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: 更新搜索接口,支持分别指定节点和边的返回数量;添加多个示例查询

yifanfeng97181590364858
diff --git a/examples/auto_graph_demo.py b/examples/auto_graph_demo.py
index 46918a8..419691c 100644
--- a/examples/auto_graph_demo.py
+++ b/examples/auto_graph_demo.py
@@ -349,7 +349,7 @@ def run_demo():
print(f"\n❓ Query: {q1}")
try:
# 默认同时搜索节点和边,返回 Tuple[nodes, edges]
- nodes, edges = graph.search(q1, top_k=3)
+ nodes, edges = graph.search(q1, top_k_nodes=3, top_k_edges=3)
interactions = edges
if interactions:
for i in interactions:
@@ -363,7 +363,7 @@ def run_demo():
q2 = "谁是治疗职业的?"
print(f"\n❓ Query: {q2}")
try:
- nodes, edges = graph.search(q2, top_k=3)
+ nodes, edges = graph.search(q2, top_k_nodes=3, top_k_edges=3)
healers = nodes
if healers:
for h in healers:
@@ -377,7 +377,7 @@ def run_demo():
q3 = "谁和谁发生了冲突?"
print(f"\n❓ Query: {q3}")
try:
- nodes, edges = graph.search(q3, top_k=3)
+ nodes, edges = graph.search(q3, top_k_nodes=3, top_k_edges=3)
conflicts = edges
if conflicts:
for c in conflicts:
@@ -391,7 +391,7 @@ def run_demo():
q4 = "谁是主坦克?"
print(f"\n❓ Query: {q4}")
try:
- nodes, edges = graph.search(q4, top_k=3)
+ nodes, edges = graph.search(q4, top_k_nodes=3, top_k_edges=3)
tanks = nodes
if tanks:
for t in tanks:
diff --git a/examples/auto_spatial_graph_demo.py b/examples/auto_spatial_graph_demo.py
index e2e3014..6f7d957 100644
--- a/examples/auto_spatial_graph_demo.py
+++ b/examples/auto_spatial_graph_demo.py
@@ -247,7 +247,7 @@ def run_demo():
q1 = "空间站里有哪些怪物或威胁?它们在哪?"
print(f"\n❓ Search Query: {q1}")
try:
- nodes, edges = graph.search(q1, top_k=5)
+ nodes, edges = graph.search(q1, top_k_nodes=5, top_k_edges=5)
if nodes:
print(f" 📍 相关实体:")
for n in nodes:
@@ -265,7 +265,7 @@ def run_demo():
q2 = "黑匣子在哪里找到的?"
print(f"\n❓ Search Query: {q2}")
try:
- nodes, edges = graph.search(q2, top_k=3)
+ nodes, edges = graph.search(q2, top_k_nodes=3, top_k_edges=3)
if nodes:
print(f" 📍 相关实体:")
for n in nodes:
@@ -283,7 +283,7 @@ def run_demo():
q3 = "动力反应堆舱(底层)里有什么东西?"
print(f"\n❓ Search Query: {q3}")
try:
- nodes, edges = graph.search(q3, top_k=5)
+ nodes, edges = graph.search(q3, top_k_nodes=5, top_k_edges=5)
if nodes:
print(f" 📍 相关实体:")
for n in nodes:
@@ -301,7 +301,7 @@ def run_demo():
q4 = "权限卡在哪里?"
print(f"\n❓ Search Query: {q4}")
try:
- nodes, edges = graph.search(q4, top_k=3)
+ nodes, edges = graph.search(q4, top_k_nodes=3, top_k_edges=3)
if nodes:
print(f" 📍 相关实体:")
for n in nodes:
diff --git a/examples/auto_temporal_graph_demo.py b/examples/auto_temporal_graph_demo.py
index e48512d..5d56917 100644
--- a/examples/auto_temporal_graph_demo.py
+++ b/examples/auto_temporal_graph_demo.py
@@ -219,7 +219,7 @@ def run_demo():
q1 = "Elias做了什么?"
print(f"\n❓ Query: {q1}")
try:
- nodes, edges = graph.search(q1, top_k=5)
+ nodes, edges = graph.search(q1, top_k_nodes=5, top_k_edges=5)
if edges:
for e in edges:
print(f" ⚡ {e}")
@@ -232,7 +232,7 @@ def run_demo():
q2 = "发生在地铁站的事件有哪些?"
