Spaces:
Running
Running
File size: 13,989 Bytes
acb800d 9b7e0d3 acb800d 9b7e0d3 acb800d 9b7e0d3 acb800d 9b7e0d3 acb800d 9b7e0d3 acb800d 9b7e0d3 acb800d |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 |
from langchain_community.vectorstores import FAISS
from langchain_google_genai import GoogleGenerativeAIEmbeddings, ChatGoogleGenerativeAI
from langchain.schema import Document
from langchain.chains import RetrievalQA
from langchain.chains.summarize import load_summarize_chain
from langchain.prompts import PromptTemplate
import aiohttp
import re
import os
from sqlalchemy.orm import Session
from ..db import crud
import dotenv
dotenv.load_dotenv()
API_KEY_COMMENTS = os.getenv('API_KEY_COMMENTS')
API_KEY_VIDEO = os.getenv('API_KEY_VIDEO')
BASE_URL = "https://www.googleapis.com/youtube/v3"
GOOGLE_API_KEY = os.getenv('GOOGLE_API_KEY')
embed_model = "gemini-embedding-001"
llm = ChatGoogleGenerativeAI(
model="gemini-flash-lite-latest",
google_api_key=GOOGLE_API_KEY,
temperature=1.0,
max_output_tokens=2048
)
tools = [{"google_search": {}}]
llm_with_grounding = llm.bind_tools(tools)
embedding_model = GoogleGenerativeAIEmbeddings(
google_api_key=GOOGLE_API_KEY,
model=embed_model
)
local_cache = {}
async def fetch_comments_data(video_id, max_results=100, order="relevance"):
url = f"{BASE_URL}/commentThreads?part=snippet&videoId={video_id}&key={API_KEY_COMMENTS}&maxResults={max_results}&order={order}"
async with aiohttp.ClientSession() as session:
async with session.get(url) as response:
if response.status == 200:
return await response.json()
return None
async def fetch_channel_details(channel_id):
url = f"{BASE_URL}/channels?part=snippet%2CcontentDetails%2Cstatistics&id={channel_id}&key={API_KEY_VIDEO}"
async with aiohttp.ClientSession() as session:
async with session.get(url) as response:
if response.status == 200:
result = await response.json()
if result["items"]:
return result["items"][0].get("statistics", {})
return {}
async def fetch_video_details(video_id):
video_details_url = f"{BASE_URL}/videos?part=snippet,statistics,contentDetails&id={video_id}&key={API_KEY_VIDEO}"
async with aiohttp.ClientSession() as session:
async with session.get(video_details_url) as response:
if response.status == 200:
result = await response.json()
if result["items"]:
video_details = result["items"][0]
return video_details
return None
async def extract_comments(video_id):
if video_id in local_cache and "Comments" in local_cache[video_id]:
print(f"Using cached comments for video ID: {video_id}")
return local_cache[video_id]["Comments"]
video = await fetch_video_details(video_id)
if not video:
return []
com_cnt = int(video['statistics']['commentCount']) if 'commentCount' in video['statistics'] else 0
all_comments = []
# Fetch comments by relevance
if com_cnt > 0:
com_data_rel = await fetch_comments_data(video_id, min(100, com_cnt), "relevance")
if com_data_rel and "items" in com_data_rel:
for item in com_data_rel["items"]:
snippet = item["snippet"]["topLevelComment"]["snippet"]
all_comments.append({
"Author": snippet["authorDisplayName"],
"CommentText": snippet["textOriginal"],
"LikeCount": snippet["likeCount"],
"PublishDate": snippet["publishedAt"],
"AuthorLogoUrl": snippet["authorProfileImageUrl"],
"SortBy": "relevance"
})
# Fetch additional comments by time if needed
if com_cnt > 100:
remaining_comments = com_cnt - len(all_comments)
if remaining_comments > 0:
com_data_time = await fetch_comments_data(video_id, min(100, remaining_comments), "time")
if com_data_time and "items" in com_data_time:
for item in com_data_time["items"]:
snippet = item["snippet"]["topLevelComment"]["snippet"]
all_comments.append({
"Author": snippet["authorDisplayName"],
"CommentText": snippet["textOriginal"],
"LikeCount": snippet["likeCount"],
"PublishDate": snippet["publishedAt"],
"AuthorLogoUrl": snippet["authorProfileImageUrl"],
"SortBy": "time"
})
# Initialize cache structure for this video
if video_id not in local_cache:
