Update app.py
Browse files
app.py
CHANGED
@@ -5,6 +5,7 @@ import plotly.graph_objects as go
|
|
5 |
import plotly.express as px
|
6 |
from plotly.subplots import make_subplots
|
7 |
import numpy as np
|
|
|
8 |
from collections import Counter, defaultdict
|
9 |
import re
|
10 |
import json
|
@@ -16,54 +17,11 @@ import logging
|
|
16 |
from functools import lru_cache
|
17 |
from dataclasses import dataclass
|
18 |
from typing import List, Dict, Optional, Tuple
|
|
|
|
|
|
|
19 |
import pandas as pd
|
20 |
|
21 |
-
# 设置日志 - 提前初始化
|
22 |
-
logging.basicConfig(level=logging.INFO)
|
23 |
-
logger = logging.getLogger(__name__)
|
24 |
-
|
25 |
-
# 尝试导入可选依赖
|
26 |
-
try:
|
27 |
-
from wordcloud import WordCloud
|
28 |
-
WORDCLOUD_AVAILABLE = True
|
29 |
-
except ImportError:
|
30 |
-
WORDCLOUD_AVAILABLE = False
|
31 |
-
logger.warning("WordCloud not available")
|
32 |
-
|
33 |
-
try:
|
34 |
-
import nltk
|
35 |
-
from nltk.corpus import stopwords
|
36 |
-
nltk.download('stopwords', quiet=True)
|
37 |
-
nltk.download('punkt', quiet=True)
|
38 |
-
STOP_WORDS = set(stopwords.words('english'))
|
39 |
-
NLTK_AVAILABLE = True
|
40 |
-
except:
|
41 |
-
NLTK_AVAILABLE = False
|
42 |
-
STOP_WORDS = {'the', 'a', 'an', 'and', 'or', 'but', 'in', 'on', 'at', 'to', 'for', 'of', 'with', 'by'}
|
43 |
-
logger.warning("NLTK not available, using basic stopwords")
|
44 |
-
|
45 |
-
try:
|
46 |
-
import langdetect
|
47 |
-
LANGDETECT_AVAILABLE = True
|
48 |
-
except ImportError:
|
49 |
-
LANGDETECT_AVAILABLE = False
|
50 |
-
logger.warning("langdetect not available, using fallback language detection")
|
51 |
-
|
52 |
-
# 尝试导入SHAP和LIME
|
53 |
-
try:
|
54 |
-
import shap
|
55 |
-
SHAP_AVAILABLE = True
|
56 |
-
except ImportError:
|
57 |
-
SHAP_AVAILABLE = False
|
58 |
-
logger.warning("SHAP not available, using basic analysis")
|
59 |
-
|
60 |
-
try:
|
61 |
-
from lime.lime_text import LimeTextExplainer
|
62 |
-
LIME_AVAILABLE = True
|
63 |
-
except ImportError:
|
64 |
-
LIME_AVAILABLE = False
|
65 |
-
logger.warning("LIME not available, using basic analysis")
|
66 |
-
|
67 |
# Configuration
|
68 |
@dataclass
|
69 |
class Config:
|
@@ -83,13 +41,10 @@ class Config:
|
|
83 |
'sv': 'Swedish'
|
84 |
}
|
85 |
|
86 |
-
# 使用更稳定的模型
|
87 |
MODELS = {
|
88 |
'en': "cardiffnlp/twitter-roberta-base-sentiment-latest",
|
89 |
'multilingual': "cardiffnlp/twitter-xlm-roberta-base-sentiment",
|
90 |
-
'zh': "uer/roberta-base-finetuned-dianping-chinese"
|
91 |
-
# 备用模型
|
92 |
-
'fallback': "distilbert-base-uncased-finetuned-sst-2-english"
|
93 |
}
|
94 |
|
95 |
# Color themes
|
@@ -102,80 +57,62 @@ class Config:
|
|
102 |
|
103 |
config = Config()
|
104 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
105 |
class ModelManager:
|
106 |
-
"""Manages multiple language models
|
107 |
def __init__(self):
|
108 |
self.models = {}
|
109 |
self.tokenizers = {}
|
110 |
-
self.device =
|
111 |
-
self.