print(f"\n❓ Query: {q2}")
try:
- nodes, edges = graph.search(q2, top_k=5)
+ nodes, edges = graph.search(q2, top_k_nodes=5, top_k_edges=5)
if edges:
for e in edges:
print(f" 📍 {e}")
@@ -245,7 +245,7 @@ def run_demo():
q3 = "Viktor和Elias之间的冲突"
print(f"\n❓ Query: {q3}")
try:
- nodes, edges = graph.search(q3, top_k=5)
+ nodes, edges = graph.search(q3, top_k_nodes=5, top_k_edges=5)
if edges:
for e in edges:
print(f" 🔥 {e}")
@@ -258,7 +258,7 @@ def run_demo():
q4 = "下午的行动"
print(f"\n❓ Query: {q4}")
try:
- nodes, edges = graph.search(q4, top_k=5)
+ nodes, edges = graph.search(q4, top_k_nodes=5, top_k_edges=5)
if edges:
for e in edges:
print(f" ⏰ {e}")
@@ -271,7 +271,7 @@ def run_demo():
q5 = "CIA和KGB分别做了什么?"
print(f"\n❓ Query: {q5}")
try:
- nodes, edges = graph.search(q5, top_k=5)
+ nodes, edges = graph.search(q5, top_k_nodes=5, top_k_edges=5)
if nodes:
print(f" 🏢 相关实体:")
for n in nodes:
diff --git a/examples/hyper_rag_demo.py b/examples/hyper_rag_demo.py
index 3d53a4a..3ac1ea6 100644
--- a/examples/hyper_rag_demo.py
+++ b/examples/hyper_rag_demo.py
@@ -236,9 +236,10 @@ if __name__ == "__main__":
print("Engaging in multi-turn dialogue based on extracted knowledge...\n")
chat_queries = [
- "What are the main themes in this text?",
- "How are different topics interconnected?",
- "What insights can you draw from the relationships in this graph?"
+ "虚空水晶现在在哪里?",
+ "黑蛇和议员Valerius有什么阴谋?",
+ "谁参与了泰坦空间站的秘密会议?",
+ "Leona目前的状况如何?",
]
for q in chat_queries:
diff --git a/hyperextract/graphs/base.py b/hyperextract/graphs/base.py
index 77ff6ff..2af3720 100644
--- a/hyperextract/graphs/base.py
+++ b/hyperextract/graphs/base.py
@@ -661,55 +661,47 @@ class AutoGraph(BaseAutoType[AutoGraphSchema[NodeSchema, EdgeSchema]], Generic[N
def search(
self,
query: str,
- top_k: int = 3,
- search_nodes: bool = True,
- search_edges: bool = True,
- ) -> List[NodeSchema] | List[EdgeSchema] | Tuple[List[NodeSchema], List[EdgeSchema]]:
+ top_k_nodes: int = 3,
+ top_k_edges: int = 3,
+ ) -> Tuple[List[NodeSchema], List[EdgeSchema]]:
"""Unified graph search interface.
- By default, searches both nodes and edges simultaneously to provide comprehensive results.
- When searching both, returns a tuple of (nodes_results, edges_results).
- When searching one type only, returns a flat list.
+ Retrieves nodes and edges semantically related to the query.
+ Always returns a tuple of (nodes, edges). If a count is 0, the corresponding list will be empty.
Args:
query: Search query string.
- top_k: Number of results to return per type (default: 3).
- search_nodes: Whether to search nodes (default: True).
- search_edges: Whether to search edges (default: True).
+ top_k_nodes: Number of node results to return (default: 3). Set to 0 to disable node search.
+ top_k_edges: Number of edge results to return (default: 3). Set to 0 to disable edge search.
Returns:
- - Tuple[List[NodeSchema], List[EdgeSchema]] if both search_nodes and search_edges are True (default).
- - List[NodeSchema] if only search_nodes is True.
- - List[EdgeSchema] if only search_edges is True.
+ Tuple[List[NodeSchema], List[EdgeSchema]]: A tuple containing:
+ - List of matching nodes (empty if top_k_nodes <= 0)
+ - List of matching edges (empty if top_k_edges <= 0)
Raises:
- ValueError: If neither search_nodes nor search_edges is True.
