Tether / app.py
SamanthaStorm's picture
Update app.py
e46fbeb verified
raw
history blame
17.6 kB
import gradio as gr
import torch
import numpy as np
from transformers import pipeline, RobertaForSequenceClassification, RobertaTokenizer
from motif_tagging import detect_motifs
import re
import matplotlib.pyplot as plt
import io
from PIL import Image
from datetime import datetime
# --- Timeline Visualization Function ---
def generate_abuse_score_chart(dates, scores, labels):
import matplotlib.pyplot as plt
import io
from PIL import Image
from datetime import datetime
try:
parsed_dates = [datetime.strptime(d, "%Y-%m-%d") for d in dates]
except Exception:
parsed_dates = list(range(len(dates)))
fig, ax = plt.subplots(figsize=(8, 3))
ax.plot(parsed_dates, scores, marker='o', linestyle='-', color='darkred', linewidth=2)
for i, (x, y) in enumerate(zip(parsed_dates, scores)):
label = labels[i]
ax.text(x, y + 2, f"{label}\n{int(y)}%", ha='center', fontsize=8, color='black')
ax.set_title("Abuse Intensity Over Time")
ax.set_xlabel("Date")
ax.set_ylabel("Abuse Score (%)")
ax.set_ylim(0, 105)
ax.grid(True)
plt.tight_layout()
buf = io.BytesIO()
plt.savefig(buf, format='png')
buf.seek(0)
return Image.open(buf)
# --- SST Sentiment Model ---
sst_pipeline = pipeline("sentiment-analysis", model="distilbert-base-uncased-finetuned-sst-2-english")
# --- Abuse Model ---
model_name = "SamanthaStorm/autotrain-jlpi4-mllvp"
model = RobertaForSequenceClassification.from_pretrained(model_name)
tokenizer = RobertaTokenizer.from_pretrained(model_name)
LABELS = [
"blame shifting", "contradictory statements", "control", "dismissiveness",
"gaslighting", "guilt tripping", "insults", "obscure language",
"projection", "recovery phase", "threat"
]
THRESHOLDS = {
"blame shifting": 0.3, "contradictory statements": 0.36, "control": 0.48, "dismissiveness": 0.45,
"gaslighting": 0.30, "guilt tripping": 0.20, "insults": 0.34, "obscure language": 0.25,
"projection": 0.35, "recovery phase": 0.25, "threat": 0.25
}
PATTERN_WEIGHTS = {
"gaslighting": 1.3,
"control": 1.2,
"dismissiveness": 0.8,
"blame shifting": 0.8,
"contradictory statements": 0.75,
"threat": 1.5 # 🔧 New: raise weight for threat
}
RISK_STAGE_LABELS = {
1: "🌀 Risk Stage: Tension-Building\nThis message reflects rising emotional pressure or subtle control attempts.",
2: "🔥 Risk Stage: Escalation\nThis message includes direct or aggressive patterns, suggesting active harm.",
3: "🌧️ Risk Stage: Reconciliation\nThis message reflects a reset attempt—apologies or emotional repair without accountability.",
4: "🌸 Risk Stage: Calm / Honeymoon\nThis message appears supportive but may follow prior harm, minimizing it."
