attribute information for a single subject from the text to build an encyclopedia item infobox.
+
+### Extraction Rules
+1. Identify the core subject of the text (Person, Location, Organization, Event, Concept, etc.)
+2. Assign a type to the subject: Person, Location, Organization, Event, Concept, Item, Other
+3. Extract all aliases, common names, or foreign names
+4. Extract category tags (e.g., "Scientist", "Mathematics", "Ancient China")
+5. Write a 100-200 word summary
+6. Extract key attributes, each in format 'AttributeName: AttributeValue' (e.g., 'Date of Birth: 429', 'Nationality: China')
+7. Extract important events or milestones
+8. Extract related concepts or entries
+
+### Constraints
+- Only extract information explicitly mentioned in the text, do not add additional content
+- Keep attribute names concise and clear
+- Maintain objectivity and accuracy, consistent with encyclopedia style
+
+### Source text:
+"""
+
+
+class EncyclopediaItem(AutoModel[EncyclopediaInfo]):
+ """
+ Applicable documents: Wikipedia entries, encyclopedia entries, professional dictionary entries
+
+ Function introduction:
+ Extract structured attribute information for a single subject (similar to an Infobox). Suitable for Wikipedia, professional dictionary entries, etc.
+
+ Example:
+ >>> template = EncyclopediaItem(llm_client=llm, embedder=embedder)
+ >>> template.feed_text("Zu Chongzhi (429-500), courtesy name Wenyuan, was from Qiuxian, Fanyang...")
+ >>> print(template.data.title)
+ >>> print(template.data.summary)
+ """
+
+ def __init__(
+ self,
+ llm_client: BaseChatModel,
+ embedder: Embeddings,
+ *,
+ max_workers: int = 10,
+ verbose: bool = False,
+ **kwargs: Any,
+ ):
+ """
+ Initialize encyclopedia item template.
+
+ Args:
+ llm_client: LLM client for knowledge extraction
+ embedder: Embedding model for semantic search
+ max_workers: Maximum number of worker threads, default: 10
+ verbose: Whether to output detailed logs, default: False
+ **kwargs: Other technical parameters, passed to base class
+ """
+ super().__init__(
+ data_schema=EncyclopediaInfo,
+ llm_client=llm_client,
+ embedder=embedder,
+ prompt=_PROMPT,
+ max_workers=max_workers,
+ verbose=verbose,
+ **kwargs,
+ )
+
+ def show(
+ self,
+ *,
+ top_k: int = 3,
+ ):
+ """
+ Display encyclopedia item.
+
+ Args:
+ top_k: Number of items to use for chat, default: 3
+ """
+ def label_extractor(item: EncyclopediaInfo) -> str:
+ return f"{item.title} ({item.entityType})"
+
+ super().show(
+ label_extractor=label_extractor,
+ top_k=top_k,
+ )
diff --git a/hyperextract/templates/en/general/entity_registry.py b/hyperextract/templates/en/general/entity_registry.py
new file mode 100644
index 0000000..3efad83
--- /dev/null
+++ b/hyperextract/templates/en/general/entity_registry.py
@@ -0,0 +1,100 @@
+"""Entity Registry - Deduplicate and summarize all unique entities appearing in the text.
+
+Suitable for entity discovery, named entity recognition, etc.
+"""
+
+from typing import Any
+from pydantic import BaseModel, Field
+from langchain_core.language_models import BaseChatModel
+from langchain_core.embeddings import Embeddings
+from hyperextract.types import AutoSet
+
+
+class RegistryEntry(BaseModel):
+ """Registry entry"""
+ name: str = Field(description="Entity name, consistent with original text")
+ category: str = Field(description="Entity type: Person, Organization, Location, Product, Concept, Other")
+ description: str = Field(description="Brief description", default="")
+
+
+_PROMPT = """You are a professional entity recognition expert. Please extract all unique entities from the text to form an entity registry.
+
+### Extraction Rules
+1. Extract all entities: Person, Organization, Location, Product, Concept, etc.
+2. Assign a type to each entity: Person, Organization, Location, Product, Concept, Other
+3. Keep entity names consistent with the original text
+4. Add a brief description for each entity
+
+### Constraints
+- Deduplication: Only keep one record for the same entity
+- Maintain objectivity and accuracy, do not add information not in the text
+- Entity name case should be consistent with the original text
+
+### Source text:
+"""
+
+
+class EntityRegistry(AutoSet[RegistryEntry]):
+ """
+ Applicable documents: Arbitrary text, web scraped content, news reports
+
+ Function introduction:
+ Deduplicate and summarize all unique entities (e.g., person names, organization names) appearing in the text to form an entity registry. Supports entity discovery and named entity recognition.
+
+ Example:
+ >>> template = EntityRegistry(llm_client=llm, embedder=embedder)
+ >>> template.feed_text("Galaxy Interstellar announces successful maiden flight of Shenzhou-50...")
+ >>> print(list(template))
+ """
+
+ def __init__(
+ self,
+ llm_client: BaseChatModel,
+ embedder: Embeddings,
+ *,
+ max_workers: int = 10,
+ verbose: bool = False,
+ **kwargs: Any,
+ ):
+ """
+ Initialize entity registry template.
+
+ Args:
+ llm_client: LLM client for knowledge extraction
+ embedder: Embedding model for semantic search
+ max_workers: Maximum number of worker threads, default: 10
+ verbose: Whether to output detailed logs, default: False
+ **kwargs: Other technical parameters, passed to base class
+ """
+ super().__init__(
+ item_schema=RegistryEntry,
+ key_extractor=lambda x: x.name,
+ llm_client=llm_client,
+ embedder=embedder,
+ prompt=_PROMPT,
+ max_workers=max_workers,
+ verbose=verbose,
+ **kwargs,
+ )
+
+ def show(
+ self,
+ *,
+ top_k_for_search: int = 3,
+ top_k_for_chat: int = 3,
+ ):
+ """
+ Display entity registry.
+
+ Args:
+ top_k_for_search: Number of items to return for semantic search, default: 3
+ top_k_for_chat: Number of items to use for chat, default: 3
+ """
+ def itemLabelExtractor(item: RegistryEntry) -> str:
+ return f"{item.name} ({item.category})"
+
+ super().show(
+ item_label_extractor=itemLabelExtractor,
+ top_k_for_search=top_k_for_search,
+ top_k_for_chat=top_k_for_chat,
+ )
diff --git a/hyperextract/templates/en/general/keyword_list.py b/hyperextract/templates/en/general/keyword_list.py
new file mode 100644
index 0000000..71f712f
--- /dev/null
+++ b/hyperextract/templates/en/general/keyword_list.py
@@ -0,0 +1,98 @@
+"""Keyword List - Extract keywords from text with category and description information.
+
+Suitable for content tag generation, topic analysis, etc.
+"""
+
+from typing import Any
+from pydantic import BaseModel, Field
+from langchain_core.language_models import BaseChatModel
+from langchain_core.embeddings import Embeddings
+from hyperextract.types import AutoList
+
+
+class Keyword(BaseModel):
+ """Keyword entry"""
+ term: str = Field(description="Keyword")
+ category: str = Field(description="Category: CoreConcept, Term, PersonName, OrganizationName, Other")
+ description: str = Field(description="Brief description or context explanation of the keyword", default="")
+
+
+_PROMPT = """You are a professional keyword extraction expert. Please extract all keywords from the text with category and description information.
+
+### Extraction Rules
+1. Extract all keywords and core phrases from the text
+2. Assign a category to each keyword: CoreConcept, Term, PersonName, OrganizationName, Other
+3. Add a brief description for each keyword, explaining its meaning or context in the text
+
+### Constraints
+- Ensure keyword accuracy, do not add information not in the text
+- Keep keyword case consistent with the original text
+
+### Source text:
+"""
+
+
+class KeywordList(AutoList[Keyword]):
+ """
+ Applicable documents: Arbitrary text, blog articles, academic papers
+
+ Function introduction:
+ Extract keywords from text with category and description information. Suitable for content tag generation and topic analysis.
+
+ Example:
+ >>> template = KeywordList(llm_client=llm, embedder=embedder)
+ >>> template.feed_text("Galaxy Interstellar announces successful maiden flight of Shenzhou-50...")
+ >>> for keyword in template:
+ ... print(f"{keyword.term} ({keyword.category}): {keyword.description}")
+ """
+
+ def __init__(
+ self,
+ llm_client: BaseChatModel,
+ embedder: Embeddings,
+ *,
+ max_workers: int = 10,
+ verbose: bool = False,
+ **kwargs: Any,
+ ):
+ """
+ Initialize keyword list template.
