File size: 10,603 Bytes
a453c29 e97ac87 a453c29 e97ac87 a453c29 e97ac87 a453c29 e97ac87 a453c29 e97ac87 a453c29 e97ac87 a453c29 e97ac87 a453c29 e97ac87 a453c29 e97ac87 a453c29 |
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 |
"""Service for finding similar topics using Google Generative AI embeddings"""
import logging
import json
import hashlib
from pathlib import Path
from typing import Optional, List, Dict
from datetime import datetime
import numpy as np
from sklearn.metrics.pairwise import cosine_similarity
from google import genai
from google.genai import types
from config import GOOGLE_API_KEY, PROJECT_ROOT
logger = logging.getLogger(__name__)
# Paths for topics and cache files
TOPICS_FILE = PROJECT_ROOT / "data" / "topics.json"
EMBEDDINGS_CACHE_FILE = PROJECT_ROOT / "data" / "topic_embeddings_cache.json"
class TopicSimilarityService:
"""Service for finding the most similar topic from a predefined list using embeddings"""
def __init__(self):
self.client = None
self.topics = []
self.topic_embeddings = None
self.initialized = False
self.model_name = "models/text-embedding-004"
def initialize(self):
"""Initialize the Google Generative AI client and load topic embeddings"""
if self.initialized:
logger.info("Topic similarity service already initialized")
return
if not GOOGLE_API_KEY:
raise ValueError("GOOGLE_API_KEY not found in environment variables")
try:
logger.info("Initializing topic similarity service with Google Generative AI")
# Create Google Generative AI client
self.client = genai.Client(api_key=GOOGLE_API_KEY)
# Load topics
self.topics = self._load_topics()
logger.info(f"Loaded {len(self.topics)} topics from {TOPICS_FILE}")
# Load or generate topic embeddings
self.topic_embeddings = self._get_topic_embeddings()
logger.info(f"Loaded {len(self.topic_embeddings)} topic embeddings")
self.initialized = True
logger.info("✓ Topic similarity service initialized successfully")
except Exception as e:
logger.error(f"Error initializing topic similarity service: {str(e)}")
raise RuntimeError(f"Failed to initialize topic similarity service: {str(e)}")
def _load_topics(self) -> List[str]:
"""Load topics from topics.json file"""
# Ensure path is absolute
topics_file = Path(TOPICS_FILE).resolve()
if not topics_file.exists():
raise FileNotFoundError(
f"Topics file not found: {topics_file}\n"
f"Current working directory: {Path.cwd()}\n"
f"PROJECT_ROOT: {PROJECT_ROOT}\n"
f"TOPICS_FILE path: {TOPICS_FILE}"
)
try:
with open(topics_file, 'r', encoding='utf-8') as f:
data = json.load(f)
topics = data.get("topics", [])
if not topics:
raise ValueError(f"No topics found in {topics_file}")
return topics
except (json.JSONDecodeError, KeyError) as e:
raise ValueError(f"Error loading topics from {topics_file}: {str(e)}")
def _get_topics_hash(self, topics: List[str]) -> str:
"""Generate a hash of the topics list to verify cache validity"""
topics_str = json.dumps(topics, sort_keys=True)
return hashlib.md5(topics_str.encode('utf-8')).hexdigest()
def _load_cached_embeddings(self) -> Optional[np.ndarray]:
"""Load cached topic embeddings if they exist and are valid"""
# Ensure path is absolute
cache_file = Path(EMBEDDINGS_CACHE_FILE).resolve()
if not cache_file.exists():
return None
try:
with open(cache_file, 'r', encoding='utf-8') as f:
cache_data = json.load(f)
# Verify cache is valid by checking topics hash
current_hash = self._get_topics_hash(self.topics)
if cache_data.get("topics_hash") == current_hash:
# Convert list embeddings back to numpy arrays
embeddings = [np.array(emb) for emb in cache_data.get("embeddings", [])]
logger.info(f"Loaded {len(embeddings)} topic embeddings from cache")
return np.array(embeddings)
else:
# Topics have changed, cache is invalid
logger.info("Topics have changed, cache is invalid")
return None
except (json.JSONDecodeError, KeyError, ValueError) as e:
logger.warning(f"Could not load cached embeddings: {e}")
return None
def _save_cached_embeddings(self, embeddings: np.ndarray):
"""Save topic embeddings to cache file"""
topics_hash = self._get_topics_hash(self.topics)
# Convert numpy arrays to lists for JSON serialization
embeddings_list = [emb.tolist() for emb in embeddings]