}
ESCALATION_QUESTIONS = [
("Partner has access to firearms or weapons", 4),
("Partner threatened to kill you", 3),
("Partner threatened you with a weapon", 3),
("Partner has ever choked you, even if you considered it consensual at the time", 4),
("Partner injured or threatened your pet(s)", 3),
("Partner has broken your things, punched or kicked walls, or thrown things ", 2),
("Partner forced or coerced you into unwanted sexual acts", 3),
("Partner threatened to take away your children", 2),
("Violence has increased in frequency or severity", 3),
("Partner monitors your calls/GPS/social media", 2)
]
DARVO_PATTERNS = {
"blame shifting", "projection", "dismissiveness", "guilt tripping", "contradictory statements"
}
DARVO_MOTIFS = [
"I never said that.", "You’re imagining things.", "That never happened.",
"You’re making a big deal out of nothing.", "It was just a joke.", "You’re too sensitive.",
"I don’t know what you’re talking about.", "You’re overreacting.", "I didn’t mean it that way.",
"You’re twisting my words.", "You’re remembering it wrong.", "You’re always looking for something to complain about.",
"You’re just trying to start a fight.", "I was only trying to help.", "You’re making things up.",
"You’re blowing this out of proportion.", "You’re being paranoid.", "You’re too emotional.",
"You’re always so dramatic.", "You’re just trying to make me look bad.",
"You’re crazy.", "You’re the one with the problem.", "You’re always so negative.",
"You’re just trying to control me.", "You’re the abusive one.", "You’re trying to ruin my life.",
"You’re just jealous.", "You’re the one who needs help.", "You’re always playing the victim.",
"You’re the one causing all the problems.", "You’re just trying to make me feel guilty.",
"You’re the one who can’t let go of the past.", "You’re the one who’s always angry.",
"You’re the one who’s always complaining.", "You’re the one who’s always starting arguments.",
"You’re the one who’s always making things worse.", "You’re the one who’s always making me feel bad.",
"You’re the one who’s always making me look like the bad guy.",
"You’re the one who’s always making me feel like a failure.",
"You’re the one who’s always making me feel like I’m not good enough.",
"I can’t believe you’re doing this to me.", "You’re hurting me.",
"You’re making me feel like a terrible person.", "You’re always blaming me for everything.",
"You’re the one who’s abusive.", "You’re the one who’s controlling.", "You’re the one who’s manipulative.",
"You’re the one who’s toxic.", "You’re the one who’s gaslighting me.",
"You’re the one who’s always putting me down.", "You’re the one who’s always making me feel bad.",
"You’re the one who’s always making me feel like I’m not good enough.",
"You’re the one who’s always making me feel like I’m the problem.",
"You’re the one who’s always making me feel like I’m the bad guy.",
"You’re the one who’s always making me feel like I’m the villain.",
"You’re the one who’s always making me feel like I’m the one who needs to change.",
"You’re the one who’s always making me feel like I’m the one who’s wrong.",
"You’re the one who’s always making me feel like I’m the one who’s crazy.",
"You’re the one who’s always making me feel like I’m the one who’s abusive.",
"You’re the one who’s always making me feel like I’m the one who’s toxic."
]
def detect_contradiction(message):
patterns = [
(r"\b(i love you).{0,15}(i hate you|you ruin everything)", re.IGNORECASE),
(r"\b(i’m sorry).{0,15}(but you|if you hadn’t)", re.IGNORECASE),
(r"\b(i’m trying).{0,15}(you never|why do you)", re.IGNORECASE),
(r"\b(do what you want).{0,15}(you’ll regret it|i always give everything)", re.IGNORECASE),
(r"\b(i don’t care).{0,15}(you never think of me)", re.IGNORECASE),
(r"\b(i guess i’m just).{0,15}(the bad guy|worthless|never enough)", re.IGNORECASE)
]
return any(re.search(p, message, flags) for p, flags in patterns)
def calculate_darvo_score(patterns, sentiment_before, sentiment_after, motifs_found, contradiction_flag=False):
pattern_hits = len([p for p in patterns if p in DARVO_PATTERNS])
pattern_score = pattern_hits / len(DARVO_PATTERNS)
sentiment_shift_score = max(0.0, sentiment_after - sentiment_before)
motif_hits = len([
motif for motif in motifs_found
if any(phrase.lower() in motif.lower() for phrase in DARVO_MOTIFS)
])
motif_score = motif_hits / len(DARVO_MOTIFS)
contradiction_score = 1.0 if contradiction_flag else 0.0
return round(min(
0.3 * pattern_score +
0.3 * sentiment_shift_score +
0.25 * motif_score +
0.15 * contradiction_score, 1.0