+
+ Args:
+ llm_client: LLM client for knowledge extraction
+ embedder: Embedding model for semantic search
+ max_workers: Maximum number of worker threads, default: 10
+ verbose: Whether to output detailed logs, default: False
+ **kwargs: Other technical parameters, passed to base class
+ """
+ super().__init__(
+ item_schema=Keyword,
+ llm_client=llm_client,
+ embedder=embedder,
+ prompt=_PROMPT,
+ max_workers=max_workers,
+ verbose=verbose,
+ **kwargs,
+ )
+
+ def show(
+ self,
+ *,
+ top_k_for_search: int = 3,
+ top_k_for_chat: int = 3,
+ ):
+ """
+ Display keyword list.
+
+ Args:
+ top_k_for_search: Number of keywords to return for semantic search, default: 3
+ top_k_for_chat: Number of keywords to use for chat, default: 3
+ """
+ def itemLabelExtractor(item: Keyword) -> str:
+ return f"{item.term} ({item.category})"
+
+ super().show(
+ item_label_extractor=itemLabelExtractor,
+ top_k_for_search=top_k_for_search,
+ top_k_for_chat=top_k_for_chat,
+ )
diff --git a/hyperextract/templates/en/general/knowledge_graph.py b/hyperextract/templates/en/general/knowledge_graph.py
new file mode 100644
index 0000000..b4ea04d
--- /dev/null
+++ b/hyperextract/templates/en/general/knowledge_graph.py
@@ -0,0 +1,147 @@
+"""General Knowledge Graph - Extract entities and relationships from any text.
+
+Suitable for unstructured documents like arbitrary text, web scraped content, etc., to extract general entities and their relationships.
+"""
+
+from typing import Any, Optional, Tuple
+from pydantic import BaseModel, Field
+from langchain_core.language_models import BaseChatModel
+from langchain_core.embeddings import Embeddings
+from hyperextract.types import AutoGraph
+
+
+class GeneralEntity(BaseModel):
+ """General entity node"""
+ name: str = Field(description="Entity name, e.g., person name, organization name, product name")
+ category: str = Field(description="Entity type: Person, Organization, Location, Product, Concept, Other")
+ description: str = Field(description="Brief description", default="")
+
+
+class GeneralRelation(BaseModel):
+ """General relationship edge"""
+ source: str = Field(description="Source entity")
+ target: str = Field(description="Target entity")
+ relationType: str = Field(description="Relationship type: BelongsTo, LocatedIn, CollaboratesWith, CompetesWith, InventedBy, CreatedBy, RelatedTo, etc.")
+ details: str = Field(description="Detailed description", default="")
+
+
+_PROMPT = """You are a professional knowledge graph extraction expert. Please extract all entities (nodes) and their relationships (edges) from the text.
+
+### Node Extraction Rules
+1. Extract all entities: Person, Organization, Location, Product, Concept, etc.
+2. Assign a type to each entity: Person, Organization, Location, Product, Concept, Other
+3. Keep entity names consistent with the original text
+4. Add a brief description for each entity
+
+### Edge Extraction Rules
+1. Only create edges from extracted entities
+2. Relationship types include: BelongsTo, LocatedIn, CollaboratesWith, CompetesWith, InventedBy, CreatedBy, RelatedTo, etc.
+
+### Constraints
+- Each edge must connect extracted nodes
+- Do not create entities or relationships not mentioned in the text
+- Maintain objectivity and accuracy, do not add information not in the text
+
+### Source text:
+"""
+
+_NODE_PROMPT = """You are a professional entity recognition expert. Please extract all key entities as nodes from the text.
+
+### Extraction Rules
+1. Extract all entities: Person, Organization, Location, Product, Concept, etc.
+2. Assign a type to each entity: Person, Organization, Location, Product, Concept, Other
+3. Keep entity names consistent with the original text
+4. Add a brief description for each entity
+
+### Source text:
+"""
+
+_EDGE_PROMPT = """You are a professional relationship extraction expert. Please extract relationships between entities from the given entity list.
+
+### Constraints
+1. Only extract edges from the known entity list below
+2. Do not create unlisted entities
+3. Relationship types include: BelongsTo, LocatedIn, CollaboratesWith, CompetesWith, InventedBy, CreatedBy, RelatedTo, etc.
+
+"""
+
+
+class KnowledgeGraph(AutoGraph[GeneralEntity, GeneralRelation]):
+ """
+ Applicable documents: Arbitrary text, web scraped content, blog articles, news reports
+
+ Function introduction:
+ Extract general entities and their relationships from any text to build a knowledge graph. Supports multiple entity types including Person, Organization, Location, Product, Concept, etc.
+
+ Example:
+ >>> template = KnowledgeGraph(llm_client=llm, embedder=embedder)
+ >>> template.feed_text("Galaxy Interstellar announces successful maiden flight of Shenzhou-50...")
+ >>> template.show()
+ """
+
+ def __init__(
+ self,
+ llm_client: BaseChatModel,
+ embedder: Embeddings,
+ *,
+ extraction_mode: str = "two_stage",
+ max_workers: int = 10,
+ verbose: bool = False,
+ **kwargs: Any,
+ ):
+ """
+ Initialize general knowledge graph template.
+
+ Args:
+ llm_client: LLM client for knowledge extraction
+ embedder: Embedding model for semantic search
+ extraction_mode: Extraction mode, either "one_stage" (extract nodes and edges simultaneously)
+ or "two_stage" (extract nodes first, then edges), default: "two_stage"
+ max_workers: Maximum number of worker threads, default: 10
+ verbose: Whether to output detailed logs, default: False
+ **kwargs: Other technical parameters, passed to base class
+ """
+ super().__init__(
+ node_schema=GeneralEntity,
+ edge_schema=GeneralRelation,
+ node_key_extractor=lambda x: x.name,
+ edge_key_extractor=lambda x: f"{x.source}|{x.relationType}|{x.target}",
+ nodes_in_edge_extractor=lambda x: (x.source, x.target),
+ llm_client=llm_client,
+ embedder=embedder,
+ extraction_mode=extraction_mode,
+ max_workers=max_workers,
+ verbose=verbose,
+ prompt=_PROMPT,
+ prompt_for_node_extraction=_NODE_PROMPT,
+ prompt_for_edge_extraction=_EDGE_PROMPT,
+ **kwargs,
+ )
+
+ def show(
+ self,
+ *,
+ top_k_for_search: int = 3,
+ top_k_for_chat: int = 3,
+ ):
+ """
+ Display knowledge graph.
+
+ Args:
+ top_k_for_search: Number of nodes/edges to return for semantic search, default: 3
+ top_k_for_chat: Number of nodes/edges to use for chat, default: 3
+ """
+ def node_label_extractor(node: GeneralEntity) -> str:
+ return f"{node.name} ({node.category})"
+
+ def edge_label_extractor(edge: GeneralRelation) -> str:
+ return edge.relationType
+
+ super().show(
+ node_label_extractor=node_label_extractor,
+ edge_label_extractor=edge_label_extractor,
+ top_k_nodes_for_search=top_k_for_search,
+ top_k_edges_for_search=top_k_for_search,
+ top_k_nodes_for_chat=top_k_for_chat,
+ top_k_edges_for_chat=top_k_for_chat,
+ )
diff --git a/hyperextract/templates/en/general/life_event_timeline.py b/hyperextract/templates/en/general/life_event_timeline.py
new file mode 100644
index 0000000..04729dc
--- /dev/null
+++ b/hyperextract/templates/en/general/life_event_timeline.py
@@ -0,0 +1,230 @@
+"""Life Event Timeline - Extract timestamped events and arrange them chronologically.
+
+Suitable for biographies, chronologies, memoirs, etc.
+"""
+
+from typing import Any, Optional
+from pydantic import BaseModel, Field
+from langchain_core.language_models import BaseChatModel
+from langchain_core.embeddings import Embeddings
+from hyperextract.types import AutoTemporalGraph
+
+
+class LifeEntity(BaseModel):
+ """Life entity node"""
+ name: str = Field(description="Entity name")
+ type: str = Field(description="Entity type: Person, Location, Organization, Item, Concept, Other")
+ description: str = Field(description="Brief description", default="")
+
+
+class LifeEvent(BaseModel):
+ """Life event edge (with optional timestamp)"""
+ source: str = Field(description="Event subject/related entity")
+ target: str = Field(description="Event object/related entity")
+ eventType: str = Field(description="Event type: Birth, Death, Education, Work, Achievement, Interaction, Travel, Other")
+ eventDate: Optional[str] = Field(description="Event date, format: YYYY-MM-DD or YYYY, optional", default=None)
+ details: str = Field(description="Detailed event description", default="")
+
+
+_PROMPT = """You are a professional biography chronology expert. Please extract entities such as people, locations, organizations, etc., and life events from the text to build a life event timeline.
+
+### Node Extraction Rules
+1. Extract all related entities: Person, Location, Organization, Item, Concept, etc.
