File size: 3,566 Bytes
8c85b97 |
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 |
import json
from datetime import datetime
from typing import Dict, Any, List
from langchain_openai import ChatOpenAI
from langchain_core.messages import SystemMessage, HumanMessage
from src.config.settings import Config
class MetadataExtractor:
"""
Service to extract intelligent metadata from meeting transcripts using an LLM.
Extracts: Title, Summary, Date, and Speaker Identities.
"""
def __init__(self):
# Use a cost-effective model for metadata extraction if possible
# defaulting to the configured model
self.llm = ChatOpenAI(
model=Config.METADATA_MODEL,
temperature=0, # Deterministic output
openai_api_key=Config.OPENAI_API_KEY
)
def extract_metadata(self, transcript_text: str) -> Dict[str, Any]:
"""
Analyze transcript to extract title, summary, date, and speaker mapping.
"""
# Truncate transcript if too long to avoid token limits (e.g., first 15k chars)
# usually enough for context
analysis_text = transcript_text[:15000]
system_prompt = """You are a Metadata Extraction Expert. Analyze the provided meeting transcript and extract the following information in JSON format:
1. "title": A concise, meaningful title for the meeting (e.g., "Q3 Marketing Strategy Review").
2. "summary": A brief 2-3 sentence summary of the meeting.
3. "meeting_date": The date the meeting likely took place, if mentioned (format: YYYY-MM-DD). If not explicitly mentioned, return null.
4. "speaker_mapping": A dictionary mapping generic speaker labels (SPEAKER_00, SPEAKER_01) to likely real names based on introductions or context. If unknown, leave empty.
Example Output:
{
"title": "Project Alpha Kickoff",
"summary": "The team discussed the timeline for Project Alpha. John assigned tasks to Sarah and Mike.",
"meeting_date": "2023-10-12",
"speaker_mapping": {
"SPEAKER_00": "John Smith",
"SPEAKER_01": "Sarah Jones"
}
}
"""
try:
response = self.llm.invoke([
SystemMessage(content=system_prompt),
HumanMessage(content=f"Transcript:\n{analysis_text}")
])
# Parse JSON from response
content = response.content.strip()
if "```json" in content:
content = content.split("```json")[1].split("```")[0].strip()
elif "```" in content:
content = content.split("```")[1].split("```")[0].strip()
metadata = json.loads(content)
return metadata
except Exception as e:
print(f"Error extracting metadata: {e}")
# Return safe defaults
return {
"title": "Untitled Meeting",
"summary": "No summary available.",
"meeting_date": None,
"speaker_mapping": {}
}
def apply_speaker_mapping(self, transcript: str, mapping: Dict[str, str]) -> str:
"""
Replace generic speaker labels with identified names in the transcript.
"""
if not mapping:
return transcript
updated_transcript = transcript
for generic, real_name in mapping.items():
# Replace "SPEAKER_00" with "John Smith"
# We handle common variations in formatting
updated_transcript = updated_transcript.replace(generic, real_name)
return updated_transcript
|