File size: 11,778 Bytes
ea99abb |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 |
from typing import List, Dict, Any, Tuple, Optional
import spacy
import networkx as nx
import matplotlib.pyplot as plt
from io import BytesIO
import base64
import re
import json
from langchain_core.messages import HumanMessage
from langchain.chat_models import init_chat_model
from dotenv import load_dotenv
import os
# Interactive visualization
from pyvis.network import Network
# Load environment variables
_ = load_dotenv()
class LLMKnowledgeGraph:
def __init__(self, model: str = "gemini-2.0-flash", model_provider: str = "google_genai"):
"""Initialize the LLM for knowledge graph generation."""
self.llm = init_chat_model(
model=model,
model_provider=model_provider,
temperature=0.1, # Lower temperature for more deterministic results
max_tokens=2000
)
self.entity_prompt = """
Extract all named entities from the following text and categorize them into the following types:
- PERSON: People, including fictional
- ORG: Companies, agencies, institutions, etc.
- GPE: Countries, cities, states
- DATE: Absolute or relative dates or periods
- MONEY: Monetary values
- PERCENT: Percentage values
- QUANTITY: Measurements, weights, distances
- EVENT: Named hurricanes, battles, wars, sports events, etc.
- WORK_OF_ART: Titles of books, songs, etc.
- LAW: Legal document titles
- LANGUAGE: Any named language
Return the entities in JSON format with the following structure:
[
{"text": "entity text", "label": "ENTITY_TYPE", "start": character_start, "end": character_end}
]
Text: """
self.relation_prompt = """
Analyze the following text and extract relationships between entities in the form of subject-relation-object triples.
For each relation, provide:
- The subject (entity that is the source of the relation)
- The relation type (e.g., 'works at', 'located in', 'part of')
- The object (entity that is the target of the relation)
Return the relations in JSON format with the following structure:
[
{"subject": "subject text", "relation": "relation type", "object": "object text"}
]
Text: """
def extract_entities_with_llm(self, text: str) -> List[Dict[str, Any]]:
"""Extract entities from text using LLM."""
try:
response = self.llm.invoke([HumanMessage(content=self.entity_prompt + text)])
# Handle case where response might be a string or a message object
if hasattr(response, 'content'):
content = response.content
else:
content = str(response)
# Clean the response to ensure it's valid JSON
content = content.strip()
if content.startswith('```json'):
content = content[content.find('['):content.rfind(']')+1]
elif content.startswith('['):
content = content[:content.rfind(']')+1]
entities = json.loads(content)
return entities
except Exception as e:
print(f"Error extracting entities with LLM: {str(e)}")
print(f"Response content: {getattr(response, 'content', str(response))}")
return []
def extract_relations_with_llm(self, text: str) -> List[Dict[str, str]]:
"""Extract relations between entities using LLM."""
try:
response = self.llm.invoke([HumanMessage(content=self.relation_prompt + text)])
# Handle case where response might be a string or a message object
if hasattr(response, 'content'):
content = response.content
else:
content = str(response)
# Clean the response to ensure it's valid JSON
content = content.strip()
if content.startswith('```json'):
content = content[content.find('['):content.rfind(']')+1]
elif content.startswith('['):
content = content[:content.rfind(']')+1]
relations = json.loads(content)
return relations
except Exception as e:
print(f"Error extracting relations with LLM: {str(e)}")
print(f"Response content: {getattr(response, 'content', str(response))}")
return []
def extract_relations(text: str, model_name: str = "gemini-2.0-flash", use_llm: bool = True) -> Dict[str, Any]:
"""
Extract entities and their relations from text to build a knowledge graph.
Args:
text: Input text to process
model_name: Name of the model to use (spaCy model or LLM)
use_llm: Whether to use LLM for relation extraction (default: True)
Returns:
Dictionary containing nodes and edges for the knowledge graph
"""
if use_llm:
# Use LLM for both entity and relation extraction
kg_extractor = LLMKnowledgeGraph(model=model_name)
# Extract entities using LLM
entities = kg_extractor.extract_entities_with_llm(text)
# Extract relations using LLM
relations = kg_extractor.extract_relations_with_llm(text)
else:
# Fallback to spaCy for entity and relation extraction
try:
nlp = spacy.load(model_name)
except OSError:
# If model is not found, download it
import subprocess
import sys
subprocess.check_call([sys.executable, "-m", "spacy", "download", model_name])
nlp = spacy.load(model_name)
# Process the text
doc = nlp(text)
# Extract entities
entities = [{"text": ent.text, "label": ent.label_, "start": ent.start_char, "end": ent.end_char}
for ent in doc.ents]
# Extract relations (subject-verb-object)
relations = []
for sent in doc.sents:
for token in sent:
if token.dep_ in ("ROOT", "nsubj", "dobj"):
subj = ""
obj = ""
relation = ""
# Find subject
if token.dep_ == "nsubj" and token.head.pos_ == "VERB":
subj = token.text
relation = token.head.lemma_
# Find object
for child in token.head.children:
if child.dep_ == "dobj":
obj = child.text
break
if subj and obj and relation:
relations.append({
"subject": subj,
"relation": relation,
"object": obj
})
return {
"entities": entities,
"relations": relations
}
def build_nx_graph(entities: List[Dict], relations: List[Dict]) -> nx.DiGraph:
"""Build a NetworkX DiGraph from entities and relations. Ensure all nodes have a 'label'."""