+2. Assign a type to each entity: Person, Location, Organization, Item, Concept, Other
+3. Add a brief description for each entity
+
+### Event Extraction Rules
+1. Only create event edges from extracted entities
+2. Event types include:
+ - Birth: Birth events
+ - Death: Death events
+ - Education: Learning, education events
+ - Work: Work, career events
+ - Achievement: Achievement, honor events
+ - Interaction: Interpersonal interaction, social activity events
+ - Travel: Travel, relocation events
+ - Other: Other events
+3. Event date is optional, only extract when explicitly mentioned in the text
+4. Date format: YYYY-MM-DD (e.g., 429-01-01) or year only (e.g., 429)
+5. Each edge must connect extracted nodes
+
+### English Time Calculation Rules
+Current observation date: {observation_time}
+
+1. **Relative Time Parsing**: Must parse relative time expressions based on observation date
+ - "last year" → Calculate as {observation_time} minus one year
+ - "this year" → The year of {observation_time}
+ - "next year" → {observation_time} plus one year
+ - "yesterday" → {observation_time} minus one day
+ - "today" → {observation_time}
+ - "tomorrow" → {observation_time} plus one day
+ - "last month" → First day of the month before {observation_time}
+ - "this month" → First day of the month of {observation_time}
+ - "next month" → First day of the month after {observation_time}
+ - "ten years ago" → {observation_time} minus ten years
+ - "five years later" → {observation_time} plus five years
+ - "beginning of year" → January 1st of the current year
+ - "end of year" → December 31st of the current year
+ - "beginning of century" → The 1st year of the century (e.g., 2000)
+ - "end of century" → The last year of the century (e.g., 2099)
+
+2. **Explicit Dates**: Keep explicit date formats from the original text unchanged
+ - Year only (e.g., 429, 2023) → 429, 2023
+ - Year-month (e.g., March 429) → 429-03
+ - Year-month-day (e.g., March 14, 429) → 429-03-14
+
+3. **Missing Time**: If there is no time information in the text, leave eventDate blank (None), do not fabricate dates
+
+### Constraints
+- Do not create entities or events not mentioned in the text
+- Maintain objectivity and accuracy, do not add information not in the text
+
+### Source text:
+"""
+
+_NODE_PROMPT = """You are a professional entity recognition expert. Please extract all related entities as nodes from the text.
+
+### Extraction Rules
+1. Extract all related entities: Person, Location, Organization, Item, Concept, etc.
+2. Assign a type to each entity: Person, Location, Organization, Item, Concept, Other
+3. Add a brief description for each entity
+
+### Source text:
+"""
+
+_EDGE_PROMPT = """You are a professional life event extraction expert. Please extract life events from the given entity list.
+
+### Event Type Explanation
+- Birth: Birth events
+- Death: Death events
+- Education: Learning, education events
+- Work: Work, career events
+- Achievement: Achievement, honor events
+- Interaction: Interpersonal interaction, social activity events
+- Travel: Travel, relocation events
+- Other: Other events
+
+### English Time Calculation Rules
+Current observation date: {observation_time}
+
+1. **Relative Time Parsing**: Must parse relative time expressions based on observation date
+ - "last year" → Calculate as {observation_time} minus one year
+ - "this year" → The year of {observation_time}
+ - "next year" → {observation_time} plus one year
+ - "yesterday" → {observation_time} minus one day
+ - "today" → {observation_time}
+ - "tomorrow" → {observation_time} plus one day
+ - "last month" → First day of the month before {observation_time}
+ - "this month" → First day of the month of {observation_time}
+ - "next month" → First day of the month after {observation_time}
+ - "ten years ago" → {observation_time} minus ten years
+ - "five years later" → {observation_time} plus five years
+ - "beginning of year" → January 1st of the current year
+ - "end of year" → December 31st of the current year
+ - "beginning of century" → The 1st year of the century (e.g., 2000)
+ - "end of century" → The last year of the century (e.g., 2099)
+
+2. **Explicit Dates**: Keep explicit date formats from the original text unchanged
+ - Year only (e.g., 429, 2023) → 429, 2023
+ - Year-month (e.g., March 429) → 429-03
+ - Year-month-day (e.g., March 14, 429) → 429-03-14
+
+3. **Missing Time**: If there is no time information in the text, leave eventDate blank (None), do not fabricate dates
+
+### Date Format Requirements
+- All date information should be converted to "YYYY-MM-DD" format (e.g., 429-01-01) or year only (e.g., 429)
+- Time information is optional, only extract when explicitly mentioned in the text
+
+### Constraints
+1. Only extract events from the known entity list below
+2. Do not create unlisted entities
+3. Time information is optional, only extract when explicitly mentioned in the text
+
+"""
+
+
+class LifeEventTimeline(AutoTemporalGraph[LifeEntity, LifeEvent]):
+ """
+ Applicable documents: Biographies, chronologies, memoirs, autobiographies
+
+ Function introduction:
+ Extract timestamped events and arrange them chronologically. Suitable for biographies, chronologies, memoirs, etc.
+
+ Example:
+ >>> template = LifeEventTimeline(llm_client=llm, embedder=embedder)
+ >>> template.feed_text("Zu Chongzhi (429-500), courtesy name Wenyuan, was from Qiuxian, Fanyang...")
+ >>> template.show()
+ """
+
+ def __init__(
+ self,
+ llm_client: BaseChatModel,
+ embedder: Embeddings,
+ *,
+ observation_time: str | None = None,
+ extraction_mode: str = "two_stage",
+ max_workers: int = 10,
+ verbose: bool = False,
+ **kwargs: Any,
+ ):
+ """
+ Initialize life event timeline template.
+
+ Args:
+ llm_client: LLM client for knowledge extraction
+ embedder: Embedding model for semantic search
+ observation_time: Observation date, used for parsing relative time expressions, default: today
+ extraction_mode: Extraction mode, either "one_stage" (extract nodes and edges simultaneously)
+ or "two_stage" (extract nodes first, then edges), default: "two_stage"
+ max_workers: Maximum number of worker threads, default: 10
+ verbose: Whether to output detailed logs, default: False
+ **kwargs: Other technical parameters, passed to base class
+ """
+ super().__init__(
+ node_schema=LifeEntity,
+ edge_schema=LifeEvent,
+ node_key_extractor=lambda x: x.name,
+ edge_key_extractor=lambda x: f"{x.source}|{x.eventType}|{x.target}",
+ time_in_edge_extractor=lambda x: x.eventDate,
+ nodes_in_edge_extractor=lambda x: (x.source, x.target),
+ llm_client=llm_client,
+ embedder=embedder,
+ observation_time=observation_time,
+ extraction_mode=extraction_mode,
+ max_workers=max_workers,
+ verbose=verbose,
+ prompt=_PROMPT,
+ prompt_for_node_extraction=_NODE_PROMPT,
+ prompt_for_edge_extraction=_EDGE_PROMPT,
+ **kwargs,
+ )
+
+ def show(
+ self,
+ *,
+ top_k_for_search: int = 3,
+ top_k_for_chat: int = 3,
+ ):
+ """
+ Display life event timeline.
+
+ Args:
+ top_k_for_search: Number of nodes/edges to return for semantic search, default: 3
+ top_k_for_chat: Number of nodes/edges to use for chat, default: 3
+ """
+ def node_label_extractor(node: LifeEntity) -> str:
+ return f"{node.name} ({node.type})"
+
+ def edge_label_extractor(edge: LifeEvent) -> str:
+ if edge.eventDate:
+ return f"{edge.eventType} ({edge.eventDate})"
+ return edge.eventType
+
+ super().show(
+ node_label_extractor=node_label_extractor,
+ edge_label_extractor=edge_label_extractor,
+ top_k_nodes_for_search=top_k_for_search,
+ top_k_edges_for_search=top_k_for_search,
+ top_k_nodes_for_chat=top_k_for_chat,
+ top_k_edges_for_chat=top_k_for_chat,
+ )
diff --git a/hyperextract/templates/en/general/operational_procedure.py b/hyperextract/templates/en/general/operational_procedure.py
new file mode 100644
index 0000000..d88f004
--- /dev/null
+++ b/hyperextract/templates/en/general/operational_procedure.py
@@ -0,0 +1,156 @@
+"""Operational Procedure - Extract the specific sequential steps for declaration, approval, or operations specified in regulations.
+
+Suitable for operation guidance, SOP process visualization, etc.
