Spaces:
Runtime error
Runtime error
File size: 32,304 Bytes
aa015c0 743efb8 aa015c0 dd2fc2e aa015c0 358d0ad aa015c0 cd60e73 aa015c0 9ce56d0 aa015c0 9ce56d0 743efb8 aa015c0 743efb8 9ce56d0 3e55ee4 dd2fc2e 3e55ee4 9ce56d0 743efb8 3e55ee4 dd2fc2e 3e55ee4 dd2fc2e 3e55ee4 dd2fc2e 3e55ee4 dd2fc2e 3e55ee4 dd2fc2e 3e55ee4 dd2fc2e 3e55ee4 dd2fc2e 3e55ee4 743efb8 aa015c0 9ce56d0 aa015c0 9ce56d0 aa015c0 9ce56d0 aa015c0 9ce56d0 aa015c0 9ce56d0 aa015c0 743efb8 aa015c0 cd60e73 aa015c0 743efb8 9ce56d0 743efb8 9ce56d0 aa015c0 743efb8 9ce56d0 743efb8 3e55ee4 aa015c0 cd60e73 743efb8 aa015c0 743efb8 aa015c0 743efb8 9ce56d0 743efb8 9ce56d0 cd60e73 9ce56d0 743efb8 aa015c0 3e55ee4 dd2fc2e 3e55ee4 9ce56d0 743efb8 aa015c0 9ce56d0 dd2fc2e 3e55ee4 dd2fc2e aa015c0 743efb8 cd60e73 743efb8 9ce56d0 dd2fc2e cd60e73 743efb8 cd60e73 9ce56d0 dd2fc2e 743efb8 3e55ee4 dd2fc2e 3e55ee4 dd2fc2e 3e55ee4 9ce56d0 cd60e73 9ce56d0 aa015c0 9ce56d0 aa015c0 9ce56d0 aa015c0 9ce56d0 aa015c0 743efb8 9ce56d0 aa015c0 743efb8 9ce56d0 743efb8 9ce56d0 743efb8 9ce56d0 743efb8 9ce56d0 743efb8 9ce56d0 743efb8 cd60e73 3e55ee4 cd60e73 aa015c0 9ce56d0 aa015c0 9ce56d0 aa015c0 9ce56d0 aa015c0 9ce56d0 42b797d dd2fc2e aa015c0 42b797d 9ce56d0 aa015c0 9ce56d0 42b797d aa015c0 9ce56d0 aa015c0 743efb8 aa015c0 9ce56d0 42b797d dd2fc2e aa015c0 9ce56d0 42b797d 743efb8 aa015c0 9ce56d0 42b797d dd2fc2e aa015c0 56b11b0 f423228 aa015c0 9ce56d0 81e292c aa015c0 dd2fc2e 81e292c aa015c0 dd2fc2e 81e292c dd2fc2e f423228 358d0ad aa015c0 dd2fc2e 358d0ad f423228 358d0ad aa015c0 dd2fc2e 49a0eef aa015c0 358d0ad aa015c0 dd2fc2e 49a0eef aa015c0 358d0ad aa015c0 dd2fc2e 49a0eef f423228 aa015c0 dd2fc2e 49a0eef aa015c0 49a0eef aa015c0 9ce56d0 aa015c0 9ce56d0 aa015c0 9ce56d0 aa015c0 743efb8 aa015c0 cd60e73 dd2fc2e aa015c0 81e292c aa015c0 743efb8 aa015c0 cd60e73 aa015c0 743efb8 dd2fc2e aa015c0 743efb8 aa015c0 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 | import os
import sys
import json
import pickle
import torch
import torch.nn as nn
import numpy as np
import re
from typing import Dict, List, Any, Optional
from collections import defaultdict, Counter
import networkx as nx
import pymorphy3
import requests
from fastapi import FastAPI, Request, Form, HTTPException
from fastapi.responses import HTMLResponse, JSONResponse
from fastapi.templating import Jinja2Templates
import uvicorn
from transformers import BertTokenizer, BertModel
from sklearn.preprocessing import LabelEncoder
import warnings
warnings.filterwarnings('ignore')
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
print(f"Используется устройство: {device}")
def clean_russian_text(text):
if not isinstance(text, str):
return ""
text = text.lower()
text = re.sub(r'http\S+|www\S+|https\S+', '', text)
text = re.sub(r'\S+@\S+', '', text)
smileys = {
':)': ' смайлик_радость ', ')': ' смайлик_радость ',
':(': ' смайлик_грусть ', '(': ' смайлик_грусть ',
':D': ' смайлик_смех ', ';)': ' смайлик_подмигивание ',
}
for smiley, replacement in smileys.items():
text = text.replace(smiley, replacement)
text = re.sub(r'[^\w\sа-яё.,!?;:)(-]', ' ', text)
text = re.sub(r'\s+', ' ', text).strip()
