toy-intelligence / src /analyzer.py
dwb2023's picture
update classifier model
af4ca43
raw
history blame
8.61 kB
# src/analyzer.py
from typing import Dict, List, Any, Optional, Union
import asyncio
from concurrent.futures import ThreadPoolExecutor
from transformers import pipeline
from datetime import datetime
from .ontology import OntologyRegistry
from .relationships import RelationshipEngine
class EventAnalyzer:
"""Main analyzer class for event processing."""
def __init__(self) -> None:
"""Initialize the event analyzer with required components."""
self.ontology = OntologyRegistry()
self.relationship_engine = RelationshipEngine()
self.executor = ThreadPoolExecutor(max_workers=3)
# Initialize NLP pipelines
self.ner_pipeline = pipeline("ner", model="dbmdz/bert-large-cased-finetuned-conll03-english")
self.classifier = pipeline("zero-shot-classification", model="facebook/bart-large-mnli")
async def extract_entities(self, text: str) -> Dict[str, List[str]]:
"""Extract entities from text using NER pipeline."""
def _extract():
return self.ner_pipeline(text)
ner_results = await asyncio.get_event_loop().run_in_executor(
self.executor, _extract
)
entities = {
"people": [],
"organizations": [],
"locations": [],
"hashtags": [word for word in text.split() if word.startswith('#')]
}
for item in ner_results:
if item["entity"].endswith("PER"):
entities["people"].append(item["word"])
elif item["entity"].endswith("ORG"):
entities["organizations"].append(item["word"])
elif item["entity"].endswith("LOC"):
entities["locations"].append(item["word"])
return entities
def extract_temporal(self, text: str) -> List[str]:
"""Extract temporal expressions from text."""
return self.ontology.validate_pattern(text, 'temporal')
async def extract_locations(self, text: str) -> List[str]:
"""Extract locations using both NER and pattern matching."""
entities = await self.extract_entities(text)
ml_locations = entities.get('locations', [])
pattern_locations = self.ontology.validate_pattern(text, 'location')
return list(set(ml_locations + pattern_locations))
def calculate_confidence(self,
entities: Dict[str, List[str]],
temporal_data: List[str],
related_events: List[Any]) -> float:
"""Calculate confidence score for extracted information."""
# Base confidence from entity presence
base_confidence = min(1.0, (
0.2 * bool(entities["people"]) +
0.2 * bool(entities["organizations"]) +
0.3 * bool(entities["locations"]) +
0.3 * bool(temporal_data)
))
# Get entity parameters for frequency calculation
entity_params = [
*entities["people"],
*entities["organizations"],
*entities["locations"]
]
if not entity_params:
return base_confidence
# Calculate entity frequency boost
query = f'''
SELECT AVG(frequency) as avg_freq
FROM entities
WHERE entity_text IN ({','.join(['?']*len(entity_params))})
'''
cursor = self.relationship_engine.conn.execute(query, entity_params)
avg_frequency = cursor.fetchone()[0] or 1
frequency_boost = min(0.2, (avg_frequency - 1) * 0.05)
# Calculate relationship confidence boost
relationship_confidence = 0
if related_events:
relationship_scores = []
for event in related_events:
cursor = self.relationship_engine.conn.execute('''
SELECT COUNT(*) as shared_entities
FROM event_entities ee1
JOIN event_entities ee2 ON ee1.entity_id = ee2.entity_id
WHERE ee1.event_id = ? AND ee2.event_id = ?
''', (event[0], event[0]))
shared_count = cursor.fetchone()[0]
relationship_scores.append(min(0.3, shared_count * 0.1))
if relationship_scores:
relationship_confidence = max(relationship_scores)
return min(1.0, base_confidence + frequency_boost + relationship_confidence)
async def analyze_event(self, text: str) -> Dict[str, Any]:
"""Analyze event text and extract structured information."""
try:
# Parallel extraction
entities_future = self.extract_entities(text)
temporal_data = self.extract_temporal(text)
locations_future = self.extract_locations(text)
# Gather async results
entities, locations = await asyncio.gather(
entities_future, locations_future
)
# Merge locations and add temporal data
entities['locations'] = locations
entities['temporal'] = temporal_data
# Find related events
related_events = self.relationship_engine.find_related_events({
'text': text,
'entities': entities
})
# Calculate confidence
confidence = self.calculate_confidence(entities, temporal_data, related_events)
# Store event if confidence meets threshold
cursor = None
if confidence >= 0.6:
cursor = self.relationship_engine.conn.execute(
'INSERT INTO events (text, timestamp, confidence) VALUES (?, ?, ?)',
(text, datetime.now().isoformat(), confidence)
)
event_id = cursor.lastrowid
# Store entities and update relationships
self.relationship_engine.store_entities(event_id, {
'person': entities['people'],
'organization': entities['organizations'],
'location': entities['locations'],
'temporal': temporal_data,
'hashtag': entities['hashtags']
})
self.relationship_engine.update_entity_relationships(event_id)
self.relationship_engine.conn.commit()
# Get entity relationships for output
entity_relationships = []
if cursor and cursor.lastrowid:
entity_relationships = self.relationship_engine.get_entity_relationships(cursor.lastrowid)
return {
"text": text,
"entities": entities,
"confidence": confidence,
"verification_needed": confidence < 0.6,
"related_events": [
{
"text": event[1],
"timestamp": event[2],
"confidence": event[3],
"shared_entities": event[4] if len(event) > 4 else None
}
for event in related_events
],
"entity_relationships": entity_relationships
}
except Exception as e:
return {"error": str(e)}
def get_entity_statistics(self) -> Dict[str, List[tuple]]:
"""Get statistics about stored entities and relationships."""
stats = {}
# Entity counts by type
cursor = self.relationship_engine.conn.execute('''
SELECT entity_type, COUNT(*) as count, AVG(frequency) as avg_frequency
FROM entities
GROUP BY entity_type
''')
stats['entity_counts'] = cursor.fetchall()
# Most frequent entities
cursor = self.relationship_engine.conn.execute('''
SELECT entity_text, entity_type, frequency
FROM entities
ORDER BY frequency DESC
LIMIT 10
''')
stats['frequent_entities'] = cursor.fetchall()
# Relationship statistics
cursor = self.relationship_engine.conn.execute('''
SELECT relationship_type, COUNT(*) as count, AVG(confidence) as avg_confidence
FROM entity_relationships
GROUP BY relationship_type
''')
stats['relationship_stats'] = cursor.fetchall()
return stats