+"""
+
+from typing import Any
+from pydantic import BaseModel, Field
+from langchain_core.language_models import BaseChatModel
+from langchain_core.embeddings import Embeddings
+from hyperextract.types import AutoGraph
+
+
+class ProcedureStep(BaseModel):
+ """Procedure step node"""
+ stepId: str = Field(description="Step ID, e.g., Step 1, Step 2")
+ name: str = Field(description="Step name")
+ description: str = Field(description="Detailed step description")
+ role: str = Field(description="Executor role", default="")
+ inputRequired: str = Field(description="Required inputs/materials", default="")
+ outputResult: str = Field(description="Output/result", default="")
+
+
+class ProcedureTransition(BaseModel):
+ """Procedure transition edge"""
+ source: str = Field(description="Source step ID")
+ target: str = Field(description="Target step ID")
+ condition: str = Field(description="Transition condition", default="")
+ details: str = Field(description="Detailed explanation", default="")
+
+
+_PROMPT = """You are a professional process analysis expert. Please extract the specific sequential steps for declaration, approval, or operations specified in regulations from the text to build an operational procedure flowchart.
+
+### Node Extraction Rules
+1. Extract all process steps
+2. Assign an ID to each step (e.g., Step 1, Step 2)
+3. Add a name and detailed description to each step
+4. Record the executor role (if available)
+5. Record required inputs or materials (if available)
+6. Record output or result (if available)
+
+### Edge Extraction Rules
+1. Only create transition edges from extracted steps
+2. Record the sequential relationship between steps
+3. Record transition conditions (if available)
+4. Record detailed explanations (if available)
+
+### Constraints
+- Ensure the sequential order of process steps is accurate
+- Maintain objectivity and accuracy, do not add information not in the text
+
+### Source text:
+"""
+
+_NODE_PROMPT = """You are a professional process step recognition expert. Please extract all process steps as nodes from the text.
+
+### Extraction Rules
+1. Extract all process steps
+2. Assign an ID to each step (e.g., Step 1, Step 2)
+3. Add a name and detailed description to each step
+4. Record the executor role (if available)
+5. Record required inputs or materials (if available)
+6. Record output or result (if available)
+
+### Source text:
+"""
+
+_EDGE_PROMPT = """You are a professional process transition recognition expert. Please extract the sequential relationship between steps from the given step list.
+
+### Constraints
+1. Only extract edges from the known step list below
+2. Do not create unlisted steps
+
+"""
+
+
+class OperationalProcedure(AutoGraph[ProcedureStep, ProcedureTransition]):
+ """
+ Applicable documents: Operation manuals, SOPs, compliance guidelines, approval processes
+
+ Function introduction:
+ Extract the specific sequential steps for declaration, approval, or operations specified in regulations. Suitable for operation guidance, SOP process visualization, etc.
+
+ Example:
+ >>> template = OperationalProcedure(llm_client=llm, embedder=embedder)
+ >>> template.feed_text("Universe First Slacker Company Leave Approval Process...")
+ >>> template.show()
+ """
+
+ def __init__(
+ self,
+ llm_client: BaseChatModel,
+ embedder: Embeddings,
+ *,
+ extraction_mode: str = "two_stage",
+ max_workers: int = 10,
+ verbose: bool = False,
+ **kwargs: Any,
+ ):
+ """
+ Initialize operational procedure template.
+
+ Args:
+ llm_client: LLM client for knowledge extraction
+ embedder: Embedding model for semantic search
+ extraction_mode: Extraction mode, either "one_stage" (extract nodes and edges simultaneously)
+ or "two_stage" (extract nodes first, then edges), default: "two_stage"
+ max_workers: Maximum number of worker threads, default: 10
+ verbose: Whether to output detailed logs, default: False
+ **kwargs: Other technical parameters, passed to base class
+ """
+ super().__init__(
+ node_schema=ProcedureStep,
+ edge_schema=ProcedureTransition,
+ node_key_extractor=lambda x: x.stepId,
+ edge_key_extractor=lambda x: f"{x.source}|{x.target}",
+ nodes_in_edge_extractor=lambda x: (x.source, x.target),
+ llm_client=llm_client,
+ embedder=embedder,
+ extraction_mode=extraction_mode,
+ max_workers=max_workers,
+ verbose=verbose,
+ prompt=_PROMPT,
+ prompt_for_node_extraction=_NODE_PROMPT,
+ prompt_for_edge_extraction=_EDGE_PROMPT,
+ **kwargs,
+ )
+
+ def show(
+ self,
+ *,
+ top_k_for_search: int = 3,
+ top_k_for_chat: int = 3,
+ ):
+ """
+ Display operational procedure flowchart.
+
+ Args:
+ top_k_for_search: Number of nodes/edges to return for semantic search, default: 3
+ top_k_for_chat: Number of nodes/edges to use for chat, default: 3
+ """
+ def node_label_extractor(node: ProcedureStep) -> str:
+ return f"{node.stepId}: {node.name}"
+
+ def edge_label_extractor(edge: ProcedureTransition) -> str:
+ if edge.condition:
+ return edge.condition
+ return "→"
+
+ super().show(
+ node_label_extractor=node_label_extractor,
+ edge_label_extractor=edge_label_extractor,
+ top_k_nodes_for_search=top_k_for_search,
+ top_k_edges_for_search=top_k_for_search,
+ top_k_nodes_for_chat=top_k_for_chat,
+ top_k_edges_for_chat=top_k_for_chat,
+ )
diff --git a/hyperextract/templates/en/general/penalty_mapping.py b/hyperextract/templates/en/general/penalty_mapping.py
new file mode 100644
index 0000000..a641b86
--- /dev/null
+++ b/hyperextract/templates/en/general/penalty_mapping.py
@@ -0,0 +1,145 @@
+"""Penalty Mapping - Describe the evolution path from "violation behavior" to "handling procedure" to "final result".
+
+Suitable for risk tracing, penalty logic analysis, etc.
+"""
+
+from typing import Any
+from pydantic import BaseModel, Field
+from langchain_core.language_models import BaseChatModel
+from langchain_core.embeddings import Embeddings
+from hyperextract.types import AutoGraph
+
+
+class PenaltyNode(BaseModel):
+ """Penalty causal node"""
+ name: str = Field(description="Node name")
+ type: str = Field(description="Node type: Violation, HandlingProcedure, PenaltyResult, Other")
+ description: str = Field(description="Detailed description")
+
+
+class PenaltyPath(BaseModel):
+ """Penalty causal path edge"""
+ source: str = Field(description="Source node name")
+ target: str = Field(description="Target node name")
+ relation: str = Field(description="Relationship type: Causes, Triggers, Executes, Produces, Other")
+ details: str = Field(description="Detailed explanation", default="")
+
+
+_PROMPT = """You are a professional penalty logic analysis expert. Please extract the evolution path from "violation behavior" to "handling procedure" to "final result" from the text to build a penalty mapping.
+
+### Node Extraction Rules
+1. Extract all key nodes: Violation, HandlingProcedure, PenaltyResult, etc.
+2. Assign a type to each node: Violation, HandlingProcedure, PenaltyResult, Other
+3. Add a detailed description to each node
+
+### Edge Extraction Rules
+1. Only create causal path edges from extracted nodes
+2. Record the causal relationship between nodes
+3. Relationship types include: Causes, Triggers, Executes, Produces, Other
+4. Add detailed explanations (if available)
+
+### Constraints
+- Ensure the logical order of causal relationships is correct
+- Maintain objectivity and accuracy, do not add information not in the text
+
+### Source text:
+"""
+
+_NODE_PROMPT = """You are a professional penalty node recognition expert. Please extract all key nodes from the text.
+
+### Extraction Rules
+1. Extract all key nodes: Violation, HandlingProcedure, PenaltyResult, etc.
+2. Assign a type to each node: Violation, HandlingProcedure, PenaltyResult, Other
+3. Add a detailed description to each node
+
+### Source text:
+"""
+
+_EDGE_PROMPT = """You are a professional penalty causal relationship recognition expert. Please extract the causal relationship between nodes from the given node list.
+
+### Constraints
+1. Only extract edges from the known node list below
+2. Do not create unlisted nodes
+
+"""
+
+
+class PenaltyMapping(AutoGraph[PenaltyNode, PenaltyPath]):
+ """
+ Applicable documents: Company internal management systems, administrative regulations, compliance guidelines, penalty cases
+
+ Function introduction:
+ Describe the evolution path from "violation behavior" to "handling procedure" to "final result". Suitable for risk tracing, penalty logic analysis, etc.
+
+ Example:
+ >>> template = PenaltyMapping(llm_client=llm, embedder=embedder)
+ >>> template.feed_text("Universe First Slacker Company Employee Penalty Regulations...")
+ >>> template.show()
+ """
+
+ def __init__(
+ self,
+ llm_client: BaseChatModel,
+ embedder: Embeddings,
+ *,
+ extraction_mode: str = "two_stage",
+ max_workers: int = 10,
+ verbose: bool = False,
+ **kwargs: Any,
+ ):
+ """
+ Initialize penalty mapping template.