+ ValueError: If both top_k_nodes and top_k_edges are <= 0.
+ ValueError: If search is requested (top_k > 0) but the corresponding index is not built.
"""
- if not search_nodes and not search_edges:
+ if top_k_nodes <= 0 and top_k_edges <= 0:
raise ValueError(
- "At least one of search_nodes or search_edges must be True."
+ "At least one of top_k_nodes or top_k_edges must be positive."
)
- if search_nodes and search_edges:
- if not self._node_memory.has_index():
- raise ValueError("Node index not built. Call build_index() first.")
- if not self._edge_memory.has_index():
- raise ValueError("Edge index not built. Call build_index() first.")
- nodes = self.search_nodes(query, top_k)
- edges = self.search_edges(query, top_k)
- return (nodes, edges)
+ nodes: List[NodeSchema] = []
+ edges: List[EdgeSchema] = []
- if search_nodes:
+ if top_k_nodes > 0:
if not self._node_memory.has_index():
raise ValueError("Node index not built. Call build_index() first.")
- nodes = self.search_nodes(query, top_k)
- return nodes
+ nodes = self.search_nodes(query, top_k=top_k_nodes)
- if search_edges:
+ if top_k_edges > 0:
if not self._edge_memory.has_index():
raise ValueError("Edge index not built. Call build_index() first.")
- edges = self.search_edges(query, top_k)
- return edges
+ edges = self.search_edges(query, top_k=top_k_edges)
+
+ return nodes, edges
def search_nodes(self, query: str, top_k: int = 3) -> List[NodeSchema]:
"""Semantic search for nodes/entities only.
diff --git a/hyperextract/hypergraphs/base.py b/hyperextract/hypergraphs/base.py
index 6ef7892..67bde4c 100644
--- a/hyperextract/hypergraphs/base.py
+++ b/hyperextract/hypergraphs/base.py
@@ -598,53 +598,47 @@ class AutoHypergraph(
def search(
self,
query: str,
- top_k: int = 3,
- search_nodes: bool = True,
- search_edges: bool = True,
- ) -> List[Node] | List[Edge] | Tuple[List[Node], List[Edge]]:
+ top_k_nodes: int = 3,
+ top_k_edges: int = 3,
+ ) -> Tuple[List[Node], List[Edge]]:
"""Unified hypergraph search interface.
- By default, searches both nodes and hyperedges simultaneously to provide comprehensive results.
- When searching both, returns a tuple of (nodes_results, edges_results).
- When searching one type only, returns a flat list.
+ Retrieves nodes and hyperedges semantically related to the query.
+ Always returns a tuple of (nodes, edges). If a count is 0, the corresponding list will be empty.
Args:
query: Search query string.
- top_k: Number of results to return per type (default: 3).
- search_nodes: Whether to search nodes (default: True).
- search_edges: Whether to search hyperedges (default: True).
+ top_k_nodes: Number of node results to return (default: 3). Set to 0 to disable node search.
+ top_k_edges: Number of hyperedge results to return (default: 3). Set to 0 to disable hyperedge search.
Returns:
- - Tuple[List[Node], List[Edge]] if both search_nodes and search_edges are True (default).
- - List[Node] if only search_nodes is True.
- - List[Edge] if only search_edges is True.
+ Tuple[List[Node], List[Edge]]: A tuple containing:
+ - List of matching nodes (empty if top_k_nodes <= 0)
+ - List of matching hyperedges (empty if top_k_edges <= 0)
Raises:
- ValueError: If neither search_nodes nor search_edges is True.
+ ValueError: If both top_k_nodes and top_k_edges are <= 0.
+ ValueError: If search is requested (top_k > 0) but the corresponding index is not built.
"""
- if not search_nodes and not search_edges:
+ if top_k_nodes <= 0 and top_k_edges <= 0:
raise ValueError(
- "At least one of search_nodes or search_edges must be True."
+ "At least one of top_k_nodes or top_k_edges must be positive."
)
- if search_nodes and search_edges:
- if not self._node_memory.has_index():
- raise ValueError("Node index not built. Call build_index() first.")
- if not self._edge_memory.has_index():
- raise ValueError("Edge index not built. Call build_index() first.")