return text
# ============================================================
# ПОЛНЫЙ КЛАСС ОНТОЛОГИИ (исправленный)
# ============================================================
class OntologyEmotionModel:
def __init__(self, emotions: List[str], train_texts: List[str] = None, train_labels: List[int] = None):
self.emotions = emotions
self.morph = pymorphy3.MorphAnalyzer()
self.ontology_graph = nx.DiGraph()
self.empirical_base = defaultdict(list)
self.hypotheses_db = {}
self.verified_hypotheses = defaultdict(list)
self.sentiment_lexicon = {}
self.rule_stats = {}
if train_texts is not None and train_labels is not None:
self._build_sentiment_lexicon(train_texts, train_labels)
self._load_rusentilex()
self.init_ontology_level1()
self.init_ontology_level2()
def _build_sentiment_lexicon(self, texts: List[str], labels: List[int]):
word_class_counts = defaultdict(lambda: np.zeros(len(self.emotions)))
for text, label in zip(texts, labels):
words = set(clean_russian_text(text).split())
for word in words:
lemma = self.morph.parse(word)[0].normal_form
word_class_counts[lemma][label] += 1
for lemma, counts in word_class_counts.items():
prob = counts / (counts.sum() + 1e-10)
if prob.max() > 0.6 and counts.sum() > 5:
dominant_class = self.emotions[np.argmax(prob)]
self.sentiment_lexicon[lemma] = dominant_class
def _parse_rusentilex(self, content):
lines = content.splitlines()
added = 0
for line in lines[1:]: # пропускаем заголовок
try:
parts = line.strip().split(',')
if len(parts) >= 3:
word = parts[0].strip().lower()
sentiment = parts[2].strip().lower()
lemma = self.morph.parse(word)[0].normal_form
if sentiment == 'positive':
self.sentiment_lexicon[lemma] = 'радость'
added += 1
elif sentiment == 'negative':
self.sentiment_lexicon[lemma] = 'грусть'
added += 1
except Exception as e:
continue
print(f" Добавлено слов из RuSentiLex: {added}")
def _load_rusentilex(self):
"""Загружает RuSentiLex из локального файла в папке model"""
import os
# Пути для поиска файла RuSentiLex
possible_paths = [
'model/rusentilex.csv',
'rusentilex.csv',
'/app/model/rusentilex.csv',
os.path.join(os.path.dirname(__file__), 'model', 'rusentilex.csv')
]
loaded = False
print("📂 Поиск RuSentiLex...")
# Пробуем загрузить из локального файла
for path in possible_paths:
if os.path.exists(path):
try:
with open(path, 'r', encoding='utf-8') as f:
content = f.read()
self._parse_rusentilex(content)
print(f"✅ RuSentiLex загружен из файла: {path}")
loaded = True
break
except Exception as e:
print(f"⚠️ Ошибка при загрузке {path}: {e}")
# Если локально не нашли, пробуем скачать из интернета
if not loaded:
print("⚠️ Локальный файл RuSentiLex не найден, пробуем скачать...")
url = "https://raw.githubusercontent.com/nicolay-r/sentiment-relation-classifiers/master/data/rusentilex.csv"
try:
r = requests.get(url, timeout=10)
if r.status_code == 200:
self._parse_rusentilex(r.text)
print("✅ RuSentiLex загружен из репозитория")
loaded = True
except Exception as e:
print(f"⚠️ Не удалось загрузить RuSentiLex из репозитория: {e}")
if not loaded:
print("⚠️ RuSentiLex не загружен. Используется только статистический лексикон.")