), 3)
def detect_weapon_language(text):
weapon_keywords = [
"knife", "knives", "stab", "cut you", "cutting",
"gun", "shoot", "rifle", "firearm", "pistol",
"bomb", "blow up", "grenade", "explode",
"weapon", "armed", "loaded", "kill you", "take you out"
]
text_lower = text.lower()
return any(word in text_lower for word in weapon_keywords)
def get_risk_stage(patterns, sentiment):
if "threat" in patterns or "insults" in patterns:
return 2
elif "recovery phase" in patterns:
return 3
elif "control" in patterns or "guilt tripping" in patterns:
return 1
elif sentiment == "supportive" and any(p in patterns for p in ["projection", "dismissiveness"]):
return 4
return 1
def generate_risk_snippet(abuse_score, top_label, escalation_score, stage):
if abuse_score >= 85 or escalation_score >= 16:
risk_level = "high"
elif abuse_score >= 60 or escalation_score >= 8:
risk_level = "moderate"
elif stage == 2 and abuse_score >= 40:
risk_level = "moderate" # 🔧 New rule for escalation stage
else:
risk_level = "low"
pattern_label = top_label.split(" – ")[0]
pattern_score = top_label.split(" – ")[1] if " – " in top_label else ""
WHY_FLAGGED = {
"control": "This message may reflect efforts to restrict someone’s autonomy, even if it's framed as concern or care.",
"gaslighting": "This message could be manipulating someone into questioning their perception or feelings.",
"dismissiveness": "This message may include belittling, invalidating, or ignoring the other person’s experience.",
"insults": "Direct insults often appear in escalating abusive dynamics and can erode emotional safety.",
"threat": "This message includes threatening language, which is a strong predictor of harm.",
"blame shifting": "This message may redirect responsibility to avoid accountability, especially during conflict.",
"guilt tripping": "This message may induce guilt in order to control or manipulate behavior.",
"recovery phase": "This message may be part of a tension-reset cycle, appearing kind but avoiding change.",
"projection": "This message may involve attributing the abuser’s own behaviors to the victim.",
"default": "This message contains language patterns that may affect safety, clarity, or emotional autonomy."
}
explanation = WHY_FLAGGED.get(pattern_label.lower(), WHY_FLAGGED["default"])
base = f"\n\n🛑 Risk Level: {risk_level.capitalize()}\n"
base += f"This message shows strong indicators of **{pattern_label}**. "
if risk_level == "high":
base += "The language may reflect patterns of emotional control, even when expressed in soft or caring terms.\n"
elif risk_level == "moderate":
base += "There are signs of emotional pressure or indirect control that may escalate if repeated.\n"
else:
base += "The message does not strongly indicate abuse, but it's important to monitor for patterns.\n"
base += f"\n💡 *Why this might be flagged:*\n{explanation}\n"
base += f"\nDetected Pattern: **{pattern_label} ({pattern_score})**\n"
base += "🧠 You can review the pattern in context. This tool highlights possible dynamics—not judgments."
return base
def analyze_single_message(text, thresholds):
motif_hits, matched_phrases = detect_motifs(text)
result = sst_pipeline(text)[0]
sentiment = "supportive" if result['label'] == "POSITIVE" else "undermining"
sentiment_score = result['score'] if sentiment == "undermining" else 0.0
weapon_flag = detect_weapon_language(text)
adjusted_thresholds = {
k: v + 0.05 if sentiment == "supportive" else v
for k, v in thresholds.items()
}
contradiction_flag = detect_contradiction(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()
threshold_labels = [
label for label, score in zip(LABELS, scores)
if score > adjusted_thresholds[label]
]
motifs = [phrase for _, phrase in matched_phrases]
darvo_score = calculate_darvo_score(
threshold_labels,
sentiment_before=0.0,
sentiment_after=sentiment_score,
motifs_found=motifs,
contradiction_flag=contradiction_flag
)
top_patterns = sorted(
[(label, score) for label, score in zip(LABELS, scores)],
key=lambda x: x[1],
reverse=True
)[:2]
# Compute weighted average across all patterns (not just top 2)
weighted_total = 0.0
weight_sum = 0.0
for label, score in zip(LABELS, scores):
weight = PATTERN_WEIGHTS.get(label, 1.0)
weighted_total += score * weight
weight_sum += weight
abuse_score_raw = (weighted_total / weight_sum) * 100
stage = get_risk_stage(threshold_labels, sentiment)
if weapon_flag:
abuse_score_raw = min(abuse_score_raw + 25, 100) # boost intensity
if weapon_flag and stage < 2:
stage = 2
if weapon_flag:
print("⚠️ Weapon-related language detected.")