- nodes = self.search_nodes(query, top_k)
- edges = self.search_edges(query, top_k)
- return (nodes, edges)
+ nodes: List[Node] = []
+ edges: List[Edge] = []
- if search_nodes:
+ if top_k_nodes > 0:
if not self._node_memory.has_index():
raise ValueError("Node index not built. Call build_index() first.")
- return self.search_nodes(query, top_k)
+ nodes = self.search_nodes(query, top_k=top_k_nodes)
- if search_edges:
+ if top_k_edges > 0:
if not self._edge_memory.has_index():
raise ValueError("Edge index not built. Call build_index() first.")
- return self.search_edges(query, top_k)
+ edges = self.search_edges(query, top_k=top_k_edges)
+
+ return nodes, edges
def search_nodes(self, query: str, top_k: int = 3) -> List[Node]:
"""Semantic search for nodes/entities only."""
diff --git a/hyperextract/hypergraphs/hyper_rag.py b/hyperextract/hypergraphs/hyper_rag.py
index 06d7524..88f8143 100644
--- a/hyperextract/hypergraphs/hyper_rag.py
+++ b/hyperextract/hypergraphs/hyper_rag.py
@@ -6,9 +6,12 @@ Extracts and manages multi-entity relationships with support for n-ary hyperedge
from typing import List
from pydantic import BaseModel, Field
from hyperextract.hypergraphs.base import AutoHypergraph
-from langchain_core.language_models import BaseChatModel
+from langchain_core.messages import AIMessage
from langchain_core.embeddings import Embeddings
+from langchain_core.language_models import BaseChatModel
+from langchain_core.prompts.chat import ChatPromptTemplate
from ontomem.merger import CustomRuleMerger
+from ..utils.logging import logger
# ============================================================================
# Node Schema
@@ -51,18 +54,34 @@ class EdgeSchema(BaseModel):
)
+# ============================================================================
+# Query Analysis Schema
+# ============================================================================
+
+
+class QueryKeywords(BaseModel):
+ """Extracted keywords for multi-granular hypergraph retrieval."""
+
+ high_level_keywords: List[str] = Field(
+ description="Overarching concepts, themes, or abstract topics (e.g., 'International trade', 'Deforestation')."
+ )
+ low_level_keywords: List[str] = Field(
+ description="Specific entities, concrete details, or tangible objects (e.g., 'Tariffs', 'Rainforest')."
+ )
+
+
# ============================================================================
# Extraction Prompts
# ============================================================================
-HYPER_RAG_NODE_EXTRACTION_PROMPT = """
+Hyper_RAG_NODE_EXTRACTION_PROMPT = """
-Goal-
Identify relevant entities from the text.
Entities will serve as participants in complex events later.\n\n
### Source Text:
"""
-HYPER_RAG_EDGE_EXTRACTION_PROMPT = """
+Hyper_RAG_EDGE_EXTRACTION_PROMPT = """
-Goal-
You are an expert hypergraph knowledge extraction assistant.
Extract "Hyperedges" that represent relationships, events, or thematic groupings involving MULTIPLE entities (2 or more) simultaneously.
@@ -79,11 +98,45 @@ It treats all involved entities as **participants** in a shared context
"""
+Hyper_RAG_KEYWORDS_EXTRACTION_PROMPT = """---Role---
+You are a helpful assistant tasked with identifying both high-level and low-level keywords in the user's query.
+
+---Goal---
+Given the query, extract both high-level and low-level keywords.
+- High-level keywords focus on overarching concepts, themes, or abstract topics.
+- Low-level keywords focus on specific entities, concrete details, or tangible objects.
+
+---Instructions---
+- Output as a structured response with two lists.
+- High-level keywords capture broader themes and relationships.
+- Low-level keywords capture specific entities and concrete terms.
+
+---Examples---
+
+Example 1:
+Query: "How does international trade influence global economic stability?"
+High-Level: ["International trade", "Global economic stability", "Economic impact"]
+Low-Level: ["Trade agreements", "Tariffs", "Currency exchange", "Imports", "Exports"]
+
+Example 2:
+Query: "What are the environmental consequences of deforestation on biodiversity?"
+High-Level: ["Environmental consequences", "Deforestation", "Biodiversity loss"]
+Low-Level: ["Species extinction", "Habitat destruction", "Carbon emissions", "Rainforest", "Ecosystem"]
+
+Example 3:
+Query: "What is the role of education in reducing poverty?"