+
+ Args:
+ llm_client: LLM client for knowledge extraction
+ embedder: Embedding model for semantic search
+ extraction_mode: Extraction mode, either "one_stage" (extract nodes and edges simultaneously)
+ or "two_stage" (extract nodes first, then edges), default: "two_stage"
+ max_workers: Maximum number of worker threads, default: 10
+ verbose: Whether to output detailed logs, default: False
+ **kwargs: Other technical parameters, passed to base class
+ """
+ super().__init__(
+ node_schema=PenaltyNode,
+ edge_schema=PenaltyPath,
+ node_key_extractor=lambda x: x.name,
+ edge_key_extractor=lambda x: f"{x.source}|{x.relation}|{x.target}",
+ nodes_in_edge_extractor=lambda x: (x.source, x.target),
+ llm_client=llm_client,
+ embedder=embedder,
+ extraction_mode=extraction_mode,
+ max_workers=max_workers,
+ verbose=verbose,
+ prompt=_PROMPT,
+ prompt_for_node_extraction=_NODE_PROMPT,
+ prompt_for_edge_extraction=_EDGE_PROMPT,
+ **kwargs,
+ )
+
+ def show(
+ self,
+ *,
+ top_k_for_search: int = 3,
+ top_k_for_chat: int = 3,
+ ):
+ """
+ Display penalty mapping.
+
+ Args:
+ top_k_for_search: Number of nodes/edges to return for semantic search, default: 3
+ top_k_for_chat: Number of nodes/edges to use for chat, default: 3
+ """
+ def node_label_extractor(node: PenaltyNode) -> str:
+ return f"{node.name} ({node.type})"
+
+ def edge_label_extractor(edge: PenaltyPath) -> str:
+ return edge.relation
+
+ super().show(
+ node_label_extractor=node_label_extractor,
+ edge_label_extractor=edge_label_extractor,
+ top_k_nodes_for_search=top_k_for_search,
+ top_k_edges_for_search=top_k_for_search,
+ top_k_nodes_for_chat=top_k_for_chat,
+ top_k_edges_for_chat=top_k_for_chat,
+ )
diff --git a/hyperextract/templates/en/general/penalty_registry.py b/hyperextract/templates/en/general/penalty_registry.py
new file mode 100644
index 0000000..1fb73f4
--- /dev/null
+++ b/hyperextract/templates/en/general/penalty_registry.py
@@ -0,0 +1,103 @@
+"""Penalty Registry - Deduplicate and summarize specific violations and their corresponding consequences to form a clear penalty ledger.
+
+Suitable for internal risk control, compliance checks, etc.
+"""
+
+from typing import Any
+from pydantic import BaseModel, Field
+from langchain_core.language_models import BaseChatModel
+from langchain_core.embeddings import Embeddings
+from hyperextract.types import AutoSet
+
+
+class PenaltyEntry(BaseModel):
+ """Violation penalty entry"""
+ violation: str = Field(description="Violation description")
+ severity: str = Field(description="Severity: Minor, Moderate, Serious, VerySerious")
+ penalty: str = Field(description="Penalty measure")
+ applicableClause: str = Field(description="Applicable clause", default="")
+ notes: str = Field(description="Notes", default="")
+
+
+_PROMPT = """You are a professional compliance analysis expert. Please extract all violations and their corresponding penalty measures from the text to form a violation penalty registry.
+
+### Extraction Rules
+1. Extract all violations and their corresponding penalty measures
+2. Assign a severity level to each violation: Minor, Moderate, Serious, VerySerious
+3. Maintain the accuracy of violations and penalty measures
+4. Record applicable clauses (if available)
+5. Add notes (if available)
+
+### Constraints
+- Deduplication: Only keep one record for the same violation and penalty measure combination
+- Maintain objectivity and accuracy, do not add information not in the text
+
+### Source text:
+"""
+
+
+class PenaltyRegistry(AutoSet[PenaltyEntry]):
+ """
+ Applicable documents: Company internal management systems, administrative regulations, compliance guidelines
+
+ Function introduction:
+ Deduplicate and summarize specific violations and their corresponding consequences to form a clear penalty ledger. Suitable for internal risk control, compliance checks, etc.
+
+ Example:
+ >>> template = PenaltyRegistry(llm_client=llm, embedder=embedder)
+ >>> template.feed_text("Universe First Slacker Company Employee Attendance Management System...")
+ >>> for entry in template:
+ ... print(f"{entry.violation} -> {entry.penalty}")
+ """
+
+ def __init__(
+ self,
+ llm_client: BaseChatModel,
+ embedder: Embeddings,
+ *,
+ max_workers: int = 10,
+ verbose: bool = False,
+ **kwargs: Any,
+ ):
+ """
+ Initialize penalty registry template.
+
+ Args:
+ llm_client: LLM client for knowledge extraction
+ embedder: Embedding model for semantic search
+ max_workers: Maximum number of worker threads, default: 10
+ verbose: Whether to output detailed logs, default: False
+ **kwargs: Other technical parameters, passed to base class
+ """
+ super().__init__(
+ item_schema=PenaltyEntry,
+ key_extractor=lambda x: f"{x.violation}_{x.penalty}",
+ llm_client=llm_client,
+ embedder=embedder,
+ prompt=_PROMPT,
+ max_workers=max_workers,
+ verbose=verbose,
+ **kwargs,
+ )
+
+ def show(
+ self,
+ *,
+ top_k_for_search: int = 3,
+ top_k_for_chat: int = 3,
+ ):
+ """
+ Display penalty registry.
+
+ Args:
+ top_k_for_search: Number of entries to return for semantic search, default: 3
+ top_k_for_chat: Number of entries to use for chat, default: 3
+ """
+ def itemLabelExtractor(item: PenaltyEntry) -> str:
+ return f"{item.violation} ({item.severity})"
+
+ super().show(
+ item_label_extractor=itemLabelExtractor,
+ top_k_for_search=top_k_for_search,
+ top_k_for_chat=top_k_for_chat,
+ )
diff --git a/hyperextract/templates/en/general/personal_profile.py b/hyperextract/templates/en/general/personal_profile.py
new file mode 100644
index 0000000..417bcd0
--- /dev/null
+++ b/hyperextract/templates/en/general/personal_profile.py
@@ -0,0 +1,114 @@
+"""Personal Profile - Aggregate static attributes of a person (birth/death dates, education, core identity).
+
+Suitable for resumes, person introductions, obituary summaries, etc.
+"""
+
+from typing import Any, Optional, List
+from pydantic import BaseModel, Field
+from langchain_core.language_models import BaseChatModel
+from langchain_core.embeddings import Embeddings
+from hyperextract.types import AutoModel
+
+
+class PersonalInfo(BaseModel):
+ """Personal profile information"""
+ name: str = Field(description="Name")
+ gender: Optional[str] = Field(description="Gender", default=None)
+ birthDate: Optional[str] = Field(description="Date of birth, format: YYYY-MM-DD or YYYY", default=None)
+ deathDate: Optional[str] = Field(description="Date of death, format: YYYY-MM-DD or YYYY", default=None)
+ nationality: Optional[str] = Field(description="Nationality", default=None)
+ birthPlace: Optional[str] = Field(description="Place of birth", default=None)
+ education: List[str] = Field(description="Education background", default_factory=list)
+ occupations: List[str] = Field(description="Occupations/identities", default_factory=list)
+ coreIdentity: List[str] = Field(description="Core identity/tags", default_factory=list)
+ majorAchievements: List[str] = Field(description="Major achievements", default_factory=list)
+ affiliations: List[str] = Field(description="Affiliated organizations/institutions", default_factory=list)
+ summary: str = Field(description="Personal introduction", default="")
+
+
+_PROMPT = """You are a professional personal profile editor. Please extract static attribute information of this person from the text to build a personal profile.
+
+### Extraction Rules
+1. Extract basic information: name, gender, date of birth, date of death, nationality, place of birth, etc.
+2. Extract education background
+3. Extract occupation or identity information
+4. Extract core identity tags
+5. Extract major achievements
+6. Extract affiliated organizations or institutions
+7. Write a 100-200 word personal introduction
+
+### Date Format Requirements
+All date information should be converted to "YYYY-MM-DD" format (e.g., 429-01-01) or year only (e.g., 429).
+
+### Constraints
+- Only extract information explicitly mentioned in the text, do not add additional content
+- Maintain objectivity and accuracy
+- Leave blank if information is missing, do not fabricate
+
+### Source text:
+"""
+
+
+class PersonalProfile(AutoModel[PersonalInfo]):
+ """
+ Applicable documents: Resumes, person introductions, obituary summaries, Wikipedia person entries
+
+ Function introduction:
+ Aggregate static attributes of a person (birth/death dates, education, core identity). Suitable for resumes, person introductions, obituary summaries, etc.
+
+ Example:
+ >>> template = PersonalProfile(llm_client=llm, embedder=embedder)
+ >>> template.feed_text("Zu Chongzhi (429-500), courtesy name Wenyuan, was from Qiuxian, Fanyang...")