local_cache[video_id] = {}
local_cache[video_id]["Comments"] = all_comments
return all_comments
def remove_links(comment):
return re.sub(r'https?://\S+|www\.\S+', '', comment).strip()
def has_multiple_timestamps(comment):
timestamps = re.findall(r'\d{1,2}(:\d{2}){1,3}', comment)
return len(timestamps) > 3
def has_char_timestamps(comment):
pattern = r'^\s*\d{1,2}(:\d{2}){1,3}[\s\u200B]*[-:|]*[\s\u200B]+[a-zA-Z]{2,}.*$'
matches = re.findall(pattern, comment, flags=re.MULTILINE)
return len(matches) > 3
def is_code_heavy(comment):
comment = comment.strip()
high_signal_keywords = re.findall(
r'\b(def|class|return|import|lambda|function|const|var|=>|try|except|elif)\b',
comment
)
code_structures = re.findall(r'(==|===|{|}|\[|\]|::|->|=)', comment)
indented_lines = re.findall(r'^\s{4,}', comment, re.MULTILINE)
num_lines = comment.count('\n') + 1
score = 0
if len(high_signal_keywords) >= 2:
score += 2
if len(code_structures) >= 3:
score += 2
if len(indented_lines) >= 2:
score += 2
if num_lines >= 3:
score += 1
if len(high_signal_keywords) > 0 and len(code_structures) > 0:
score += 1
return score >= 5
def clean_and_filter_comment(comment):
comment = remove_links(comment)
if is_code_heavy(comment):
return None
if has_multiple_timestamps(comment) and not has_char_timestamps(comment):
return None
if len(comment.strip()) == 0:
return None
return comment.strip()
def process_comments(comment_list):
cleaned_comments = []
for comment in comment_list:
cleaned = clean_and_filter_comment(comment)
if cleaned:
cleaned_comments.append(cleaned)
return cleaned_comments
def format_comment(comment):
return f"{comment['Author']}: {comment['CommentText']} (Likes: {comment['LikeCount']})"
custom_prompt = PromptTemplate(
input_variables=["text", "title", "channel_name"],
template="""
IMPORTANT: Keep your entire response under 1000 tokens. Be concise. Focus on essential insights. Avoid over-explaining or repeating.
You are a critical and insightful assistant summarizing YouTube comments.
You are summarizing the comment section of a video titled: "{title}", published by the channel: "{channel_name}".
Your tasks are to:
1. **Summarize**: Identify and summarize the main opinions, reactions, and themes across the comments.
2. **Highlight**: Highlight praise, criticism, and any notable disagreements or debates.
3. **Spot Interesting Comments**: Mention particularly insightful, surprising, or unique perspectives, especially if they contrast with the general opinion.
4. **Fact Check**: Identify any factual claims made in the comments. For each claim:
- Evaluate whether it's accurate, misleading, false, or unverifiable.
- Reference widely accepted knowledge or consensus to support your evaluation.
- When possible, include the author's handle (e.g., @username) for clarity.
Return your output in this format:
**Summary of Opinions**: ...
**Common Themes & Sentiments**: ...
**Notable or Unique Comments**: ...
**Fact Check Notes**:
- @username: "Comment content or claim..." → ✅ True / ❌ False
- Explanation: Only if a comment is False
Comments are shown below, with author names starting with @:
{text}
Summary:
"""
)
def chunk_comments(comments, chunk_size=20):
chunks = []
for i in range(0, len(comments), chunk_size):
chunk = comments[i:i + chunk_size]
if chunk:
chunks.append(Document(page_content="\n".join(chunk)))
return chunks
def get_qa_prompt(summary):
qa_prompt = PromptTemplate(
input_variables=["context", "question"],
template=f"""
You are a sharp and knowledge-driven assistant analyzing YouTube comments.
Here is a summary of the comment section:
{summary}
Here are the most relevant comment chunks, with author names:
{{context}}
Answer the following question based on the comments, but also incorporate your broader reasoning, factual understanding, and critical thinking.
Your response should:
- **Be direct, honest, and grounded in fact and logic**.
- If commenters make false or unsupported claims, clearly point them out and explain why.