# Выводим статистику
print(f"📊 Всего слов в лексиконе: {len(self.sentiment_lexicon)}")
def init_ontology_level1(self):
self.emotion_definitions = {
'радость': {
'valence': 'positive', 'arousal': 'high',
'definition': 'Позитивное эмоциональное состояние',
'opposite': ['грусть', 'злость']
},
'грусть': {
'valence': 'negative', 'arousal': 'low',
'definition': 'Негативное эмоциональное состояние',
'opposite': ['радость']
},
'злость': {
'valence': 'negative', 'arousal': 'high',
'definition': 'Негативное эмоциональное состояние',
'opposite': ['радость']
},
'страх': {
'valence': 'negative', 'arousal': 'high',
'definition': 'Эмоциональная реакция на угрозу',
'opposite': ['уверенность', 'спокойствие']
},
'сарказм': {
'valence': 'negative', 'arousal': 'high',
'definition': 'Язвительная насмешка',
'opposite': ['радость']
}
}
for emotion in self.emotions:
if emotion in self.emotion_definitions:
self.ontology_graph.add_node(emotion, **self.emotion_definitions[emotion])
else:
self.ontology_graph.add_node(emotion, valence='neutral', arousal='neutral')
for emotion, data in self.emotion_definitions.items():
if 'opposite' in data:
for opposite in data['opposite']:
if opposite in self.emotions:
self.ontology_graph.add_edge(emotion, opposite, relation='opposite')
def init_ontology_level2(self):
self.linguistic_rules = {
'усилители': {
'words': ['очень', 'сильно', 'крайне', 'чрезвычайно', 'невероятно', 'абсолютно'],
'effect': 'increase_arousal',
'weight': 0.3,
'learnable': True
},
'ослабители': {
'words': ['слегка', 'немного', 'чуть-чуть', 'отчасти', 'несколько'],
'effect': 'decrease_arousal',
'weight': -0.2,
'learnable': True
},
'отрицания': {
'words': ['не', 'ни', 'нет', 'нельзя', 'невозможно'],
'effect': 'negation',
'weight': -0.5,
'learnable': True
},
'восклицания': {
'patterns': [r'!+', r'\?+'],
'effect': 'increase_arousal',
'weight': 0.4,
'learnable': True
},
'вопросительные': {
'patterns': [r'\?+'],
'effect': 'uncertainty',
'weight': 0.2,
'learnable': True
},
'сарказм_маркеры': {
'words': ['какой', 'такой', 'прям', 'ага', 'ну да', 'конечно', 'отличная работа', 'прекрасно', 'замечательно', 'как всегда'],
'effect': 'sarcasm',
'weight': 0.6,
'learnable': True
}
}
def add_empirical_knowledge(self, text: str, emotion: str, confidence: float):
self.empirical_base[emotion].append({'text': text, 'confidence': confidence})
if len(self.empirical_base[emotion]) > 1000:
self.empirical_base[emotion] = self.empirical_base[emotion][-1000:]
def formulate_hypothesis(self, text: str, model_prediction: Dict, rule_based_prediction: Dict) -> Dict:
hypothesis_id = f"hyp_{len(self.hypotheses_db) + 1:06d}"
hypothesis = {
'id': hypothesis_id, 'text': text,
'model_prediction': model_prediction,
'rule_based_prediction': rule_based_prediction,
'disagreement': self.calculate_disagreement(model_prediction, rule_based_prediction),
'status': 'pending'
}
self.hypotheses_db[hypothesis_id] = hypothesis
return hypothesis
def verify_hypothesis(self, hypothesis_id: str, actual_emotion: str = None) -> Dict:
if hypothesis_id not in self.hypotheses_db:
return None
hypothesis = self.hypotheses_db[hypothesis_id]
if actual_emotion:
model_correct = hypothesis['model_prediction']['emotion'] == actual_emotion
rule_correct = hypothesis['rule_based_prediction']['emotion'] == actual_emotion
if model_correct and not rule_correct:
hypothesis['status'] = 'model_superior'
elif rule_correct and not model_correct:
hypothesis['status'] = 'rule_superior'
elif model_correct and rule_correct:
hypothesis['status'] = 'both_correct'
else:
hypothesis['status'] = 'both_incorrect'
return hypothesis
def apply_linguistic_rules(self, text: str) -> Dict:
rules_applied = []
adjustments = {'valence': 0, 'arousal': 0, 'uncertainty': 0, 'sarcasm': 0}
words = text.lower().split()