|
112 |
-
self._load_models()
|
113 |
|
114 |
-
def
|
115 |
-
"""
|
116 |
try:
|
117 |
-
|
118 |
-
return torch.device("cuda")
|
119 |
-
elif hasattr(torch.backends, 'mps') and torch.backends.mps.is_available():
|
120 |
-
return torch.device("mps")
|
121 |
-
else:
|
122 |
-
return torch.device("cpu")
|
123 |
-
except:
|
124 |
-
return torch.device("cpu")
|
125 |
-
|
126 |
-
def _load_models(self):
|
127 |
-
"""Load models with error handling"""
|
128 |
-
try:
|
129 |
-
# 首先尝试加载多语言模型
|
130 |
model_name = config.MODELS['multilingual']
|
131 |
-
logger.info(f"Loading model: {model_name}")
|
132 |
-
|
133 |
self.tokenizers['default'] = AutoTokenizer.from_pretrained(model_name)
|
134 |
self.models['default'] = AutoModelForSequenceClassification.from_pretrained(model_name)
|
135 |
self.models['default'].to(self.device)
|
|
|
136 |
|
137 |
-
|
138 |
-
|
|
|
|
|
|
|
|
|
139 |
|
140 |
except Exception as e:
|
141 |
-
logger.error(f"Failed to load
|
142 |
-
|
143 |
-
try:
|
144 |
-
fallback_model = config.MODELS['fallback']
|
145 |
-
logger.info(f"Trying fallback model: {fallback_model}")
|
146 |
-
|
147 |
-
self.tokenizers['default'] = AutoTokenizer.from_pretrained(fallback_model)
|
148 |
-
self.models['default'] = AutoModelForSequenceClassification.from_pretrained(fallback_model)
|
149 |
-
self.models['default'].to(self.device)
|
150 |
-
|
151 |
-
logger.info(f"Successfully loaded fallback model: {fallback_model}")
|
152 |
-
self.model_loaded = True
|
153 |
-
|
154 |
-
except Exception as e2:
|
155 |
-
logger.error(f"Failed to load fallback model: {e2}")
|
156 |
-
self.model_loaded = False
|
157 |
-
raise RuntimeError("Failed to load any sentiment analysis model")
|
158 |
|
159 |
def get_model(self, language='en'):
|
160 |
"""Get model for specific language"""
|
161 |
-
if
|
162 |
-
|
163 |
-
|
164 |
-
|
165 |
-
return self.models['default'], self.tokenizers['default']
|
166 |
|
167 |
@staticmethod
|
168 |
def detect_language(text: str) -> str:
|
169 |
-
"""Detect text language
|
170 |
-
if not LANGDETECT_AVAILABLE:
|
171 |
-
# 简单的语言检测
|
172 |
-
if re.search(r'[\u4e00-\u9fff]', text):
|
173 |
-
return 'zh'
|
174 |
-
else:
|
175 |
-
return 'en'
|
176 |
-
|
177 |
try:
|
|
|
178 |
detected = langdetect.detect(text)
|
|
|
179 |
language_mapping = {
|
180 |
'zh-cn': 'zh',
|
181 |
'zh-tw': 'zh'
|
@@ -185,23 +122,10 @@ class ModelManager:
|
|
185 |
except:
|
186 |
return 'en'
|
187 |
|
188 |
-
|
189 |
-
model_manager = None
|
190 |
-
|
191 |
-
def initialize_models():
|
192 |
-
"""延迟初始化模型"""
|
193 |
-
global model_manager
|
194 |
-
if model_manager is None:
|
195 |
-
try:
|
196 |
-
model_manager = ModelManager()
|
197 |
-
return True
|
198 |
-
except Exception as e:
|
199 |
-
logger.error(f"Model initialization failed: {e}")
|
200 |
-
return False
|
201 |
-
return True
|
202 |
|
203 |
class HistoryManager:
|
204 |
-
"""Enhanced history manager"""
|
205 |
def __init__(self):
|
206 |
self._history = []
|
207 |
|
@@ -211,6 +135,7 @@ class HistoryManager:
|
|
211 |
self._history = self._history[-config.MAX_HISTORY_SIZE:]
|
212 |
|
213 |
def add_batch_entries(self, entries: List[Dict]):
|
|
|
214 |
for entry in entries:
|
215 |
self.add_entry(entry)
|
216 |
|
@@ -218,10 +143,12 @@ class HistoryManager:
|
|
218 |
return self._history.copy()
|
219 |
|
220 |
def get_recent_history(self, n: int = 10) -> List[Dict]:
|
|
|
221 |
return self._history[-n:] if self._history else []
|
222 |
|
223 |
def filter_history(self, sentiment: str = None, language: str = None,
|
224 |
min_confidence: float = None) -> List[Dict]:
|
|
|
225 |
filtered = self._history
|
226 |
|
227 |
if sentiment:
|