cache_data = {
"topics_hash": topics_hash,
"embeddings": embeddings_list,
"model": self.model_name,
"cached_at": datetime.now().isoformat()
}
try:
# Ensure path is absolute and directory exists
cache_file = Path(EMBEDDINGS_CACHE_FILE).resolve()
cache_file.parent.mkdir(parents=True, exist_ok=True)
with open(cache_file, 'w', encoding='utf-8') as f:
json.dump(cache_data, f, indent=2)
logger.info(f"Cached {len(embeddings)} topic embeddings to {cache_file}")
except Exception as e:
logger.warning(f"Could not save cached embeddings: {e}")
def _get_topic_embeddings(self) -> np.ndarray:
"""
Get topic embeddings, loading from cache if available, otherwise generating and caching them
Returns:
numpy.ndarray: Array of topic embeddings
"""
# Try to load from cache first
cached_embeddings = self._load_cached_embeddings()
if cached_embeddings is not None:
return cached_embeddings
# Cache miss or invalid - generate embeddings
logger.info(f"Generating embeddings for {len(self.topics)} topics (this may take a moment)...")
try:
embedding_response = self.client.models.embed_content(
model=self.model_name,
contents=self.topics,
config=types.EmbedContentConfig(task_type="SEMANTIC_SIMILARITY")
)
if not hasattr(embedding_response, "embeddings") or embedding_response.embeddings is None:
raise RuntimeError("Embedding API did not return embeddings")
embeddings = [np.array(e.values) for e in embedding_response.embeddings]
embeddings_array = np.array(embeddings)
# Save to cache for future use
self._save_cached_embeddings(embeddings_array)
return embeddings_array
except Exception as e:
logger.error(f"Error generating topic embeddings: {str(e)}")
raise RuntimeError(f"Failed to generate topic embeddings: {str(e)}")
def find_most_similar_topic(self, input_text: str) -> Dict[str, any]:
"""
Compare a single input text to all topics and return the highest cosine similarity
Args:
input_text: The text to compare against topics
Returns:
dict: Contains 'topic', 'similarity', and 'index' of the most similar topic
"""
if not self.initialized:
self.initialize()
if not input_text or not isinstance(input_text, str):
raise ValueError("Input text must be a non-empty string")
input_text = input_text.strip()
if len(input_text) == 0:
raise ValueError("Input text cannot be empty")
if not self.topics:
raise ValueError("No topics found in topics.json")
try:
# Embed the input text
embedding_response = self.client.models.embed_content(
model=self.model_name,
contents=[input_text],
config=types.EmbedContentConfig(task_type="SEMANTIC_SIMILARITY")
)
if not hasattr(embedding_response, "embeddings") or embedding_response.embeddings is None:
raise RuntimeError("Embedding API did not return embeddings")
# Extract input embedding
input_embedding = np.array(embedding_response.embeddings[0].values).reshape(1, -1)
# Calculate cosine similarity between input and each topic
similarities = cosine_similarity(input_embedding, self.topic_embeddings)[0]
# Find the highest similarity
max_index = np.argmax(similarities)
max_similarity = similarities[max_index]
most_similar_topic = self.topics[max_index]
return {
"topic": most_similar_topic,
"similarity": float(max_similarity),
"index": int(max_index)
}
except Exception as e:
logger.error(f"Error finding similar topic: {str(e)}")
raise RuntimeError(f"Failed to find similar topic: {str(e)}")
def batch_find_similar_topics(self, input_texts: List[str]) -> List[str]:
"""
Find the most similar topic for each input text
Args:
input_texts: List of input texts to compare against topics
Returns:
List of most similar topics (one per input text)
"""
if not self.initialized:
self.initialize()
if not input_texts or not isinstance(input_texts, list):
raise ValueError("Input texts must be a non-empty list")
results = []
for text in input_texts:
try:
result = self.find_most_similar_topic(text)
results.append(result["topic"])
except Exception as e:
logger.error(f"Error finding similar topic for text '{text[:50]}...': {str(e)}")
results.append(None) # Or raise, depending on desired behavior
return results
# Initialize singleton instance
topic_similarity_service = TopicSimilarityService()
|