parsed = [self.morph.parse(w)[0] for w in words]
lemmas = [p.normal_form for p in parsed]
pos_tags = [p.tag.POS for p in parsed]
# Проверка на слова из лексикона
for lemma in lemmas:
sentiment = self.sentiment_lexicon.get(lemma, 'neutral')
if sentiment == 'радость':
rules_applied.append(f"позитивное слово: {lemma}")
adjustments['valence'] += 0.2
elif sentiment in ('грусть', 'злость', 'страх'):
rules_applied.append(f"негативное слово: {lemma}")
adjustments['valence'] -= 0.2
for category, rule in self.linguistic_rules.items():
if 'words' in rule:
for word in rule['words']:
if word in lemmas:
rules_applied.append(f"{category}: {word}")
effect = rule['effect']
weight = rule['weight']
if effect == 'increase_arousal':
adjustments['arousal'] += weight
elif effect == 'decrease_arousal':
adjustments['arousal'] += weight
elif effect == 'negation':
adjustments['valence'] += weight
elif effect == 'sarcasm':
adjustments['sarcasm'] += weight
if 'patterns' in rule:
for pattern in rule['patterns']:
if re.search(pattern, text):
rules_applied.append(f"{category}: {pattern}")
weight = rule['weight']
if rule['effect'] == 'increase_arousal':
adjustments['arousal'] += weight
elif rule['effect'] == 'uncertainty':
adjustments['uncertainty'] += weight
if 'не' in lemmas:
idx = lemmas.index('не')
if idx + 1 < len(lemmas) and lemmas[idx+1] == 'очень':
adjustments['arousal'] -= 0.2
adjustments['valence'] -= 0.1
rules_applied.append("сочетание: не очень")
else:
for j in range(idx+1, min(idx+4, len(lemmas))):
if pos_tags[j] in ('ADJF', 'ADJS', 'ADVB'):
target_word = lemmas[j]
sentiment = self.sentiment_lexicon.get(target_word, 'neutral')
if sentiment in ('грусть', 'злость', 'страх'):
adjustments['valence'] += 1.0
rules_applied.append(f"инверсия негатива: не {target_word}")
elif sentiment == 'радость':
adjustments['valence'] -= 1.0
rules_applied.append(f"инверсия позитива: не {target_word}")
break
pos_words = [w for w in lemmas if self.sentiment_lexicon.get(w) == 'радость']
neg_words = [w for w in lemmas if self.sentiment_lexicon.get(w) in ('грусть', 'злость', 'страх')]
if pos_words and neg_words:
adjustments['sarcasm'] += 0.5
rules_applied.append(f"контраст тональности: позитив {pos_words[:2]} vs негатив {neg_words[:2]}")
# Дополнительная проверка на саркастические фразы
sarcasm_phrases = ['конечно', 'ага', 'ну да', 'как всегда', 'отличная работа', 'прекрасно', 'замечательно']
for phrase in sarcasm_phrases:
if phrase in text.lower():
adjustments['sarcasm'] += 0.6
rules_applied.append(f"саркастическая фраза: {phrase}")
if adjustments['sarcasm'] > 0.5:
rules_applied.append("обнаружен сарказм")
return {'rules_applied': rules_applied, 'adjustments': adjustments, 'lemmas': lemmas}
def calculate_disagreement(self, pred1: Dict, pred2: Dict) -> float:
if pred1['emotion'] == pred2['emotion']:
return 0.0
emotions = list(self.emotion_definitions.keys())
idx1 = emotions.index(pred1['emotion']) if pred1['emotion'] in emotions else -1
idx2 = emotions.index(pred2['emotion']) if pred2['emotion'] in emotions else -1
if idx1 == -1 or idx2 == -1:
return 0.5
distance = abs(idx1 - idx2) / len(emotions)
return 0.7 * distance
def explain_transition(self, from_emotion: str, to_emotion: str) -> List[str]:
try:
return nx.shortest_path(self.ontology_graph, source=from_emotion, target=to_emotion)
except:
return []
def adjust_prediction_with_rules(self, prediction: Dict, rule_analysis: Dict) -> Dict:
original_emotion = prediction['emotion']
original_confidence = prediction['confidence']
adj = rule_analysis['adjustments']
rules = rule_analysis['rules_applied']
# Сохраняем исходную уверенность для проверки коррекции
original_confidence_value = original_confidence
was_corrected = len(rules) > 0
conf_mult = 1.0 + adj['arousal'] * 0.2 + adj['uncertainty'] * 0.1 - abs(adj['valence']) * 0.1