@@ -251,12 +178,12 @@ class HistoryManager:
|
|
251 |
'positive_count': sentiments.count('Positive'),
|
252 |
'negative_count': sentiments.count('Negative'),
|
253 |
'neutral_count': sentiments.count('Neutral'),
|
254 |
-
'avg_confidence': np.mean(confidences)
|
255 |
-
'max_confidence': np.max(confidences)
|
256 |
-
'min_confidence': np.min(confidences)
|
257 |
'languages_detected': len(set(languages)),
|
258 |
'most_common_language': Counter(languages).most_common(1)[0][0] if languages else 'en',
|
259 |
-
'avg_text_length': np.mean([len(item.get('full_text', '')) for item in self._history])
|
260 |
}
|
261 |
|
262 |
history_manager = HistoryManager()
|
@@ -283,12 +210,14 @@ class TextProcessor:
|
|
283 |
@staticmethod
|
284 |
def extract_keywords(text: str, top_k: int = 5) -> List[str]:
|
285 |
"""Extract key words from text"""
|
|
|
286 |
if re.search(r'[\u4e00-\u9fff]', text):
|
287 |
words = re.findall(r'[\u4e00-\u9fff]+', text)
|
288 |
all_chars = ''.join(words)
|
289 |
char_freq = Counter(all_chars)
|
290 |
return [char for char, _ in char_freq.most_common(top_k)]
|
291 |
else:
|
|
|
292 |
cleaned = TextProcessor.clean_text(text)
|
293 |
words = cleaned.split()
|
294 |
word_freq = Counter(words)
|
@@ -301,7 +230,7 @@ class TextProcessor:
|
|
301 |
return [line.strip() for line in lines if line.strip()]
|
302 |
|
303 |
class SentimentAnalyzer:
|
304 |
-
"""Enhanced sentiment analysis
|
305 |
|
306 |
@staticmethod
|
307 |
def analyze_text(text: str, language: str = 'auto', preprocessing_options: Dict = None) -> Dict:
|
@@ -309,10 +238,6 @@ class SentimentAnalyzer:
|
|
309 |
if not text.strip():
|
310 |
raise ValueError("Empty text provided")
|
311 |
|
312 |
-
# 确保模型已加载
|
313 |
-
if not initialize_models():
|
314 |
-
raise RuntimeError("Failed to initialize sentiment analysis models")
|
315 |
-
|
316 |
# Detect language if auto
|
317 |
if language == 'auto':
|
318 |
detected_lang = model_manager.detect_language(text)
|
@@ -320,13 +245,9 @@ class SentimentAnalyzer:
|
|
320 |
detected_lang = language
|
321 |
|
322 |
# Get appropriate model
|
323 |
-
|
324 |
-
model, tokenizer = model_manager.get_model(detected_lang)
|
325 |
-
except Exception as e:
|
326 |
-
logger.error(f"Failed to get model: {e}")
|
327 |
-
raise RuntimeError(f"Model loading failed: {e}")
|
328 |
|
329 |
-
# Preprocessing
|
330 |
options = preprocessing_options or {}
|
331 |
processed_text = text
|
332 |
if options.get('clean_text', False) and not re.search(r'[\u4e00-\u9fff]', text):
|
@@ -338,13 +259,8 @@ class SentimentAnalyzer:
|
|
338 |
|
339 |
try:
|
340 |
# Tokenize and analyze
|
341 |
-
inputs = tokenizer(
|
342 |
-
|
343 |
-
return_tensors="pt",
|
344 |
-
padding=True,
|
345 |
-
truncation=True,
|
346 |
-
max_length=config.MAX_TEXT_LENGTH
|
347 |
-
).to(model_manager.device)
|
348 |
|
349 |
with torch.no_grad():
|
350 |
outputs = model(**inputs)
|
@@ -391,7 +307,7 @@ class SentimentAnalyzer:
|
|
391 |
|
392 |
except Exception as e:
|
393 |
logger.error(f"Analysis failed: {e}")
|
394 |
-
raise
|
395 |
|
396 |
@staticmethod
|
397 |
def analyze_batch(texts: List[str], language: str = 'auto',
|
@@ -404,461 +320,207 @@ class SentimentAnalyzer:
|
|
404 |
result['batch_index'] = i
|
405 |
results.append(result)
|
406 |
except Exception as e:
|
|
|
407 |
results.append({
|
408 |
'sentiment': 'Error',
|
409 |
'confidence': 0.0,
|
410 |
'error': str(e),
|
411 |
'batch_index': i,
|
412 |
-
'text': text
|
413 |
})
|
414 |
return results
|
415 |
|
416 |
-
class
|
417 |
-
"""
|
418 |
|
419 |
@staticmethod
|
420 |
-
def
|
421 |
-
"""Create
|
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 |
-