G = nx.DiGraph()
# Add entities as nodes
for entity in entities:
label = entity.get("label") or entity.get("type") or "ENTITY"
text = entity.get("text") or entity.get("word")
G.add_node(text, label=label, type="entity")
# Add edges and ensure nodes exist with label
for rel in relations:
subj = rel.get("subject")
obj = rel.get("object")
rel_label = rel.get("relation", "related_to")
if subj is not None and subj not in G:
G.add_node(subj, label="ENTITY", type="entity")
if obj is not None and obj not in G:
G.add_node(obj, label="ENTITY", type="entity")
G.add_edge(subj, obj, label=rel_label)
return G
def visualize_knowledge_graph(entities: List[Dict], relations: List[Dict]) -> str:
"""
Generate a static PNG visualization of the knowledge graph, returned as base64 string for HTML embedding.
"""
G = build_nx_graph(entities, relations)
plt.figure(figsize=(12, 8))
pos = nx.spring_layout(G, k=0.5, iterations=50)
# Color nodes by entity type
entity_types = list(set([d.get('label', 'ENTITY') for n, d in G.nodes(data=True)]))
color_map = {etype: plt.cm.tab20(i % 20) for i, etype in enumerate(entity_types)}
node_colors = [color_map[d.get('label', 'ENTITY')] for n, d in G.nodes(data=True)]
nx.draw_networkx_nodes(G, pos, node_size=2000, node_color=node_colors, alpha=0.8)
nx.draw_networkx_edges(G, pos, edge_color='gray', arrows=True, arrowsize=20)
nx.draw_networkx_labels(G, pos, font_size=10, font_weight='bold')
edge_labels = {(u, v): d['label'] for u, v, d in G.edges(data=True)}
nx.draw_networkx_edge_labels(G, pos, edge_labels=edge_labels, font_size=8)
buf = BytesIO()
plt.savefig(buf, format='png', bbox_inches='tight')
plt.close()
img_str = base64.b64encode(buf.getvalue()).decode('utf-8')
return f"data:image/png;base64,{img_str}"
def visualize_knowledge_graph_interactive(entities: List[Dict], relations: List[Dict]) -> str:
"""
Generate an interactive HTML visualization of the knowledge graph using pyvis.
Returns HTML as a string for embedding in Gradio or web UI.
"""
G = build_nx_graph(entities, relations)
net = Network(height="600px", width="100%", directed=True, notebook=False)
# Color map for entity types
entity_types = list(set([d.get('label', 'ENTITY') for n, d in G.nodes(data=True)]))
color_palette = ["#e3f2fd", "#e8f5e9", "#fff8e1", "#f3e5f5", "#e8eaf6", "#e0f7fa", "#f1f8e9", "#fce4ec", "#e8f5e9", "#f5f5f5", "#fafafa", "#e1f5fe", "#fff3e0", "#d7ccc8", "#f9fbe7", "#fbe9e7", "#ede7f6", "#e0f2f1"]
color_map = {etype: color_palette[i % len(color_palette)] for i, etype in enumerate(entity_types)}
for n, d in G.nodes(data=True):
label = d.get('label', 'ENTITY')
net.add_node(n, label=n, title=f"{n}<br>Type: {label}", color=color_map[label])
for u, v, d in G.edges(data=True):
net.add_edge(u, v, label=d['label'], title=d['label'])
net.set_options('''var options = { "edges": { "arrows": {"to": {"enabled": true}}, "color": {"color": "#888"} }, "nodes": { "font": {"size": 18} }, "physics": { "enabled": true } };''')
html_buf = BytesIO()
net.write_html(html_buf)
html_buf.seek(0)
html = html_buf.read().decode('utf-8')
# Remove <html>, <body> wrappers to allow embedding in Gradio
body_start = html.find('<body>') + len('<body>')
body_end = html.find('</body>')
body_content = html[body_start:body_end]
return body_content
def build_knowledge_graph(text: str, model_name: str = "gemini-2.0-flash", use_llm: bool = True) -> Dict[str, Any]:
"""
Main function to build a knowledge graph from text.
Args:
text: Input text to process
model_name: Name of the model to use (spaCy model or LLM)
use_llm: Whether to use LLM for relation extraction (default: True)
Returns:
Dictionary containing the knowledge graph data and visualization
"""
# Extract entities and relations
result = extract_relations(text, model_name, use_llm)
# Generate visualization
if result.get("entities") and result.get("relations"):
visualization = visualize_knowledge_graph(result["entities"], result["relations"])
result["visualization"] = visualization
else:
result["visualization"] = None
return result
|