+ >>> print(template.data.name)
+ >>> print(template.data.birthDate)
+ """
+
+ def __init__(
+ self,
+ llm_client: BaseChatModel,
+ embedder: Embeddings,
+ *,
+ max_workers: int = 10,
+ verbose: bool = False,
+ **kwargs: Any,
+ ):
+ """
+ Initialize personal profile template.
+
+ Args:
+ llm_client: LLM client for knowledge extraction
+ embedder: Embedding model for semantic search
+ max_workers: Maximum number of worker threads, default: 10
+ verbose: Whether to output detailed logs, default: False
+ **kwargs: Other technical parameters, passed to base class
+ """
+ super().__init__(
+ data_schema=PersonalInfo,
+ llm_client=llm_client,
+ embedder=embedder,
+ prompt=_PROMPT,
+ max_workers=max_workers,
+ verbose=verbose,
+ **kwargs,
+ )
+
+ def show(
+ self,
+ *,
+ top_k: int = 3,
+ ):
+ """
+ Display personal profile.
+
+ Args:
+ top_k: Number of items to use for chat, default: 3
+ """
+ def label_extractor(item: PersonalInfo) -> str:
+ if item.birthDate and item.deathDate:
+ return f"{item.name} ({item.birthDate}-{item.deathDate})"
+ return item.name
+
+ super().show(
+ label_extractor=label_extractor,
+ top_k=top_k,
+ )
diff --git a/hyperextract/templates/en/general/regulation_profile.py b/hyperextract/templates/en/general/regulation_profile.py
new file mode 100644
index 0000000..487d89c
--- /dev/null
+++ b/hyperextract/templates/en/general/regulation_profile.py
@@ -0,0 +1,112 @@
+"""Regulation Profile - Extract the name, version, scope, effective date, and core purpose of a regulation.
+
+Suitable for regulation overview, version management, etc.
+"""
+
+from typing import Any, Optional, List
+from pydantic import BaseModel, Field
+from langchain_core.language_models import BaseChatModel
+from langchain_core.embeddings import Embeddings
+from hyperextract.types import AutoModel
+
+
+class RegulationInfo(BaseModel):
+ """Regulation metadata information"""
+ title: str = Field(description="Regulation title")
+ version: Optional[str] = Field(description="Version number", default=None)
+ issuer: Optional[str] = Field(description="Issuing authority", default=None)
+ issueDate: Optional[str] = Field(description="Issue date", default=None)
+ effectiveDate: Optional[str] = Field(description="Effective date", default=None)
+ expiryDate: Optional[str] = Field(description="Expiry date", default=None)
+ scope: List[str] = Field(description="Applicable scope", default_factory=list)
+ targetAudience: List[str] = Field(description="Target audience", default_factory=list)
+ corePurpose: str = Field(description="Core purpose/objective", default="")
+ keywords: List[str] = Field(description="Keywords/tags", default_factory=list)
+ relatedRegulations: List[str] = Field(description="Related regulations", default_factory=list)
+
+
+_PROMPT = """You are a professional regulation analysis expert. Please extract metadata information of this regulation from the text to build a regulation profile.
+
+### Extraction Rules
+1. Extract the full name of the regulation
+2. Extract version number (if available)
+3. Extract issuing authority
+4. Extract issue date, effective date, expiry date (if available)
+5. Extract applicable scope
+6. Extract target audience
+7. Extract core purpose or objective (100-200 words)
+8. Extract keywords or tags
+9. Extract related other regulations
+
+### Constraints
+- Only extract information explicitly mentioned in the text, do not add additional content
+- Maintain objectivity and accuracy
+- Leave blank if information is missing, do not fabricate
+
+### Source text:
+"""
+
+
+class RegulationProfile(AutoModel[RegulationInfo]):
+ """
+ Applicable documents: Company internal management systems, administrative regulations, operation manuals, compliance guidelines
+
+ Function introduction:
+ Extract the name, version, scope, effective date, and core purpose of a regulation. Suitable for regulation overview, version management, etc.
+
+ Example:
+ >>> template = RegulationProfile(llm_client=llm, embedder=embedder)
+ >>> template.feed_text("Universe First Slacker Company Employee Attendance Management System...")
+ >>> print(template.data.title)
+ >>> print(template.data.effectiveDate)
+ """
+
+ def __init__(
+ self,
+ llm_client: BaseChatModel,
+ embedder: Embeddings,
+ *,
+ max_workers: int = 10,
+ verbose: bool = False,
+ **kwargs: Any,
+ ):
+ """
+ Initialize regulation profile template.
+
+ Args:
+ llm_client: LLM client for knowledge extraction
+ embedder: Embedding model for semantic search
+ max_workers: Maximum number of worker threads, default: 10
+ verbose: Whether to output detailed logs, default: False
+ **kwargs: Other technical parameters, passed to base class
+ """
+ super().__init__(
+ data_schema=RegulationInfo,
+ llm_client=llm_client,
+ embedder=embedder,
+ prompt=_PROMPT,
+ max_workers=max_workers,
+ verbose=verbose,
+ **kwargs,
+ )
+
+ def show(
+ self,
+ *,
+ top_k: int = 3,
+ ):
+ """
+ Display regulation profile.
+
+ Args:
+ top_k: Number of items to use for chat, default: 3
+ """
+ def label_extractor(item: RegulationInfo) -> str:
+ if item.version:
+ return f"{item.title} (v{item.version})"
+ return item.title
+
+ super().show(
+ label_extractor=label_extractor,
+ top_k=top_k,
+ )
diff --git a/hyperextract/templates/en/general/social_network.py b/hyperextract/templates/en/general/social_network.py
new file mode 100644
index 0000000..47eeed2
--- /dev/null
+++ b/hyperextract/templates/en/general/social_network.py
@@ -0,0 +1,169 @@
+"""Social Network Graph - Focus on extracting interpersonal relationships, interactions, and organizational affiliations.
+
+Suitable for biography analysis, character setting research, etc.
+"""
+
+from typing import Any
+from pydantic import BaseModel, Field
+from langchain_core.language_models import BaseChatModel
+from langchain_core.embeddings import Embeddings
+from hyperextract.types import AutoGraph
+
+
+class PersonNode(BaseModel):
+ """Person/organization node"""
+ name: str = Field(description="Name")
+ type: str = Field(description="Node type: Person, Organization, Institution, Other")
+ description: str = Field(description="Brief description", default="")
+ role: str = Field(description="Role/identity", default="")
+
+
+class SocialRelation(BaseModel):
+ """Social relationship edge"""
+ source: str = Field(description="Source node")
+ target: str = Field(description="Target node")
+ relationType: str = Field(description="Relationship type: Family, Friend, Colleague, TeacherStudent, SupervisorSubordinate, Collaboration, Competition, Affiliation, Other")
+ details: str = Field(description="Detailed relationship explanation", default="")
+
+
+_PROMPT = """You are a professional social relationship analysis expert. Please extract people, organizations, and their mutual relationships from the text to build a social network graph.
+
+### Node Extraction Rules
+1. Extract all people, organizations, institutions, etc.
+2. Assign a type to each node: Person, Organization, Institution, Other
+3. Add a brief description for each node
+4. Annotate role or identity for each node
+
+### Edge Extraction Rules
+1. Only create edges from extracted nodes
+2. Relationship types include:
+ - Family: Family relationships (parents, children, spouse, siblings, etc.)
+ - Friend: Friend relationships
+ - Colleague: Work partner relationships
+ - TeacherStudent: Educational relationships
+ - SupervisorSubordinate: Management relationships
+ - Collaboration: Collaborative relationships
+ - Competition: Competitive relationships
+ - Affiliation: Organizational affiliation relationships
+ - Other: Other relationships
+3. Each edge must connect extracted nodes
+
+### Constraints
+- Do not create nodes or relationships not mentioned in the text
+- Maintain objectivity and accuracy, do not add information not in the text
+
+### Source text:
+"""
+
+_NODE_PROMPT = """You are a professional person and organization recognition expert. Please extract all people and organizations as nodes from the text.
+
+### Extraction Rules
+1. Extract all people, organizations, institutions, etc.
+2. Assign a type to each node: Person, Organization, Institution, Other
+3. Add a brief description for each node
+4. Annotate role or identity for each node
+
+### Source text:
+"""
+
+_EDGE_PROMPT = """You are a professional social relationship extraction expert. Please extract mutual relationships between nodes from the given node list.
+
+### Relationship Type Explanation
+- Family: Family relationships (parents, children, spouse, siblings, etc.)
