Spaces:
Sleeping
Sleeping
Update app.py
Browse files
app.py
CHANGED
@@ -13,7 +13,7 @@ import numpy as np
|
|
13 |
import json
|
14 |
import tempfile
|
15 |
from fastapi import FastAPI, UploadFile, File, HTTPException, Form, BackgroundTasks
|
16 |
-
from fastapi.responses import FileResponse, JSONResponse, HTMLResponse
|
17 |
from fastapi.middleware.cors import CORSMiddleware
|
18 |
from transformers import pipeline, AutoModelForQuestionAnswering, AutoTokenizer
|
19 |
from sentence_transformers import SentenceTransformer
|
@@ -31,6 +31,9 @@ from starlette.concurrency import run_in_threadpool
|
|
31 |
import gensim
|
32 |
from gensim import corpora, models
|
33 |
|
|
|
|
|
|
|
34 |
# Global cache for analysis results based on file hash
|
35 |
analysis_cache = {}
|
36 |
|
@@ -197,15 +200,13 @@ try:
|
|
197 |
nlp = spacy.load("en_core_web_sm")
|
198 |
print("✅ Loading NLP models...")
|
199 |
|
200 |
-
# Update summarizer to use
|
201 |
-
from transformers import LEDTokenizer
|
202 |
summarizer = pipeline(
|
203 |
"summarization",
|
204 |
-
model="
|
205 |
-
tokenizer="
|
206 |
device=0 if torch.cuda.is_available() else -1
|
207 |
)
|
208 |
-
# Optionally convert summarizer model to FP16 for faster inference on GPU (if supported)
|
209 |
if device == "cuda":
|
210 |
try:
|
211 |
summarizer.model.half()
|
@@ -235,8 +236,6 @@ except Exception as e:
|
|
235 |
|
236 |
from transformers import pipeline
|
237 |
qa_model = pipeline("question-answering", model="deepset/roberta-base-squad2")
|
238 |
-
|
239 |
-
# Initialize sentiment-analysis pipeline
|
240 |
sentiment_pipeline = pipeline("sentiment-analysis", model="distilbert-base-uncased-finetuned-sst-2-english", device=0 if torch.cuda.is_available() else -1)
|
241 |
|
242 |
def legal_chatbot(user_input, context):
|
@@ -263,10 +262,8 @@ async def process_video_to_text(video_file_path):
|
|
263 |
"-acodec", "pcm_s16le", "-ar", "44100", "-ac", "2",
|
264 |
temp_audio_path, "-y"
|
265 |
]
|
266 |
-
# Run ffmpeg in a separate thread
|
267 |
await run_in_threadpool(subprocess.run, cmd, check=True)
|
268 |
print(f"Audio extracted to {temp_audio_path}")
|
269 |
-
# Run speech-to-text in threadpool
|
270 |
result = await run_in_threadpool(speech_to_text, temp_audio_path)
|
271 |
transcript = result["text"]
|
272 |
print(f"Transcription completed: {len(transcript)} characters")
|
@@ -326,11 +323,61 @@ def get_enhanced_context_info(text):
|
|
326 |
enhanced["topics"] = analyze_topics(text, num_topics=5)
|
327 |
return enhanced
|
328 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
329 |
def analyze_risk_enhanced(text):
|
330 |
enhanced = get_enhanced_context_info(text)
|
331 |
avg_sentiment = enhanced["average_sentiment"]
|
332 |
risk_score = abs(avg_sentiment) if avg_sentiment < 0 else 0
|
333 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
334 |
|
335 |
def analyze_contract_clauses(text):
|
336 |
max_length = 512
|
@@ -370,7 +417,6 @@ async def analyze_legal_document(file: UploadFile = File(...)):
|
|
370 |
try:
|
371 |
content = await file.read()
|
372 |
file_hash = compute_md5(content)
|
373 |
-
# Return cached result if available
|
374 |
if file_hash in analysis_cache:
|
375 |
return analysis_cache[file_hash]
|
376 |
text = await run_in_threadpool(extract_text_from_pdf, io.BytesIO(content))
|
@@ -594,10 +640,8 @@ async def download_clause_radar_chart(task_id: str):
|
|
594 |
clauses = analyze_contract_clauses(text)
|
595 |
if not clauses:
|
596 |
raise HTTPException(status_code=404, detail="No clauses detected.")
|
597 |
-
# For radar chart, use clause types and their confidence scores
|
598 |
labels = [c["type"] for c in clauses]
|
599 |
values = [c["confidence"] for c in clauses]
|
600 |
-
# To close the radar chart, repeat the first value and label
|
601 |
labels += labels[:1]
|
602 |
values += values[:1]
|
603 |
angles = np.linspace(0, 2 * np.pi, len(labels), endpoint=False).tolist()
|
|
|
13 |
import json
|
14 |
import tempfile
|
15 |
from fastapi import FastAPI, UploadFile, File, HTTPException, Form, BackgroundTasks
|
16 |
+
from fastapi.responses import FileResponse, JSONResponse, HTMLResponse
|
17 |
from fastapi.middleware.cors import CORSMiddleware
|
18 |
from transformers import pipeline, AutoModelForQuestionAnswering, AutoTokenizer
|
19 |
from sentence_transformers import SentenceTransformer
|
|
|
31 |
import gensim
|
32 |
from gensim import corpora, models
|
33 |
|
34 |
+
# Import spacy stop words
|
35 |
+
from spacy.lang.en.stop_words import STOP_WORDS
|
36 |
+
|
37 |
# Global cache for analysis results based on file hash
|
38 |
analysis_cache = {}
|
39 |
|
|
|
200 |
nlp = spacy.load("en_core_web_sm")
|
201 |
print("✅ Loading NLP models...")
