File size: 11,850 Bytes
943be01
 
d3d5ee5
943be01
d3d5ee5
 
 
 
 
 
 
 
 
 
943be01
d3d5ee5
943be01
d3d5ee5
 
 
 
 
 
 
943be01
d3d5ee5
943be01
d3d5ee5
 
 
 
 
 
 
 
 
1033c6d
 
 
 
 
 
 
d3d5ee5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1033c6d
 
 
 
 
d3d5ee5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1033c6d
 
 
 
d3d5ee5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1033c6d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
d3d5ee5
 
 
 
 
 
 
 
 
 
 
1033c6d
 
d3d5ee5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1033c6d
d3d5ee5
1033c6d
d3d5ee5
 
943be01
 
 
d3d5ee5
 
 
 
1033c6d
d3d5ee5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1033c6d
 
 
 
 
 
 
 
 
d3d5ee5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1033c6d
d3d5ee5
943be01
 
d3d5ee5
 
 
943be01
 
d3d5ee5
 
1033c6d
d3d5ee5
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
from fastapi import FastAPI, File, UploadFile, Form
from fastapi.middleware.cors import CORSMiddleware
from transformers import LongformerTokenizer, pipeline
from PIL import Image
import pytesseract
import cv2
import re
import torch
import matplotlib.pyplot as plt
import math
import io
import base64
from typing import Dict, List, Any, Optional
import numpy as np

app = FastAPI()

app.add_middleware(
    CORSMiddleware,
    allow_origins=["*"],
    allow_credentials=True,
    allow_methods=["*"],
    allow_headers=["*"],
)

device = 0 if torch.cuda.is_available() else -1

model_id = "allenai/longformer-base-4096"
tok = LongformerTokenizer.from_pretrained(model_id)

emo_head = pipeline(
    "text-classification",
    model="j-hartmann/emotion-english-distilroberta-base",
    return_all_scores=True,
    device=device,
)

translator = pipeline(
    "translation",
    model="Helsinki-NLP/opus-mt-mul-en",
    device=device,
)

flan = pipeline("text2text-generation", model="google/flan-t5-base", device=device)

time_regex = re.compile(r"(\d{1,2}[:]\d{2}\s*(AM|PM|am|pm)?)|(\d{1,2}[/]\d{1,2}[/]\d{2,4})")
negative_keys = {"anger", "sadness", "fear", "disgust"}
positive_keys = {"joy", "surprise"}


def mask_names(names: List[str]) -> Dict[str, str]:
    return {n: f"User_{i+1}" for i, n in enumerate(names)}


def extract_time(line: str) -> str:
    m = time_regex.search(line)
    return m.group() if m else ""


def ocr_image(image: Image.Image) -> str:
    img = image.convert("RGB")
    try:
        return pytesseract.image_to_string(img, lang="eng+hin+tel")
    except Exception:
        return pytesseract.image_to_string(img)


def ocr_video_bytes(video_bytes: bytes) -> str:
    temp_path = "/tmp/temp_video.mp4"
    with open(temp_path, "wb") as f:
        f.write(video_bytes)
    cap = cv2.VideoCapture(temp_path)
    texts = []
    idx = 0
    while True:
        ret, frame = cap.read()
        if not ret:
            break
        if idx % 25 == 0:
            rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
            img = Image.fromarray(rgb)
            try:
                t = pytesseract.image_to_string(img, lang="eng+hin+tel")
            except Exception:
                t = pytesseract.image_to_string(img)
            if t.strip():
                texts.append(t)
        idx += 1
    cap.release()
    return "\n".join(texts)


def split_by_speaker(text: str, privacy: bool) -> Dict[str, str]:
    speakers: Dict[str, List[str]] = {}
    for raw in text.splitlines():
        if ":" in raw:
            name, msg = raw.split(":", 1)
            name, msg = name.strip(), msg.strip()
            if msg:
                speakers.setdefault(name, []).append(msg)
    if not speakers:
        speakers["User"] = [text]
    if privacy:
        mapping = mask_names(list(speakers.keys()))
        return {mapping[k]: " ".join(v) for k, v in speakers.items()}
    return {k: " ".join(v) for k, v in speakers.items()}


def chunk_text(text: str, max_tokens: int = 2048) -> List[str]:
    words = text.split()
    chunks: List[str] = []
    temp: List[str] = []
    for w in words:
        temp.append(w)
        enc = tok(" ".join(temp), truncation=True, max_length=max_tokens)
        if len(enc["input_ids"]) >= max_tokens:
            temp.pop()
            chunks.append(" ".join(temp))
            temp = [w]
    if temp:
        chunks.append(" ".join(temp))
    return chunks