- If a comment makes a valid point or well-reasoned argument, acknowledge and explain it.
- Avoid vague disclaimers like "this is just their opinion" unless it's truly subjective with no clear reasoning path.
- Reference author handles (e.g., @username) or comment snippets for clarity when helpful.
Question: {{question}}
Answer:
"""
)
return qa_prompt
async def ensure_processed_comments(video_id):
"""Ensure comments are processed and cached for a video."""
if video_id not in local_cache:
local_cache[video_id] = {}
# Check if processed comments are already cached
if "ProcessedComments" in local_cache[video_id]:
return local_cache[video_id]["ProcessedComments"]
try:
# Get raw comments (will use cache if available)
comments = await extract_comments(video_id)
if not comments:
raise ValueError("No raw comments found for processing.")
# Process and cache the formatted/cleaned comments
formatted_comments = [format_comment(comment) for comment in comments]
cleaned_comments = process_comments(formatted_comments)
if not cleaned_comments:
raise ValueError("No valid comments found after cleaning.")
local_cache[video_id]["ProcessedComments"] = cleaned_comments
return cleaned_comments
except ValueError as ve:
raise ve
except Exception as e:
print(f"An unexpected error occurred during comment processing for {video_id}: {e}")
raise RuntimeError(f"Failed to process comments due to an internal issue: {str(e)}")
async def summarize_comments(db: Session, video_id: str, title: str = '', channel_name: str = ''):
"""Checking if Comments Summary already exist in DB"""
cached_video = crud.get_or_create_video_store(db, video_id)
if cached_video and cached_video.comment_summary:
print(f"Using cached comment summary for video ID: {video_id}")
return cached_video.comment_summary
try:
cleaned_comments = await ensure_processed_comments(video_id)
if not cleaned_comments:
raise ValueError("No valid comments found after cleaning for summarization.")
comment_text = "\n\n".join(cleaned_comments)
all_comments_docs = Document(page_content=comment_text)
summary_chain = load_summarize_chain(llm=llm_with_grounding, chain_type="stuff", prompt=custom_prompt)
response = summary_chain.invoke({
"input_documents": [all_comments_docs],
"title": title,
"channel_name": channel_name
})
summary = response['output_text'].strip()
# Cache the summary in the database
crud.update_comment_summary(db, video_id=video_id, summary=summary)
return summary
except ValueError as ve:
raise ve
except Exception as e:
print(f"An unexpected error occurred during comment summarization for {video_id}: {e}")
raise RuntimeError(f"Failed to generate summary due to an internal issue: {str(e)}")
async def answer_question(db: Session, video_id: str, question: str):
try:
"""Answer questions based on comments, using DB for summary cache."""
summary = await summarize_comments(db, video_id)
processed_comments = await ensure_processed_comments(video_id)
if not processed_comments:
raise ValueError("No comments available to answer the question after processing.")
# Check for a cached vectorstore
if local_cache.get(video_id, {}).get("comment_vectorstore"):
print(f"Using local cache for comment vectorstore for video ID: {video_id}")
vectorstore = local_cache[video_id]["comment_vectorstore"]
else:
print(f"Creating and caching comment vectorstore for video ID: {video_id}")
if not processed_comments:
raise ValueError("Cannot create vectorstore: No cleaned comments available.")
chunked_docs = chunk_comments(processed_comments)
vectorstore = FAISS.from_documents(chunked_docs, embedding_model)
local_cache.setdefault(video_id, {})["comment_vectorstore"] = vectorstore
qa_prompt = get_qa_prompt(summary)
retriever = vectorstore.as_retriever()
qa_chain = RetrievalQA.from_chain_type(
llm=llm_with_grounding, retriever=retriever, chain_type="stuff",
chain_type_kwargs={"prompt": qa_prompt},
)
answer = qa_chain.invoke(question)
if not answer:
raise ValueError("LLM could not provide a meaningful answer to the question.")
return answer['result']
except (ValueError, RuntimeError) as e:
# Re-raise specific exceptions from sub-functions (summarize_comments, ensure_processed_comments)
raise e
except Exception as e:
print(f"Error answering comment question: {e}")
# Transform general exceptions into a RuntimeError for the API layer
raise RuntimeError(f"Error processing question: {str(e)}") |