app.py

import gradio as gr
import torch
from transformers import pipeline as hf_pipeline, AutoModelForSequenceClassification, AutoTokenizer


# ——— 1) Emotion Pipeline ————————————————————————————————————————————————
emotion_pipeline = hf_pipeline(
    "text-classification",
    model="j-hartmann/emotion-english-distilroberta-base",
    top_k=None,
    truncation=True
)

def get_emotion_profile(text):
    """
    Returns a dict of { emotion_label: score } for the input text.
    """
    results = emotion_pipeline(text)
    # some pipelines return [[…]]
    if isinstance(results, list) and isinstance(results[0], list):
        results = results[0]
    return {r["label"].lower(): round(r["score"], 3) for r in results}


# ——— 2) Abuse-patterns Model ——————————————————————————————————————————————
model_name = "SamanthaStorm/tether-multilabel-v3"
model     = AutoModelForSequenceClassification.from_pretrained(model_name)
tokenizer = AutoTokenizer.from_pretrained(model_name, use_fast=False)

LABELS = [
    "blame shifting", "contradictory statements", "control", "dismissiveness",
    "gaslighting", "guilt tripping", "insults", "obscure language",
    "projection", "recovery phase", "threat"
]

THRESHOLDS = {
    "blame shifting": 0.28, "contradictory statements": 0.27, "control": 0.08, "dismissiveness": 0.32,
    "gaslighting":     0.27, "guilt tripping":           0.31, "insults":  0.10, "obscure language": 0.55,
    "projection":      0.09, "recovery phase":           0.33, "threat":    0.15
}

# ——— 3) Emotional-tone tagging (no abuse_score / DARVO) —————————————————————————————
def get_tone_tag(emotion_profile, patterns):
    anger   = emotion_profile.get("anger",   0)
    disgust = emotion_profile.get("disgust", 0)
    sadness = emotion_profile.get("sadness", 0)
    joy     = emotion_profile.get("joy",     0)
    neutral = emotion_profile.get("neutral", 0)
    fear    = emotion_profile.get("fear",    0)

    # 1) Vulnerable: sadness high + recovery-phase
    if sadness > 0.4 and "recovery phase" in patterns:
        return "vulnerable"

    # 2) Supportive: joy very high + no other patterns (or only recovery-phase)
    if joy > 0.5 and (not patterns or patterns == ["recovery phase"]):
        return "supportive"

    # 3) Confrontational: anger/disgust high + aggressive patterns
    if (anger + disgust) > 0.5 and any(p in patterns for p in ["insults", "control", "threat"]):
        return "confrontational"

    # 4) Manipulative: neutral high + classic manipulation patterns
    if neutral > 0.4 and any(p in patterns for p in ["gaslighting", "dismissiveness", "projection", "guilt tripping", "blame shifting"]):
        return "manipulative"

    # 5) Feigned Warmth: joy high but manipulative patterns present
    if joy > 0.5 and any(p in patterns for p in ["gaslighting", "dismissiveness", "projection", "guilt tripping", "blame shifting"]):
        return "feigned warmth"

    # 6) Defensive: anger high + contradictory statements
    if anger > 0.4 and "contradictory statements" in patterns:
        return "defensive"

    # 7) Neutral: pure neutral dominates all
    if neutral > max(anger, disgust, sadness, joy, fear):
        return "neutral"

    return None


# ——— 3) Single-message analysis ——————————————————————————————————————————————
def analyze_message(text):
    """
    Runs emotion profiling, and the abuse-pattern classifier.
    Returns a dict with:
      - emotion_profile: { emotion: score }
      - active_patterns: [ labels above their threshold ]
    """
    
    emotion_profile = get_emotion_profile(text)

    # get raw model scores
    toks = tokenizer(text, return_tensors="pt", truncation=True, padding=True)
    with torch.no_grad():
        logits = model(**toks).logits.squeeze(0)
    scores = torch.sigmoid(logits).cpu().numpy()

    # pick up all labels whose score >= threshold
    active = [lab for lab, sc in zip(LABELS, scores) if sc >= THRESHOLDS[lab]]

    return {
        "emotion_profile": emotion_profile,
        "active_patterns": active,
        "tone_tag": tone_tag
    }


# ——— 5) Composite wrapper (handles .txt or image + text boxes) ——————————————————————
def analyze_composite(uploaded_file, *texts):
    outputs = []

    if uploaded_file is not None:
        raw = uploaded_file.read()
        name = uploaded_file.name.lower()
        if name.endswith((".png", ".jpg", ".jpeg", ".tiff", ".bmp", ".gif")):
            img = Image.open(io.BytesIO(raw))
            content = pytesseract.image_to_string(img)
        else:
            try:
                content = raw.decode("utf-8")
            except UnicodeDecodeError:
                content = raw.decode("latin-1")

        r = analyze_message(content)
        outputs.append(
            "── Uploaded File ──\n"
            f"Emotion Profile : {r['emotion_profile']}\n"
            f"Active Patterns : {r['active_patterns']}\n"
            f"Emotional Tone : {r['tone_tag']}\n"
        )

    for idx, txt in enumerate(texts, start=1):
        if not txt:
            continue
        r = analyze_message(txt)
        outputs.append(
            f"── Message {idx} ──\n"
            f"Emotion Profile : {r['emotion_profile']}\n"
            f"Active Patterns : {r['active_patterns']}\n"
            f"Emotional Tone : {r['tone_tag']}\n"
        )

    return "\n".join(outputs) if outputs else "Please enter at least one message."


# ——— 6) Gradio interface ————————————————————————————————————————————————
message_inputs = [gr.Textbox(label=f"Message {i+1}") for i in range(3)]

iface = gr.Interface(
    fn=analyze_composite,
    inputs=[ 
        gr.File(file_types=[".txt", ".png", ".jpg", ".jpeg"], 
                label="Upload text or image") 
    ] + message_inputs,
    outputs=gr.Textbox(label="Analysis"),
    title="Tether Analyzer (with Tone Tags)",
    description="Extracts motifs, emotions, patterns—and now an emotional tone tag—no abuse score or DARVO."
)

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