def translate_to_english(text: str) -> str:
    if not text or not text.strip():
        return text
    ascii_chars = sum(1 for ch in text if ord(ch) < 128)
    ascii_ratio = ascii_chars / max(1, len(text))
    if ascii_ratio > 0.9:
        return text
    try:
        out = translator(text, max_length=512)
        if isinstance(out, list) and out:
            return out[0]["translation_text"]
    except Exception:
        return text
    return text


def emotion_scores(text: str) -> Dict[str, float]:
    res = emo_head(text)[0]
    return {x["label"]: float(x["score"]) for x in res}


def emotions_over_chunks(chunks: List[str]) -> Dict[str, float]:
    if not chunks:
        return {}
    sums: Dict[str, float] = {}
    count = 0
    for c in chunks:
        translated = translate_to_english(c)
        e = emotion_scores(translated)
        for k, v in e.items():
            sums[k] = sums.get(k, 0.0) + v
        count += 1
    return {k: v / count for k, v in sums.items()} if count else {}


def compute_risk(emotions: Dict[str, float]) -> float:
    neg = sum(emotions.get(k, 0.0) for k in negative_keys)
    strongest_neg = max((emotions.get(k, 0.0) for k in negative_keys), default=0.0)
    risk = 0.7 * neg + 0.3 * strongest_neg
    return max(0.0, min(1.0, risk))


def dominant_emotions(emotions: Dict[str, float], top_n: int = 2, threshold: float = 0.2):
    if not emotions:
        return []
    sorted_items = sorted(emotions.items(), key=lambda x: x[1], reverse=True)
    dom = [k for k, v in sorted_items if v >= threshold]
    if not dom:
        dom = [sorted_items[0][0]]
    return dom[:top_n]


def summarize_person(name: str, text: str, risk: float, emotions: Dict[str, float]) -> str:
    emo_str_for_prompt = ", ".join(f"{k}: {round(v,3)}" for k, v in emotions.items())
    prompt = (
        "You are a clinical psychologist describing one person from a chat. "
        "Write a short summary in THIRD PERSON about this person only. "
        "Explain briefly: what they mainly talked about, what they seem to feel, "
        "and how they are coping with work or life. "
        "IMPORTANT: Do NOT copy or quote any sentences from the chat. "
        "Do NOT use lines like 'Name:' or repeat the exact wording. "
        "Write 4 to 6 ORIGINAL sentences in your own words.\n"
        f"Person name: {name}\n"
        f"Risk score (0-1): {round(risk,3)}\n"
        f"Emotion scores: {emo_str_for_prompt}\n"
        f"Conversation from this person:\n{text[:2500]}"
    )
    out = flan(prompt, max_length=220, do_sample=False)[0]["generated_text"].strip()
    return out


def hybrid_suggestions(name: str, summary: str, risk: float, emotions: Dict[str, float]) -> str:
    emo_str_for_prompt = ", ".join(f"{k}: {round(v,3)}" for k, v in emotions.items())
    neg_sum = sum(emotions.get(k, 0.0) for k in negative_keys)
    pos_sum = sum(emotions.get(k, 0.0) for k in positive_keys)
    prompt = (
        "You are a therapist AND a practical workplace coach giving advice directly to this person. "
        "Use the summary below only as background. "
        "You MUST NOT repeat sentences or phrases from the summary. "
        "Do NOT retell what happened in the chat. "
        "Instead, give 4 to 6 sentences of specific, realistic suggestions that mix emotional support "
        "and workplace strategies. Include both coping ideas (breathing, journaling, breaks, talking to someone) "
        "AND practical tips (communication, planning, boundaries, routines). "
        "Keep the tone gentle and hopeful.\n"
        f"Person name: {name}\n"
        f"Risk score (0-1): {round(risk,3)}\n"
        f"Total negative emotion: {round(neg_sum,3)}\n"
        f"Total positive emotion: {round(pos_sum,3)}\n"
        f"Emotion scores: {emo_str_for_prompt}\n"
        f"Summary of this person:\n{summary}"
    )
    out = flan(prompt, max_length=230, do_sample=False)[0]["generated_text"].strip()
    return out