+- Friend: Friend relationships
+- Colleague: Work partner relationships
+- TeacherStudent: Educational relationships
+- SupervisorSubordinate: Management relationships
+- Collaboration: Collaborative relationships
+- Competition: Competitive relationships
+- Affiliation: Organizational affiliation relationships
+- Other: Other relationships
+
+### Constraints
+1. Only extract edges from the known node list below
+2. Do not create unlisted nodes
+
+"""
+
+
+class SocialNetwork(AutoGraph[PersonNode, SocialRelation]):
+ """
+ Applicable documents: Biographies, memoirs, novels, character setting documents
+
+ Function introduction:
+ Focus on extracting interpersonal relationships, interactions, and organizational affiliations. Suitable for biography analysis, character setting research, etc.
+
+ Example:
+ >>> template = SocialNetwork(llm_client=llm, embedder=embedder)
+ >>> template.feed_text("Zu Chongzhi studied under He Chengtian and debated with Dai Faxing...")
+ >>> template.show()
+ """
+
+ def __init__(
+ self,
+ llm_client: BaseChatModel,
+ embedder: Embeddings,
+ *,
+ extraction_mode: str = "two_stage",
+ max_workers: int = 10,
+ verbose: bool = False,
+ **kwargs: Any,
+ ):
+ """
+ Initialize social network graph template.
+
+ Args:
+ llm_client: LLM client for knowledge extraction
+ embedder: Embedding model for semantic search
+ extraction_mode: Extraction mode, either "one_stage" (extract nodes and edges simultaneously)
+ or "two_stage" (extract nodes first, then edges), default: "two_stage"
+ max_workers: Maximum number of worker threads, default: 10
+ verbose: Whether to output detailed logs, default: False
+ **kwargs: Other technical parameters, passed to base class
+ """
+ super().__init__(
+ node_schema=PersonNode,
+ edge_schema=SocialRelation,
+ node_key_extractor=lambda x: x.name,
+ edge_key_extractor=lambda x: f"{x.source}|{x.relationType}|{x.target}",
+ nodes_in_edge_extractor=lambda x: (x.source, x.target),
+ llm_client=llm_client,
+ embedder=embedder,
+ extraction_mode=extraction_mode,
+ max_workers=max_workers,
+ verbose=verbose,
+ prompt=_PROMPT,
+ prompt_for_node_extraction=_NODE_PROMPT,
+ prompt_for_edge_extraction=_EDGE_PROMPT,
+ **kwargs,
+ )
+
+ def show(
+ self,
+ *,
+ top_k_for_search: int = 3,
+ top_k_for_chat: int = 3,
+ ):
+ """
+ Display social network graph.
+
+ Args:
+ top_k_for_search: Number of nodes/edges to return for semantic search, default: 3
+ top_k_for_chat: Number of nodes/edges to use for chat, default: 3
+ """
+ def node_label_extractor(node: PersonNode) -> str:
+ if node.role:
+ return f"{node.name} ({node.role})"
+ return f"{node.name} ({node.type})"
+
+ def edge_label_extractor(edge: SocialRelation) -> str:
+ return edge.relationType
+
+ super().show(
+ node_label_extractor=node_label_extractor,
+ edge_label_extractor=edge_label_extractor,
+ top_k_nodes_for_search=top_k_for_search,
+ top_k_edges_for_search=top_k_for_search,
+ top_k_nodes_for_chat=top_k_for_chat,
+ top_k_edges_for_chat=top_k_for_chat,
+ )
diff --git a/hyperextract/templates/zh/__init__.py b/hyperextract/templates/zh/__init__.py
index fa57f54..c303507 100644
--- a/hyperextract/templates/zh/__init__.py
+++ b/hyperextract/templates/zh/__init__.py
@@ -1,57 +1,34 @@
from .general import (
# 知识图谱
KnowledgeGraph,
- GeneralEntity,
- GeneralRelation,
# 实体集合
EntityRegistry,
- RegistryEntry,
# 关键词列表
KeywordList,
- Keyword,
# 百科条目
EncyclopediaItem,
- EncyclopediaInfo,
# 概念层级图
ConceptHierarchy,
- ConceptNode,
- HierarchyRelation,
# 引用网络
CrossReferenceNet,
- ReferenceNode,
- ReferenceRelation,
# 个人档案
PersonalProfile,
- PersonalInfo,
# 社会网络
SocialNetwork,
- PersonNode,
- SocialRelation,
# 生平时序图
LifeEventTimeline,
- LifeEntity,
- LifeEvent,
# 规章元数据快照
RegulationProfile,
- RegulationInfo,
# 规章条文清单
ClauseList,
- Clause,
# 违规处罚对照表
PenaltyRegistry,
- PenaltyEntry,
# 执行程序流程图
OperationalProcedure,
- ProcedureStep,
- ProcedureTransition,
# 违规处罚因果链
PenaltyMapping,
- PenaltyNode,
- PenaltyPath,
# 合规行为超图
ComplianceLogic,
- ComplianceEntity,
- ComplianceRule,
)
from .agriculture import CropCycleGraph, LivestockGraph, AgriPestControl
from .biology import TaxonomyGraph, BiologicalNetwork, MetabolicHypergraph
@@ -99,57 +76,34 @@ __all__ = [
# general
## 知识图谱
"KnowledgeGraph",
- "GeneralEntity",
- "GeneralRelation",
## 实体集合
"EntityRegistry",
- "RegistryEntry",
## 关键词列表
"KeywordList",
- "Keyword",
## 百科条目
"EncyclopediaItem",
- "EncyclopediaInfo",
## 概念层级图
"ConceptHierarchy",
- "ConceptNode",
- "HierarchyRelation",
## 引用网络
"CrossReferenceNet",
- "ReferenceNode",
- "ReferenceRelation",
## 个人档案
"PersonalProfile",
- "PersonalInfo",
## 社会网络
"SocialNetwork",
- "PersonNode",
- "SocialRelation",
## 生平时序图
"LifeEventTimeline",
- "LifeEntity",
- "LifeEvent",
## 规章元数据快照
"RegulationProfile",
- "RegulationInfo",
## 规章条文清单
"ClauseList",
- "Clause",
## 违规处罚对照表
"PenaltyRegistry",
- "PenaltyEntry",
## 执行程序流程图
"OperationalProcedure",
- "ProcedureStep",
- "ProcedureTransition",
## 违规处罚因果链
"PenaltyMapping",
- "PenaltyNode",
- "PenaltyPath",
## 合规行为超图
"ComplianceLogic",
- "ComplianceEntity",
- "ComplianceRule",
# Finance
"RiskAssessmentGraph",
"MarketSentimentGraph",
@@ -196,9 +150,11 @@ __all__ = [
"CultivationSystemMap",
"SectRelationGraph",
"ArtifactRegistry",
+ # Agriculture
"CropCycleGraph",
"LivestockGraph",
"AgriPestControl",
+ # Biology
"TaxonomyGraph",
"BiologicalNetwork",
"MetabolicHypergraph",
diff --git a/hyperextract/templates/zh/general/__init__.py b/hyperextract/templates/zh/general/__init__.py
index f6a38eb..d3a0ed0 100644
--- a/hyperextract/templates/zh/general/__init__.py
+++ b/hyperextract/templates/zh/general/__init__.py
@@ -1,71 +1,48 @@
-from .knowledge_graph import KnowledgeGraph, GeneralEntity, GeneralRelation
-from .entity_registry import EntityRegistry, RegistryEntry
-from .keyword_list import KeywordList, Keyword
-from .encyclopedia_item import EncyclopediaItem, EncyclopediaInfo
-from .concept_hierarchy import ConceptHierarchy, ConceptNode, HierarchyRelation
-from .cross_reference_net import CrossReferenceNet, ReferenceNode, ReferenceRelation
-from .personal_profile import PersonalProfile, PersonalInfo
-from .social_network import SocialNetwork, PersonNode, SocialRelation
-from .life_event_timeline import LifeEventTimeline, LifeEntity, LifeEvent
-from .regulation_profile import RegulationProfile, RegulationInfo
-from .clause_list import ClauseList, Clause
-from .penalty_registry import PenaltyRegistry, PenaltyEntry