+High-Level: ["Education", "Poverty reduction", "Socioeconomic development"]
+Low-Level: ["School access", "Literacy rates", "Job training", "Income inequality"]
+
+---Real Data---
+Query: {query}
+"""
+
# ============================================================================
# Merge Templates for LLM.CUSTOM_RULE Strategy
# ============================================================================
-NODE_MERGE_RULE = """You are an intelligent data merging assistant.
+Hyper_RAG_NODE_MERGE_RULE = """You are an intelligent data merging assistant.
You will receive a list of objects representing the same Entity.
Your task is to merge them into a SINGLE object exactly matching the schema.
@@ -93,7 +146,7 @@ Merge strategy:
2. **description**: Synthesize a single, comprehensive description containing all unique details from the input descriptions. Write it in the third person. Resolve any contradictions coherently.
"""
-EDGE_MERGE_RULE = """You are an intelligent data merging assistant.
+Hyper_RAG_EDGE_MERGE_RULE = """You are an intelligent data merging assistant.
You will receive a list of objects representing the same Hyperedge (Relationship/Event).
Your task is to merge them into a SINGLE object exactly matching the schema.
@@ -159,7 +212,7 @@ class Hyper_RAG(AutoHypergraph[NodeSchema, EdgeSchema]):
key_extractor=node_key_fn,
llm_client=llm_client,
item_schema=NodeSchema,
- rule=NODE_MERGE_RULE,
+ rule=Hyper_RAG_NODE_MERGE_RULE,
)
# Edge merger: merges relationship descriptions using EDGE_MERGE_RULE
@@ -167,7 +220,7 @@ class Hyper_RAG(AutoHypergraph[NodeSchema, EdgeSchema]):
key_extractor=edge_key_fn,
llm_client=llm_client,
item_schema=EdgeSchema,
- rule=EDGE_MERGE_RULE,
+ rule=Hyper_RAG_EDGE_MERGE_RULE,
)
# 4. Call parent class initialization
@@ -182,8 +235,8 @@ class Hyper_RAG(AutoHypergraph[NodeSchema, EdgeSchema]):
# Enforce two-stage extraction (nodes first, then edges)
extraction_mode="two_stage",
# Inject customized prompts for extraction
- prompt_for_node_extraction=HYPER_RAG_NODE_EXTRACTION_PROMPT,
- prompt_for_edge_extraction=HYPER_RAG_EDGE_EXTRACTION_PROMPT,
+ prompt_for_node_extraction=Hyper_RAG_NODE_EXTRACTION_PROMPT,
+ prompt_for_edge_extraction=Hyper_RAG_EDGE_EXTRACTION_PROMPT,
# Pass custom merger instances
node_strategy_or_merger=node_merger,
edge_strategy_or_merger=edge_merger,
@@ -196,3 +249,133 @@ class Hyper_RAG(AutoHypergraph[NodeSchema, EdgeSchema]):
max_workers=max_workers,
verbose=verbose,
)
+
+ def chat(
+ self,
+ query: str,
+ top_k_nodes: int = 3,
+ top_k_edges: int = 3,
+ ) -> AIMessage:
+ """Performs a chat-like interaction using hypergraph knowledge.
+
+ Implements the 'Hyper-RAG' dual-level retrieval strategy:
+ 1. Analyzes the query to extract 'Low-Level' keywords (specific entities/details) and 'High-Level' keywords (overarching themes).
+ 2. Retrieves Nodes using Low-Level keywords (finding specific entities).
+ 3. Retrieves Hyperedges using High-Level keywords (finding broader relationships and themes).
+ 4. Synthesizes an answer based on the combined multi-granular context.
+
+ Args:
+ query: User query string.
+ top_k_nodes: Number of relevant nodes to retrieve (default: 3). Set to 0 to disable node context.
+ top_k_edges: Number of relevant hyperedges to retrieve (default: 3). Set to 0 to disable edge context.
+
+ Returns:
+ An AIMessage object containing the LLM-generated response.
+ Access the text content via response.content.