def build_two_line_overall_summary(results: List[Dict[str, Any]], group_emo: Dict[str, float]) -> str:
    if not results:
        return "No conversation detected."
    names = [r["name"] for r in results]
    if len(names) == 1:
        name_part = names[0]
    else:
        name_part = ", ".join(names[:-1]) + " and " + names[-1]
    avg_risk = sum(r["risk"] for r in results) / len(results)
    if avg_risk > 0.7:
        risk_text = "are experiencing intense emotional strain related to this conversation."
    elif avg_risk > 0.45:
        risk_text = "are dealing with noticeable stress and emotional discomfort."
    else:
        risk_text = "show mostly manageable emotions with some moments of stress."
    if group_emo:
        top_emos = sorted(group_emo.items(), key=lambda x: x[1], reverse=True)[:3]
        emo_part = ", ".join(k for k, _ in top_emos)
        emo_text = f"The most prominent emotions in the group are {emo_part}."
    else:
        emo_text = "The emotional tone of the conversation is relatively neutral."
    return f"{name_part} {risk_text} {emo_text}"


def plot_to_base64(fig) -> str:
    buf = io.BytesIO()
    fig.savefig(buf, format="png", bbox_inches="tight")
    buf.seek(0)
    img_base64 = base64.b64encode(buf.read()).decode("utf-8")
    plt.close(fig)
    return img_base64


@app.post("/analyze")
async def analyze(
    text_input: Optional[str] = Form(None),
    privacy: str = Form("OFF"),
    images: List[UploadFile] = File(None),
    videos: List[UploadFile] = File(None),
):
    collected: List[str] = []
    if text_input and text_input.strip():
        collected.append(text_input)
    if images:
        for img_file in images:
            img_bytes = await img_file.read()
            img = Image.open(io.BytesIO(img_bytes))
            t = ocr_image(img)
            if t.strip():
                collected.append(t)
    if videos:
        for vid_file in videos:
            vid_bytes = await vid_file.read()
            t = ocr_video_bytes(vid_bytes)
            if t.strip():
                collected.append(t)
    if not collected:
        return {"error": "No readable text found."}
    combined = "\n".join(collected)
    speakers = split_by_speaker(combined, privacy == "ON")
    results: List[Dict[str, Any]] = []
    for name, txt in speakers.items():
        chunks = chunk_text(txt)
        emos = emotions_over_chunks(chunks)
        risk = compute_risk(emos)
        summary = summarize_person(name, txt, risk, emos)
        feedback = hybrid_suggestions(name, summary, risk, emos)
        results.append(
            {
                "name": name,
                "risk": risk,
                "emotions": emos,
                "summary": summary,
                "feedback": feedback,
            }
        )
    fig1, ax = plt.subplots(1, 2, figsize=(11, 4))
    names = [x["name"] for x in results]
    scores = [x["risk"] for x in results]
    ax[0].bar(names, scores, color="#B03A2E")
    ax[0].set_ylim(0, 1)
    ax[0].set_title("Risk Levels")
    group_emo: Dict[str, float] = {}
    for r in results:
        for k, v in r["emotions"].items():
            group_emo[k] = group_emo.get(k, 0.0) + v
    group_emo = {k: v / len(results) for k, v in group_emo.items()}
    ax[1].bar(list(group_emo.keys()), list(group_emo.values()), color="#2E86C1")
    ax[1].set_ylim(0, 1)
    ax[1].set_title("Group Emotion")
    plt.tight_layout()
    plot1_b64 = plot_to_base64(fig1)
    n = len(results)
    cols = min(3, n)
    rows = math.ceil(n / cols)
    fig2, ax2 = plt.subplots(rows, cols, figsize=(5 * cols, 3 * rows))
    axlist = [ax2] if n == 1 else ax2.flatten()
    for i, r in enumerate(results):
        axp = axlist[i]
        axp.bar(list(r["emotions"].keys()), list(r["emotions"].values()), color="#17A589")
        axp.set_ylim(0, 1)
        axp.set_title(r["name"])
        axp.tick_params(axis="x", rotation=45)
    for j in range(len(axlist) - n):
        axlist[n + j].axis("off")
    fig2.tight_layout()
    plot2_b64 = plot_to_base64(fig2)
    overall_summary = build_two_line_overall_summary(results, group_emo)
    return {
        "overall_summary": overall_summary,
        "results": results,
        "group_emotions": group_emo,
        "plot1": plot1_b64,
        "plot2": plot2_b64,
    }


@app.get("/")
async def root():
    return {"message": "Mental Health Chat Analyzer API"}


if __name__ == "__main__":
    import uvicorn

    uvicorn.run(app, host="0.0.0.0", port=8000)