conf_mult = np.clip(conf_mult, 0.5, 1.5)
new_confidence = original_confidence * conf_mult
new_emotion = original_emotion
# Если есть негативные слова и нет позитивных, корректируем эмоцию
has_negative = any('негативное слово' in r for r in rules)
has_positive = any('позитивное слово' in r for r in rules)
if has_negative and not has_positive:
if original_emotion == 'радость':
new_emotion = 'грусть'
new_confidence *= 0.8
rules.append("коррекция: негативные слова без позитивных")
elif original_emotion == 'сарказм':
new_emotion = 'грусть'
new_confidence *= 0.9
elif has_positive and not has_negative and original_emotion in ('грусть', 'злость', 'страх'):
new_emotion = 'радость'
rules.append("коррекция: позитивные слова")
# Инверсия на основе правил
for rule in rules:
if rule.startswith("инверсия негатива:"):
new_emotion = 'радость'
break
elif rule.startswith("инверсия позитива:"):
if adj['arousal'] > 0.3:
new_emotion = 'злость'
else:
new_emotion = 'грусть'
break
# Сарказм (контраст + маркеры)
sarcasm_flag = adj['sarcasm'] > 0.5
if sarcasm_flag:
new_emotion = 'сарказм'
new_confidence = min(new_confidence * 0.8, 0.9)
if "саркастическая фраза" in str(rules):
new_confidence = min(new_confidence * 1.1, 0.95)
# Восклицания
if any('восклицание' in r for r in rules):
new_confidence = min(new_confidence * 1.2, 1.0)
# Если онтология не применила коррекции, а уверенность была менее 90%,
# то повышаем уверенность на 10% (но не более 100%)
if not was_corrected and original_confidence_value < 0.9:
new_confidence = min(new_confidence * 1.10, 1.0)
# Ограничиваем максимум 1.0 (100%)
new_confidence = min(new_confidence, 1.0)
return {
'emotion': new_emotion,
'confidence': new_confidence,
'rules_applied': rules
}
def get_ontology_analysis(self, text: str, model_prediction: Dict) -> Dict:
rule_analysis = self.apply_linguistic_rules(text)
adjusted = self.adjust_prediction_with_rules(model_prediction, rule_analysis)
disagreement = self.calculate_disagreement(model_prediction, adjusted)
hypothesis = self.formulate_hypothesis(text, model_prediction, adjusted) if disagreement > 0.2 else None
return {
'rule_analysis': rule_analysis,
'adjusted_prediction': adjusted,
'disagreement': disagreement,
'hypothesis': hypothesis
}
def get_statistics(self) -> Dict:
return {
'ontology_nodes': len(self.ontology_graph.nodes),
'ontology_edges': len(self.ontology_graph.edges),
'linguistic_rules': len(self.linguistic_rules),
'emotions_covered': len(self.emotions),
'pending_hypotheses': len([h for h in self.hypotheses_db.values() if h['status'] == 'pending'])
}
# ============================================================
# КЛАССЫ МОДЕЛЕЙ LSTM и BERT
# ============================================================
class EmotionLSTM(nn.Module):
def __init__(self, vocab_size, embed_dim=128, hidden_dim=256, num_classes=3, dropout=0.3, num_layers=2):
super().__init__()
self.embedding = nn.Embedding(vocab_size, embed_dim, padding_idx=0)
self.lstm = nn.LSTM(embed_dim, hidden_dim, num_layers, batch_first=True, bidirectional=True, dropout=dropout)
self.dropout = nn.Dropout(dropout)
self.classifier = nn.Sequential(
nn.Linear(hidden_dim * 2, 128), nn.ReLU(), nn.Dropout(dropout),
nn.Linear(128, 64), nn.ReLU(), nn.Linear(64, num_classes)
)
def forward(self, x, return_confidence=False):
embedded = self.embedding(x)
lstm_out, (hidden, cell) = self.lstm(embedded)
lstm_last = torch.cat((hidden[-2, :, :], hidden[-1, :, :]), dim=1)
features = self.dropout(lstm_last)
logits = self.classifier(features)
if return_confidence:
probs = torch.softmax(logits, dim=1)
conf, _ = torch.max(probs, dim=1)
return logits, conf
return logits
class EmotionBERT(nn.Module):
def __init__(self, bert_model_name, num_classes, dropout=0.3):
super().__init__()
self.bert = BertModel.from_pretrained(bert_model_name)