if "threat" in threshold_labels or "control" in threshold_labels or "insults" in threshold_labels:
abuse_score = min(abuse_score_raw, 100)
else:
abuse_score = min(abuse_score_raw, 95)
print("\n--- Debug Info ---")
print(f"Text: {text}")
print(f"Sentiment: {sentiment} (raw: {result['label']}, score: {result['score']:.3f})")
print("Abuse Pattern Scores:")
for label, score in zip(LABELS, scores):
passed = "✅" if score > adjusted_thresholds[label] else "❌"
print(f" {label:25}{score:.3f} {passed}")
print(f"Motifs: {motifs}")
print(f"Contradiction: {contradiction_flag}")
print("------------------\n")
return abuse_score, threshold_labels, top_patterns, result, stage, darvo_score
def analyze_composite(msg1, date1, msg2, date2, msg3, date3, *answers_and_none):
none_selected_checked = answers_and_none[-1]
responses_checked = any(answers_and_none[:-1])
none_selected = not responses_checked and none_selected_checked
if none_selected:
escalation_score = None
risk_level = "unknown"
else:
escalation_score = sum(w for (_, w), a in zip(ESCALATION_QUESTIONS, answers_and_none[:-1]) if a)
risk_level = (
"High" if escalation_score >= 16 else
"Moderate" if escalation_score >= 8 else
"Low"
)
messages = [msg1, msg2, msg3]
dates = [date1, date2, date3]
active = [(m, d) for m, d in zip(messages, dates) if m.strip()]
if not active:
return "Please enter at least one message."
results = [(analyze_single_message(m, THRESHOLDS.copy()), d) for m, d in active]
abuse_scores = [r[0][0] for r in results]
top_labels = [r[0][2][0][0] for r in results]
top_scores = [r[0][2][0][1] for r in results]
sentiments = [r[0][3]['label'] for r in results]
stages = [r[0][4] for r in results]
darvo_scores = [r[0][5] for r in results]
dates_used = [r[1] or "Undated" for r in results] # Store dates for future mapping
composite_abuse = int(round(sum(abuse_scores) / len(abuse_scores)))
top_label = f"{top_labels[0]}{int(round(top_scores[0] * 100))}%"
most_common_stage = max(set(stages), key=stages.count)
stage_text = RISK_STAGE_LABELS[most_common_stage]
avg_darvo = round(sum(darvo_scores) / len(darvo_scores), 3)
darvo_blurb = ""
if avg_darvo > 0.25:
level = "moderate" if avg_darvo < 0.65 else "high"
darvo_blurb = f"\n\n🎭 **DARVO Score: {avg_darvo}** → This indicates a **{level} likelihood** of narrative reversal (DARVO), where the speaker may be denying, attacking, or reversing blame."
out = f"Abuse Intensity: {composite_abuse}%\n"
out += "📊 This reflects the strength and severity of detected abuse patterns in the message(s).\n\n"
if escalation_score is None:
out += "Escalation Potential: Unknown (Checklist not completed)\n"
out += "🔍 *This section was not completed. Escalation potential is unknown.*\n\n"
else:
out += f"Escalation Potential: {risk_level} ({escalation_score}/{sum(w for _, w in ESCALATION_QUESTIONS)})\n"
out += "🚨 This indicates how many serious risk factors are present based on your answers to the safety checklist.\n"
out += generate_risk_snippet(composite_abuse, top_label, escalation_score if escalation_score is not None else 0, most_common_stage)
out += f"\n\n{stage_text}"
out += darvo_blurb
pattern_labels = [r[0][2][0][0] for r in results] # top label for each message
timeline_image = generate_abuse_score_chart(dates_used, abuse_scores, pattern_labels)
return out, timeline_image
message_date_pairs = [
(
gr.Textbox(label=f"Message {i+1}"),
gr.Textbox(label=f"Date {i+1} (optional)", placeholder="YYYY-MM-DD")
)
for i in range(3)
]
textbox_inputs = [item for pair in message_date_pairs for item in pair]
quiz_boxes = [gr.Checkbox(label=q) for q, _ in ESCALATION_QUESTIONS]
none_box = gr.Checkbox(label="None of the above")
iface = gr.Interface(
fn=analyze_composite,
inputs=textbox_inputs + quiz_boxes + [none_box],
outputs=[
gr.Textbox(label="Results"),
gr.Image(label="Risk Stage Timeline", type="pil")
],
title="Abuse Pattern Detector + Escalation Quiz",
allow_flagging="manual"
)
if __name__ == "__main__":
iface.launch().