Spaces:
Sleeping
Sleeping
Upload 3 files
Browse files- app.py.py +118 -0
- moderator.py +37 -0
- requirements.txt +8 -0
app.py.py
ADDED
|
@@ -0,0 +1,118 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from fastapi import FastAPI, Depends, HTTPException
|
| 2 |
+
from fastapi.middleware.cors import CORSMiddleware
|
| 3 |
+
from sqlalchemy.orm import Session
|
| 4 |
+
from typing import List
|
| 5 |
+
import database, schemas, moderator
|
| 6 |
+
|
| 7 |
+
# Init DB
|
| 8 |
+
database.init_db()
|
| 9 |
+
|
| 10 |
+
app = FastAPI(title="SafeStream API")
|
| 11 |
+
|
| 12 |
+
# Allow all origins for dev
|
| 13 |
+
app.add_middleware(
|
| 14 |
+
CORSMiddleware,
|
| 15 |
+
allow_origins=["*"],
|
| 16 |
+
allow_credentials=True,
|
| 17 |
+
allow_methods=["*"],
|
| 18 |
+
allow_headers=["*"],
|
| 19 |
+
)
|
| 20 |
+
|
| 21 |
+
def get_db():
|
| 22 |
+
db = database.SessionLocal()
|
| 23 |
+
try:
|
| 24 |
+
yield db
|
| 25 |
+
finally:
|
| 26 |
+
db.close()
|
| 27 |
+
|
| 28 |
+
@app.on_event("startup")
|
| 29 |
+
def startup_event():
|
| 30 |
+
# Seed data if empty
|
| 31 |
+
db = database.SessionLocal()
|
| 32 |
+
if db.query(database.Video).count() == 0:
|
| 33 |
+
# Using real YouTube IDs for the iframe
|
| 34 |
+
seed_videos = [
|
| 35 |
+
database.Video(title="Daily Vlog #102", url="dQw4w9WgXcQ", description="Just another day in the life. Comment below!"),
|
| 36 |
+
database.Video(title="Gaming Highlights", url="jNQXAC9IVRw", description="Insane plays from last night."),
|
| 37 |
+
database.Video(title="Relaxing Rain Sounds", url="q76bMs-NwRk", description="Sleep aid.")
|
| 38 |
+
]
|
| 39 |
+
db.add_all(seed_videos)
|
| 40 |
+
db.commit()
|
| 41 |
+
db.close()
|
| 42 |
+
|
| 43 |
+
@app.get("/videos", response_model=List[schemas.Video])
|
| 44 |
+
def get_videos(db: Session = Depends(get_db)):
|
| 45 |
+
return db.query(database.Video).all()
|
| 46 |
+
|
| 47 |
+
@app.get("/videos/{video_id}", response_model=schemas.Video)
|
| 48 |
+
def get_video(video_id: int, db: Session = Depends(get_db)):
|
| 49 |
+
video = db.query(database.Video).filter(database.Video.id == video_id).first()
|
| 50 |
+
if not video:
|
| 51 |
+
raise HTTPException(status_code=404, detail="Video not found")
|
| 52 |
+
return video
|
| 53 |
+
|
| 54 |
+
|
| 55 |
+
@app.get("/videos/{video_id}/comments", response_model=List[schemas.Comment])
|
| 56 |
+
def get_comments(video_id: int, db: Session = Depends(get_db)):
|
| 57 |
+
return db.query(database.Comment).filter(database.Comment.video_id == video_id).all()
|
| 58 |
+
|
| 59 |
+
@app.post("/comments", response_model=schemas.Comment)
|
| 60 |
+
def create_comment(comment: schemas.CommentCreate, db: Session = Depends(get_db)):
|
| 61 |
+
# 1. Analyze for toxicity
|
| 62 |
+
analysis = moderator.moderator.analyze(comment.text)
|
| 63 |
+
|
| 64 |
+
# 2. Save to DB
|
| 65 |
+
db_comment = database.Comment(
|
| 66 |
+
video_id=comment.video_id,
|
| 67 |
+
user=comment.user,
|
| 68 |
+
text=comment.text,
|
| 69 |
+
timestamp=comment.timestamp,
|
| 70 |
+
is_toxic=analysis["is_toxic"],
|
| 71 |
+
toxicity_score=analysis["score"],
|
| 72 |
+
flagged_reason=analysis["reason"]
|
| 73 |
+
)
|
| 74 |
+
db.add(db_comment)
|
| 75 |
+
db.commit()
|
| 76 |
+
db.refresh(db_comment)
|
| 77 |
+
return db_comment
|
| 78 |
+
|
| 79 |
+
@app.delete("/comments")
|
| 80 |
+
def clear_comments(db: Session = Depends(get_db)):