hidden = self.bert.config.hidden_size
self.classifier = nn.Sequential(
nn.Dropout(dropout), nn.Linear(hidden, 256), nn.ReLU(),
nn.Dropout(dropout), nn.Linear(256, 128), nn.ReLU(),
nn.Linear(128, num_classes)
)
def forward(self, input_ids, attention_mask, return_confidence=False):
out = self.bert(input_ids, attention_mask, return_dict=True)
cls = out.last_hidden_state[:, 0, :]
logits = self.classifier(cls)
if return_confidence:
probs = torch.softmax(logits, dim=1)
conf, _ = torch.max(probs, dim=1)
return logits, conf
return logits
# ============================================================
# КАСКАДНЫЙ КЛАССИФИКАТОР
# ============================================================
class CascadeEmotionClassifier:
def __init__(self, lstm_model, bert_model, vocab, tokenizer, label_encoder, ontology_model, threshold=0.95, device='cpu', max_length_lstm=100, max_length_bert=128):
self.lstm_model = lstm_model
self.bert_model = bert_model
self.vocab = vocab
self.tokenizer = tokenizer
self.label_encoder = label_encoder
self.ontology_model = ontology_model
self.threshold = threshold
self.device = device
self.max_length_lstm = max_length_lstm
self.max_length_bert = max_length_bert
self.lstm_model.eval()
self.bert_model.eval()
self.lstm_model.to(device)
self.bert_model.to(device)
self.stats = {'total': 0, 'lstm': 0, 'bert': 0, 'corrections': 0}
def text_to_sequence(self, text):
words = str(text).split()[:self.max_length_lstm]
sequence = [self.vocab.get(word, self.vocab.get('<UNK>', 1)) for word in words]
if len(sequence) < self.max_length_lstm:
sequence += [self.vocab.get('<PAD>', 0)] * (self.max_length_lstm - len(sequence))
return sequence[:self.max_length_lstm]
def predict(self, text):
self.stats['total'] += 1
text_clean = clean_russian_text(text)
seq = torch.LongTensor([self.text_to_sequence(text_clean)]).to(self.device)
with torch.no_grad():
lstm_logits, lstm_conf = self.lstm_model(seq, return_confidence=True)
lstm_probs = torch.softmax(lstm_logits, dim=1)
lstm_pred = lstm_probs.argmax().item()
lstm_emo = self.label_encoder.inverse_transform([lstm_pred])[0]
lstm_pred_dict = {'emotion': lstm_emo, 'confidence': lstm_conf.item(), 'probabilities': lstm_probs[0].cpu().numpy().tolist()}
# Применяем онтологию к LSTM
lstm_onto = self.ontology_model.get_ontology_analysis(text_clean, lstm_pred_dict)
if lstm_onto['adjusted_prediction']['confidence'] >= self.threshold:
self.stats['lstm'] += 1
final = lstm_onto['adjusted_prediction']
used = "LSTM + онтология"
rules_applied = lstm_onto['rule_analysis']['rules_applied']
else:
self.stats['bert'] += 1
enc = self.tokenizer(text_clean, truncation=True, padding=True, max_length=self.max_length_bert, return_tensors='pt').to(self.device)
with torch.no_grad():
bert_logits, bert_conf = self.bert_model(enc['input_ids'], enc['attention_mask'], return_confidence=True)
bert_probs = torch.softmax(bert_logits, dim=1)
bert_pred = bert_probs.argmax().item()
bert_emo = self.label_encoder.inverse_transform([bert_pred])[0]
bert_pred_dict = {'emotion': bert_emo, 'confidence': bert_conf.item(), 'probabilities': bert_probs[0].cpu().numpy().tolist()}
# Применяем онтологию к BERT
bert_onto = self.ontology_model.get_ontology_analysis(text_clean, bert_pred_dict)
final = bert_onto['adjusted_prediction']
used = "BERT + онтология"
rules_applied = bert_onto['rule_analysis']['rules_applied']
return {
'text': text,
'predicted_emotion': final['emotion'],
'confidence': float(final['confidence']),
'used_model': used,
'rules_applied': rules_applied,
'was_corrected_by_ontology': len(rules_applied) > 0
}
# ============================================================
# ЗАГРУЗКА МОДЕЛИ
# ============================================================
def load_model():
print("Загрузка модели...")
