app.py

import gradio as gr
import torch
import numpy as np
from transformers import AutoModelForSequenceClassification, AutoTokenizer, RobertaForSequenceClassification, RobertaTokenizer
from motif_tagging import detect_motifs

# Load models
sentiment_model = AutoModelForSequenceClassification.from_pretrained("SamanthaStorm/tether-sentiment")
sentiment_tokenizer = AutoTokenizer.from_pretrained("SamanthaStorm/tether-sentiment")
model_name = "SamanthaStorm/autotrain-c1un8-p8vzo"
model = RobertaForSequenceClassification.from_pretrained(model_name, trust_remote_code=True)
tokenizer = RobertaTokenizer.from_pretrained(model_name, trust_remote_code=True)

LABELS = [...]
THRESHOLDS = {...}
PATTERN_LABELS = LABELS[:15]
DANGER_LABELS = LABELS[15:18]
EXPLANATIONS = {...}
PATTERN_WEIGHTS = {...}


def custom_sentiment(text):
    inputs = sentiment_tokenizer(text, return_tensors="pt", truncation=True, padding=True)
    with torch.no_grad():
        outputs = sentiment_model(**inputs)
        probs = torch.nn.functional.softmax(outputs.logits, dim=1)
        label_idx = torch.argmax(probs).item()
    label_map = {0: "supportive", 1: "undermining"}
    label = label_map[label_idx]
    score = probs[0][label_idx].item()
    return {"label": label, "score": score}


def calculate_abuse_level(scores, thresholds, motif_hits=None):
    weighted_scores = []
    for label, score in zip(LABELS, scores):
        if score > thresholds[label]:
            weight = PATTERN_WEIGHTS.get(label, 1.0)
            weighted_scores.append(score * weight)
    base_score = round(np.mean(weighted_scores) * 100, 2) if weighted_scores else 0.0
    motif_hits = motif_hits or []
    if any(label in motif_hits for label in DANGER_LABELS):
        base_score = max(base_score, 75.0)
    return base_score


def interpret_abuse_level(score):
    if score > 80:
        return "Extreme / High Risk"
    elif score > 60:
        return "Severe / Harmful Pattern Present"
    elif score > 40:
        return "Likely Abuse"
    elif score > 20:
        return "Mild Concern"
    return "Very Low / Likely Safe"


def analyze_single_message(text):
    if not text.strip():
        return "No input provided."
    sentiment = custom_sentiment(text)
    thresholds = {k: v * 0.8 for k, v in THRESHOLDS.items()} if sentiment['label'] == "undermining" else THRESHOLDS.copy()
    motif_flags, matched_phrases = detect_motifs(text)
    inputs = tokenizer(text, return_tensors="pt", truncation=True, padding=True)
    with torch.no_grad():
        outputs = model(**inputs)
    scores = torch.sigmoid(outputs.logits.squeeze(0)).numpy()
    abuse_score = calculate_abuse_level(scores, thresholds, [label for label, _ in matched_phrases])
    summary = interpret_abuse_level(abuse_score)
    return f"Abuse Risk Score: {abuse_score}% — {summary}\nSentiment: {sentiment['label']} ({sentiment['score']*100:.2f}%)"


def analyze_composite(msg1, msg2, msg3):
    results = [analyze_single_message(t) for t in [msg1, msg2, msg3]]
    composite_score = np.mean([
        float(line.split('%')[0].split()[-1]) if 'Abuse Risk Score:' in line else 0
        for line in results
    ])
    final_summary = interpret_abuse_level(composite_score)
    composite_result = f"\n\nComposite Abuse Risk Score: {composite_score:.2f}% — {final_summary}"
    return results[0], results[1], results[2], composite_result


iface = gr.Interface(
    fn=analyze_composite,
    inputs=[
        gr.Textbox(label="Message 1"),
        gr.Textbox(label="Message 2"),
        gr.Textbox(label="Message 3")
    ],
    outputs=[
        gr.Textbox(label="Message 1 Result"),
        gr.Textbox(label="Message 2 Result"),
        gr.Textbox(label="Message 3 Result"),
        gr.Textbox(label="Composite Score Summary")
    ],
    title="Abuse Pattern Detector (Multi-Message)",
    live=False,
    allow_flagging="manual"
)

if __name__ == "__main__":
    iface.launch()