+
+ Example:
+ >>> # Dual-level retrieval: specific entities + broader themes
+ >>> response = hg.chat("How does education reduce poverty?", top_k_nodes=5, top_k_edges=3)
+ >>> print(response.content)
+ """
+ # Step 1: Validation
+ if top_k_nodes <= 0 and top_k_edges <= 0:
+ raise ValueError(
+ "At least one of top_k_nodes or top_k_edges must be positive."
+ )
+
+ # Step 2: Extract Keywords (Dual-Level Strategy)
+ keyword_prompt = ChatPromptTemplate.from_template(
+ Hyper_RAG_KEYWORDS_EXTRACTION_PROMPT
+ )
+ keyword_chain = keyword_prompt | self.llm_client.with_structured_output(
+ QueryKeywords
+ )
+
+ # Containers for keywords
+ low_level_targets = []
+ high_level_targets = []
+ extraction_success = False
+
+ try:
+ keywords_result: QueryKeywords = keyword_chain.invoke({"query": query})
+ if keywords_result.low_level_keywords:
+ low_level_targets = keywords_result.low_level_keywords
+ if keywords_result.high_level_keywords:
+ high_level_targets = keywords_result.high_level_keywords
+
+ extraction_success = True
+
+ if self.verbose:
+ logger.info(
+ f"Query Analysis:\n"
+ f" - Low-Level Targets: {low_level_targets}\n"
+ f" - High-Level Targets: {high_level_targets}"
+ )
+
+ except Exception as e:
+ if self.verbose:
+ logger.warning(
+ f"Keyword extraction failed ({e}), falling back to raw query."
+ )
+
+ context_parts = []
+
+ # Step 3: Retrieve and format nodes context (Using Low-Level Keywords)
+ if top_k_nodes > 0:
+ found_nodes = []
+ if extraction_success and low_level_targets:
+ # Search for EACH keyword independently
+ for kw in low_level_targets:
+ # Append ALL results directly (no deduplication)
+ found_nodes.extend(self.search_nodes(kw, top_k=top_k_nodes))
+ else:
+ # Fallback: search with original query
+ found_nodes = self.search_nodes(query, top_k=top_k_nodes)
+
+ if found_nodes:
+ context_parts.append("=== Relevant Nodes (Entity Level) ===")
+ for node in found_nodes:
+ assert isinstance(node, BaseModel), (
+ "Node must be a Pydantic BaseModel"
+ )
+ context_parts.append(node.model_dump_json(indent=2))
+
+ # Step 4: Retrieve and format edges context (Using High-Level Keywords)
+ if top_k_edges > 0:
+ found_edges = []
+ if extraction_success and high_level_targets:
+ # Search for EACH keyword independently
+ for kw in high_level_targets:
+ # Append ALL results directly (no deduplication)
+ found_edges.extend(self.search_edges(kw, top_k=top_k_edges))
+ else:
+ # Fallback: search with original query
+ found_edges = self.search_edges(query, top_k=top_k_edges)
+
+ if found_edges:
+ context_parts.append("=== Relevant Hyperedges (Thematic Level) ===")
+ for edge in found_edges:
+ assert isinstance(edge, BaseModel), (
+ "Edge must be a Pydantic BaseModel"
+ )
+ context_parts.append(edge.model_dump_json(indent=2))
+
+ # Step 5: Combine context or use fallback
+ if not context_parts:
+ context = "No relevant information found in the knowledge base."
+ else:
+ context = "\n\n".join(context_parts)
+
+ # Step 6: Invoke LLM with structured context
+ qa_prompt = ChatPromptTemplate.from_template(
+ "Based on the following Hypergraph Knowledge, answer the user's question.\n"
+ "The context includes both specific entities (Nodes) found via low-level keywords \n"
+ "and broader relationships/themes (Hyperedges) found via high-level keywords.\n\n"
+ "{context}\n\n"
+ "Question: {question}\n\n"
+ "Answer:"
+ )
+
+ qa_chain = qa_prompt | self.llm_client
+ return qa_chain.invoke({"context": context, "question": query})
diff --git a/hyperextract/templates/__init__.py b/hyperextract/templates/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/hyperextract/templates/graphs/__init__.py b/hyperextract/templates/graphs/__init__.py
new file mode 100644
index 0000000..e69de29
diff --git a/hyperextract/templates/hypergraph/__init__.py b/hyperextract/templates/hypergraph/__init__.py
new file mode 100644
index 0000000..e69de29