model_dir = 'model'
# Загружаем информацию о модели
with open(f'{model_dir}/model_info.json', 'r', encoding='utf-8') as f:
model_info = json.load(f)
# Загружаем vocab
with open(f'{model_dir}/vocab.json', 'r', encoding='utf-8') as f:
vocab = json.load(f)
# СОЗДАЁМ label_encoder из model_info
print("📂 Создание label_encoder...")
label_encoder = LabelEncoder()
label_encoder.classes_ = np.array(model_info['classes'])
print(f"✅ label_encoder создан, классы: {list(label_encoder.classes_)}")
# СОЗДАЁМ онтологию
print("📂 Создание онтологии...")
ontology_model = OntologyEmotionModel(
emotions=list(label_encoder.classes_),
train_texts=None,
train_labels=None
)
print("✅ Онтология создана")
# LSTM
print("📂 Загрузка LSTM...")
lstm_model = EmotionLSTM(
vocab_size=len(vocab),
embed_dim=model_info.get('embed_dim', 300),
hidden_dim=256,
num_classes=model_info['num_classes'],
dropout=0.3,
num_layers=2
)
checkpoint = torch.load(f'{model_dir}/lstm_model.pth', map_location=device, weights_only=False)
lstm_model.load_state_dict(checkpoint['model_state_dict'])
print("✅ LSTM загружена")
# BERT
print("📂 Загрузка BERT...")
bert_model = EmotionBERT(
bert_model_name=model_info['bert_model_name'],
num_classes=model_info['num_classes'],
dropout=0.3
)
bert_model.load_state_dict(torch.load(f'{model_dir}/bert_model.pth', map_location=device, weights_only=False))
print("✅ BERT загружена")
# Токенизатор
print("📂 Загрузка токенизатора...")
try:
tokenizer = BertTokenizer.from_pretrained(model_dir)
print("✅ Токенизатор загружен из model_dir")
except Exception as e:
print(f"⚠️ Ошибка: {e}")
print("🔄 Загружаем токенизатор из Hugging Face...")
tokenizer = BertTokenizer.from_pretrained('DeepPavlov/rubert-base-cased')
print("✅ Токенизатор загружен из Hugging Face")
# Каскад
print("📂 Создание каскадного классификатора...")
cascade = CascadeEmotionClassifier(
lstm_model=lstm_model,
bert_model=bert_model,
vocab=vocab,
tokenizer=tokenizer,
label_encoder=label_encoder,
ontology_model=ontology_model,
threshold=model_info.get('threshold', 0.95),
device=device,
max_length_lstm=model_info.get('max_length_lstm', 100),
max_length_bert=model_info.get('max_length_bert', 128)
)
print("✅ Модель успешно загружена!")
return cascade, model_info
# ============================================================
# FASTAPI ПРИЛОЖЕНИЕ
# ============================================================
app = FastAPI(title="Emotion Analysis with BERT and Ontology")
templates = Jinja2Templates(directory="templates")
classifier = None
model_info = None
@app.on_event("startup")
async def startup_event():
global classifier, model_info
classifier, model_info = load_model()
@app.get("/", response_class=HTMLResponse)
async def home(request: Request):
return templates.TemplateResponse("index.html", {"request": request})
@app.post("/predict")
async def predict(text: str = Form(...)):
if not classifier:
raise HTTPException(status_code=503, detail="Модель не загружена")
if not text or len(text.strip()) < 3:
return JSONResponse({"error": "Введите хотя бы 3 символа."}, status_code=400)
try:
result = classifier.predict(text)
# Форматируем правила для отображения
rules_display = []
for rule in result['rules_applied'][:10]:
if ':' in rule:
cat, val = rule.split(':', 1)
rules_display.append(f"<span class='rule-tag'>{cat}: {val}</span>")
else:
rules_display.append(f"<span class='rule-tag'>{rule}</span>")
return JSONResponse({
"success": True,
"emotion": result['predicted_emotion'],
"confidence": f"{result['confidence']*100:.1f}%",
"used_model": result['used_model'],
"rules": "".join(rules_display) if rules_display else "Нет правил",
"was_corrected": str(result['was_corrected_by_ontology'])
})
except Exception as e:
return JSONResponse({"error": str(e)}, status_code=500)
@app.get("/health")
async def health_check():
return {"status": "healthy", "model_loaded": classifier is not None}
if __name__ == "__main__":
uvicorn.run(app, host="0.0.0.0", port=7860) |