-from .operational_procedure import OperationalProcedure, ProcedureStep, ProcedureTransition
-from .penalty_mapping import PenaltyMapping, PenaltyNode, PenaltyPath
-from .compliance_logic import ComplianceLogic, ComplianceEntity, ComplianceRule
+from .knowledge_graph import KnowledgeGraph
+from .entity_registry import EntityRegistry
+from .keyword_list import KeywordList
+from .encyclopedia_item import EncyclopediaItem
+from .concept_hierarchy import ConceptHierarchy
+from .cross_reference_net import CrossReferenceNet
+from .personal_profile import PersonalProfile
+from .social_network import SocialNetwork
+from .life_event_timeline import LifeEventTimeline
+from .regulation_profile import RegulationProfile
+from .clause_list import ClauseList
+from .penalty_registry import PenaltyRegistry
+from .operational_procedure import OperationalProcedure
+from .penalty_mapping import PenaltyMapping
+from .compliance_logic import ComplianceLogic
__all__ = [
# 知识图谱
"KnowledgeGraph",
- "GeneralEntity",
- "GeneralRelation",
# 实体集合
"EntityRegistry",
- "RegistryEntry",
# 关键词列表
"KeywordList",
- "Keyword",
# 百科条目
"EncyclopediaItem",
- "EncyclopediaInfo",
# 概念层级图
"ConceptHierarchy",
- "ConceptNode",
- "HierarchyRelation",
# 引用网络
"CrossReferenceNet",
- "ReferenceNode",
- "ReferenceRelation",
# 个人档案
"PersonalProfile",
- "PersonalInfo",
# 社会网络
"SocialNetwork",
- "PersonNode",
- "SocialRelation",
# 生平时序图
"LifeEventTimeline",
- "LifeEntity",
- "LifeEvent",
# 规章元数据快照
"RegulationProfile",
- "RegulationInfo",
# 规章条文清单
"ClauseList",
- "Clause",
# 违规处罚对照表
"PenaltyRegistry",
- "PenaltyEntry",
# 执行程序流程图
"OperationalProcedure",
- "ProcedureStep",
- "ProcedureTransition",
# 违规处罚因果链
"PenaltyMapping",
- "PenaltyNode",
- "PenaltyPath",
# 合规行为超图
- "ComplianceLogic",
- "ComplianceEntity",
- "ComplianceRule"
+ "ComplianceLogic"
]
diff --git a/hyperextract/templates/zh/general/compliance_logic.py b/hyperextract/templates/zh/general/compliance_logic.py
index 51cf98d..571ae9f 100644
--- a/hyperextract/templates/zh/general/compliance_logic.py
+++ b/hyperextract/templates/zh/general/compliance_logic.py
@@ -1,78 +1,86 @@
-"""合规行为超图 - 建模“在何种条件下,何种主体,必须/禁止执行何种操作”的复杂逻辑。
+"""合规行为超图 - 建模"在何种条件下,何种主体,必须/禁止执行何种操作"的复杂逻辑。
适用于合规审计、风险识别等。
"""
-from typing import Any, List, Tuple
+from typing import Any, List
from pydantic import BaseModel, Field
from langchain_core.language_models import BaseChatModel
from langchain_core.embeddings import Embeddings
from hyperextract.types import AutoHypergraph
-class ComplianceEntity(BaseModel):
- """合规实体节点"""
- name: str = Field(description="实体名称")
- type: str = Field(description="实体类型:主体、操作、条件、约束、其他")
+class ComplianceElement(BaseModel):
+ """合规要素 - 节点"""
+ name: str = Field(description="要素名称")
+ type: str = Field(description="要素类型:主体、操作、条件、其他")
description: str = Field(description="简要描述", default="")
class ComplianceRule(BaseModel):
- """合规规则超边(连接多个实体)"""
- ruleType: str = Field(description="规则类型:必须、禁止、允许、其他")
- participants: List[str] = Field(description="参与实体名称列表")
- condition: str = Field(description="适用条件", default="")
- action: str = Field(description="操作内容")
- details: str = Field(description="详细说明", default="")
+ """合规规则 - 超边"""
+ rule_type: str = Field(description="规则类型:必须、禁止、允许、其他")
+ action: str = Field(description="要求执行或禁止的操作内容")
+ condition: str = Field(description="适用条件/前置要求", default="")
+ penalty: str = Field(description="违规后果/处罚措施", default="")
+ source_clause: str = Field(description="规则来源条款", default="")
+ notes: str = Field(description="补充说明", default="")
+ participants: List[str] = Field(description="参与要素名称列表")
-_PROMPT = """你是一位专业的合规分析专家。请从文本中提取合规实体和规则,建模“在何种条件下,何种主体,必须/禁止执行何种操作”的复杂逻辑,构建合规行为超图。
+_PROMPT = """你是一位专业的合规分析专家。请从文本中提取合规要素和规则,建模"在何种条件下,何种主体,必须/禁止执行何种操作"的复杂逻辑,构建合规行为超图。
### 节点提取规则
-1. 提取所有合规相关实体
-2. 为每个实体指定类型:主体、操作、条件、约束、其他
-3. 为每个实体添加简要描述
+1. 提取规则中涉及的要素作为节点
+2. 为每个要素指定类型:主体、操作、条件、其他
+3. 为每个要素添加简要描述
### 超边提取规则
-1. 仅从提取的实体中创建合规规则超边
+1. 仅从提取的要素中创建合规规则(超边)
2. 指定规则类型:必须、禁止、允许、其他
-3. 列出所有参与实体(主体、操作、条件等)
-4. 描述适用条件(如有)
-5. 描述操作内容
-6. 添加详细说明(如有)
+3. 填写要求执行或禁止的操作内容(必须)
+4. 提取适用条件/前置要求(如有)
+5. 提取违规后果/处罚措施(如有)
+6. 提取规则来源条款(如有)
+7. 添加补充说明(如有)
+8. 列出所有参与要素名称(必须)
### 约束条件
- 确保规则逻辑准确
- 保持客观准确,不添加文本中没有的信息
+- 每条超边必须包含 rule_type、action 和 participants
+- 其他字段尽量填充,缺失的使用空字符串
### 源文本:
"""
-_NODE_PROMPT = """你是一位专业的合规实体识别专家。请从文本中提取所有合规相关实体作为节点。
+_NODE_PROMPT = """你是一位专业的合规要素识别专家。请从文本中提取所有合规相关要素作为节点。
### 提取规则
-1. 提取所有合规相关实体
-2. 为每个实体指定类型:主体、操作、条件、约束、其他
-3. 为每个实体添加简要描述
+1. 提取规则中涉及的要素
+2. 为每个要素指定类型:主体、操作、条件、其他
+3. 为每个要素添加简要描述
### 源文本:
"""
-_EDGE_PROMPT = """你是一位专业的合规规则提取专家。请从给定实体列表中提取合规规则超边。
+_EDGE_PROMPT = """你是一位专业的合规规则提取专家。请从给定节点(要素)列表中提取合规规则超边。
### 约束条件
-1. 仅从下方已知实体列表中提取超边
-2. 不要创建未列出的实体
+1. 仅从下方已知要素列表中提取规则(超边)
+2. 不要创建未列出的要素
+3. 每条规则必须包含 rule_type、action 和 participants
+4. 其他字段尽量填充,缺失的使用空字符串
"""
-class ComplianceLogic(AutoHypergraph[ComplianceEntity, ComplianceRule]):
+class ComplianceLogic(AutoHypergraph[ComplianceElement, ComplianceRule]):
"""
适用文档: 合规指南、行政法规、公司管理制度
功能介绍:
- 建模“在何种条件下,何种主体,必须/禁止执行何种操作”的复杂逻辑。适用于合规审计、风险识别等。
+ 建模"在何种条件下,何种主体,必须/禁止执行何种操作"的复杂逻辑。适用于合规审计、风险识别等。
Example:
>>> template = ComplianceLogic(llm_client=llm, embedder=embedder)
@@ -92,7 +100,7 @@ class ComplianceLogic(AutoHypergraph[ComplianceEntity, ComplianceRule]):
):
"""
初始化合规行为超图模板。
-
+
Args:
llm_client: LLM 客户端,用于知识提取
embedder: 嵌入模型,用于语义检索
@@ -103,10 +111,10 @@ class ComplianceLogic(AutoHypergraph[ComplianceEntity, ComplianceRule]):
**kwargs: 其他技术参数,传递给基类
"""
super().__init__(
- node_schema=ComplianceEntity,
+ node_schema=ComplianceElement,
edge_schema=ComplianceRule,
node_key_extractor=lambda x: x.name,
- edge_key_extractor=lambda x: f"{x.ruleType}_{sorted(x.participants)}",
+ edge_key_extractor=lambda x: f"{sorted(x.participants)}_{x.rule_type}_{x.action}",
nodes_in_edge_extractor=lambda x: tuple(x.participants),
llm_client=llm_client,
embedder=embedder,
@@ -127,17 +135,17 @@ class ComplianceLogic(AutoHypergraph[ComplianceEntity, ComplianceRule]):
):
"""
展示合规行为超图。
-
+
Args:
top_k_for_search: 语义检索返回的节点/边数量,默认为 3
top_k_for_chat: 问答使用的节点/边数量,默认为 3
"""
- def node_label_extractor(node: ComplianceEntity) -> str:
+ def node_label_extractor(node: ComplianceElement) -> str:
return f"{node.name} ({node.type})"
-
+
def edge_label_extractor(edge: ComplianceRule) -> str:
- return edge.ruleType
-
+ return f"[{edge.rule_type}] {edge.action}"
+
super().show(
node_label_extractor=node_label_extractor,
edge_label_extractor=edge_label_extractor,