|
202 |
|
203 |
+
# Update summarizer to use facebook/bart-large-cnn for summarization
|
|
|
204 |
summarizer = pipeline(
|
205 |
"summarization",
|
206 |
+
model="facebook/bart-large-cnn",
|
207 |
+
tokenizer="facebook/bart-large-cnn",
|
208 |
device=0 if torch.cuda.is_available() else -1
|
209 |
)
|
|
|
210 |
if device == "cuda":
|
211 |
try:
|
212 |
summarizer.model.half()
|
|
|
236 |
|
237 |
from transformers import pipeline
|
238 |
qa_model = pipeline("question-answering", model="deepset/roberta-base-squad2")
|
|
|
|
|
239 |
sentiment_pipeline = pipeline("sentiment-analysis", model="distilbert-base-uncased-finetuned-sst-2-english", device=0 if torch.cuda.is_available() else -1)
|
240 |
|
241 |
def legal_chatbot(user_input, context):
|
|
|
262 |
"-acodec", "pcm_s16le", "-ar", "44100", "-ac", "2",
|
263 |
temp_audio_path, "-y"
|
264 |
]
|
|
|
265 |
await run_in_threadpool(subprocess.run, cmd, check=True)
|
266 |
print(f"Audio extracted to {temp_audio_path}")
|
|
|
267 |
result = await run_in_threadpool(speech_to_text, temp_audio_path)
|
268 |
transcript = result["text"]
|
269 |
print(f"Transcription completed: {len(transcript)} characters")
|
|
|
323 |
enhanced["topics"] = analyze_topics(text, num_topics=5)
|
324 |
return enhanced
|
325 |
|
326 |
+
# New function to create a detailed, dynamic explanation for each topic
|
327 |
+
def explain_topics(topics):
|
328 |
+
explanation = {}
|
329 |
+
for topic_idx, topic_str in topics:
|
330 |
+
# Split topic string into individual weighted terms
|
331 |
+
parts = topic_str.split('+')
|
332 |
+
terms = []
|
333 |
+
for part in parts:
|
334 |
+
part = part.strip()
|
335 |
+
if '*' in part:
|
336 |
+
weight_str, word = part.split('*', 1)
|
337 |
+
word = word.strip().strip('\"').strip('\'')
|
338 |
+
try:
|
339 |
+
weight = float(weight_str)
|
340 |
+
except:
|
341 |
+
weight = 0.0
|
342 |
+
# Filter out common stop words
|
343 |
+
if word.lower() not in STOP_WORDS and len(word) > 1:
|
344 |
+
terms.append((weight, word))
|
345 |
+
terms.sort(key=lambda x: -x[0])
|
346 |
+
# Create a plain language label based on dominant words
|
347 |
+
if terms:
|
348 |
+
if any("liability" in word.lower() for weight, word in terms):
|
349 |
+
label = "Liability & Penalty Risk"
|
350 |
+
elif any("termination" in word.lower() for weight, word in terms):
|
351 |
+
label = "Termination & Refund Risk"
|
352 |
+
elif any("compliance" in word.lower() for weight, word in terms):
|
353 |
+
label = "Compliance & Regulatory Risk"
|
354 |
+
else:
|
355 |
+
label = "General Risk Language"
|
356 |
+
else:
|
357 |
+
label = "General Risk Language"
|
358 |
+
explanation_text = (
|
359 |
+
f"Topic {topic_idx} ({label}) is characterized by dominant terms: " +
|
360 |
+
", ".join([f"'{word}' ({weight:.3f})" for weight, word in terms[:5]])
|
361 |
+
)
|
362 |
+
explanation[topic_idx] = {
|
363 |
+
"label": label,
|
364 |
+
"explanation": explanation_text,
|
365 |
+
"terms": terms
|
366 |
+
}
|
367 |
+
return explanation
|
368 |
+
|
369 |
def analyze_risk_enhanced(text):
|
370 |
enhanced = get_enhanced_context_info(text)
|
371 |
avg_sentiment = enhanced["average_sentiment"]
|
372 |
risk_score = abs(avg_sentiment) if avg_sentiment < 0 else 0
|
373 |
+
topics_raw = enhanced["topics"]
|
374 |
+
topics_explanation = explain_topics(topics_raw)
|
375 |
+
return {
|
376 |
+
"risk_score": risk_score,
|
377 |
+
"average_sentiment": avg_sentiment,
|
378 |
+
"topics": topics_raw,
|
379 |
+
"topics_explanation": topics_explanation
|
380 |
+
}
|
381 |
|
382 |
def analyze_contract_clauses(text):
|
383 |
max_length = 512
|
|
|
417 |
try:
|
418 |
content = await file.read()
|
419 |
file_hash = compute_md5(content)
|
|
|
420 |
if file_hash in analysis_cache:
|
421 |
return analysis_cache[file_hash]
|
422 |
text = await run_in_threadpool(extract_text_from_pdf, io.BytesIO(content))
|
|
|
640 |
clauses = analyze_contract_clauses(text)
|
641 |
if not clauses:
|
642 |
raise HTTPException(status_code=404, detail="No clauses detected.")
|
|
|
643 |
labels = [c["type"] for c in clauses]
|
644 |
values = [c["confidence"] for c in clauses]
|
|
|
645 |
labels += labels[:1]
|
646 |
values += values[:1]
|
647 |
angles = np.linspace(0, 2 * np.pi, len(labels), endpoint=False).tolist()
|