|
| 81 |
+
db.query(database.Comment).delete()
|
| 82 |
+
db.commit()
|
| 83 |
+
return {"message": "All comments cleared"}
|
| 84 |
+
|
| 85 |
+
@app.get("/stats")
|
| 86 |
+
def get_stats(db: Session = Depends(get_db)):
|
| 87 |
+
comments = db.query(database.Comment).all()
|
| 88 |
+
total = len(comments)
|
| 89 |
+
toxic_comments = [c for c in comments if c.is_toxic]
|
| 90 |
+
toxic_count = len(toxic_comments)
|
| 91 |
+
|
| 92 |
+
# Custom Classification Logic for Demo
|
| 93 |
+
types = {
|
| 94 |
+
"Insult": 0,
|
| 95 |
+
"Identity Hate": 0,
|
| 96 |
+
"Threat": 0,
|
| 97 |
+
"Online Harassment": 0
|
| 98 |
+
}
|
| 99 |
+
|
| 100 |
+
for c in toxic_comments:
|
| 101 |
+
text = c.text.lower()
|
| 102 |
+
if any(w in text for w in ["kill", "die", "hurt", "attack", "gun", "shoot", "murder"]):
|
| 103 |
+
types["Threat"] += 1
|
| 104 |
+
elif any(w in text for w in ["hate", "racist", "gay", "lesbian", "black", "white", "jew", "muslim", "women", "men", "trans"]):
|
| 105 |
+
types["Identity Hate"] += 1
|
| 106 |
+
elif any(w in text for w in ["stupid", "idiot", "dumb", "ugly", "fat", "loser", "moron", "coward"]):
|
| 107 |
+
types["Insult"] += 1
|
| 108 |
+
else:
|
| 109 |
+
# Catch-all for other toxic comments
|
| 110 |
+
types["Online Harassment"] += 1
|
| 111 |
+
|
| 112 |
+
return {
|
| 113 |
+
"total_comments": total,
|
| 114 |
+
"toxic_comments": toxic_count,
|
| 115 |
+
"toxic_ratio": (toxic_count/total) if total > 0 else 0,
|
| 116 |
+
"types_breakdown": [{"name": k, "value": v} for k, v in types.items()]
|
| 117 |
+
}
|
| 118 |
+
|
moderator.py
ADDED
|
@@ -0,0 +1,37 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
from transformers import pipeline
|
| 2 |
+
import torch
|
| 3 |
+
|
| 4 |
+
class ToxicityModel:
|
| 5 |
+
def __init__(self):
|
| 6 |
+
print("Loading Toxicity Model...")
|
| 7 |
+
# Using a model that is good for toxic comment detection
|
| 8 |
+
# 'distilbert-base-uncased-finetuned-sst-2-english' is Sentiment (Pos/Neg)
|
| 9 |
+
# Neg is a decent proxy for toxic for a demo.
|
| 10 |
+
# Alternatively: 'unitary/toxic-bert' is better but heavier.
|
| 11 |
+
try:
|
| 12 |
+
self.classifier = pipeline("sentiment-analysis", model="distilbert-base-uncased-finetuned-sst-2-english")
|
| 13 |
+
except Exception as e:
|
| 14 |
+
print(f"Error loading model: {e}")
|
| 15 |
+
self.classifier = None
|
| 16 |
+
|
| 17 |
+
def analyze(self, text: str):
|
| 18 |
+
if not self.classifier:
|
| 19 |
+
return {"is_toxic": False, "score": 0.0, "reason": "Model inactive"}
|
| 20 |
+
|
| 21 |
+
results = self.classifier(text)
|
| 22 |
+
# Format: [{'label': 'NEGATIVE', 'score': 0.99}]
|
| 23 |
+
result = results[0]
|
| 24 |
+
|
| 25 |
+
label = result['label']
|
| 26 |
+
score = result['score']
|
| 27 |
+
|
| 28 |
+
# Heuristic: If Negative and high confidence, flag it.
|
| 29 |
+
is_toxic = (label == 'NEGATIVE' and score > 0.6)
|
| 30 |
+
|
| 31 |
+
return {
|
| 32 |
+
"is_toxic": is_toxic,
|
| 33 |
+
"score": score,
|
| 34 |
+
"reason": f"Classified as {label} with confidence {score:.2f}"
|
| 35 |
+
}
|
| 36 |
+
|
| 37 |
+
moderator = ToxicityModel()
|
requirements.txt
ADDED
|
@@ -0,0 +1,8 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
fastapi
|
| 2 |
+
uvicorn
|
| 3 |
+
pydantic
|
| 4 |
+
transformers
|
| 5 |
+
torch
|
| 6 |
+
scikit-learn
|
| 7 |
+
python-multipart
|
| 8 |
+
sqlalchemy
|