predict_fn = ExplainabilityAnalyzer.create_prediction_function(model, tokenizer, device)
|
454 |
-
|
455 |
-
# Test prediction function first
|
456 |
-
test_probs = predict_fn([text])
|
457 |
-
if len(test_probs) == 0:
|
458 |
-
return {'method': 'LIME', 'error': 'Prediction function failed'}
|
459 |
-
|
460 |
-
# Determine class names based on model output
|
461 |
-
num_classes = len(test_probs[0])
|
462 |
-
if num_classes == 3:
|
463 |
-
class_names = ['Negative', 'Neutral', 'Positive']
|
464 |
-
else:
|
465 |
-
class_names = ['Negative', 'Positive']
|
466 |
-
|
467 |
-
# Initialize LIME explainer
|
468 |
-
explainer = LimeTextExplainer(
|
469 |
-
class_names=class_names,
|
470 |
-
feature_selection='auto',
|
471 |
-
split_expression=r'\W+',
|
472 |
-
bow=False
|
473 |
-
)
|
474 |
-
|
475 |
-
# Generate explanation
|
476 |
-
explanation = explainer.explain_instance(
|
477 |
-
text,
|
478 |
-
predict_fn,
|
479 |
-
num_features=min(num_features, len(text.split())),
|
480 |
-
num_samples=50 # Reduced for faster processing
|
481 |
-
)
|
482 |
-
|
483 |
-
# Extract feature importance
|
484 |
-
feature_importance = explanation.as_list()
|
485 |
-
|
486 |
-
return {
|
487 |
-
'method': 'LIME',
|
488 |
-
'feature_importance': feature_importance,
|
489 |
-
'class_names': class_names,
|
490 |
-
'success': True
|
491 |
-
}
|
492 |
-
|
493 |
-
except Exception as e:
|
494 |
-
logger.error(f"LIME analysis failed: {e}")
|
495 |
-
return {'method': 'LIME', 'error': str(e)}
|
496 |
-
|
497 |
-
@staticmethod
|
498 |
-
def analyze_with_attention(text: str, model, tokenizer, device) -> Dict:
|
499 |
-
"""Analyze text with attention weights - simplified version"""
|
500 |
-
try:
|
501 |
-
# Tokenize input
|
502 |
-
inputs = tokenizer(text, return_tensors="pt", padding=True,
|
503 |
-
truncation=True, max_length=config.MAX_TEXT_LENGTH).to(device)
|
504 |
-
|
505 |
-
# Get tokens for display
|
506 |
-
tokens = tokenizer.convert_ids_to_tokens(inputs['input_ids'][0])
|
507 |
-
|
508 |
-
# Simple attention simulation based on input importance
|
509 |
-
try:
|
510 |
-
with torch.no_grad():
|
511 |
-
outputs = model(**inputs, output_attentions=True)
|
512 |
-
if hasattr(outputs, 'attentions') and outputs.attentions is not None:
|
513 |
-
attentions = outputs.attentions
|
514 |
-
# Average attention across layers and heads
|
515 |
-
avg_attention = torch.mean(torch.stack(attentions), dim=(0, 1, 2)).cpu().numpy()
|
516 |
-
else:
|
517 |
-
raise AttributeError("No attention outputs")
|
518 |
-
except:
|
519 |
-
# Fallback: simulate attention based on token position and type
|
520 |
-
avg_attention = np.random.uniform(0.1, 1.0, len(tokens))
|
521 |
-
# Give higher attention to non-special tokens
|
522 |
-
for i, token in enumerate(tokens):
|
523 |
-
if token in ['[CLS]', '[SEP]', '<s>', '</s>', '<pad>']:
|
524 |
-
avg_attention[i] *= 0.3
|
525 |
-
|
526 |
-
# Create attention weights for each token
|
527 |
-
attention_weights = []
|
528 |
-
for i, token in enumerate(tokens):
|
529 |
-
if i < len(avg_attention):
|
530 |
-
# Clean token for display
|
531 |
-
clean_token = token.replace('Ġ', '').replace('##', '')
|
532 |
-
if clean_token.strip():
|
533 |
-
attention_weights.append((clean_token, float(avg_attention[i])))
|
534 |
-
|
535 |
-
return {
|
536 |
-
'method': 'Attention',
|
537 |
-
'tokens': [t[0] for t in attention_weights],
|
538 |
-
'attention_weights': attention_weights,
|
539 |
-
'success': True
|
540 |
-
}
|
541 |
-
|
542 |
-
except Exception as e:
|
543 |
-
logger.error(f"Attention analysis failed: {e}")
|
544 |
-
return {'method': 'Attention', 'error': str(e)}
|
545 |
-
|
546 |
-
class AdvancedVisualizer:
|
547 |
-
"""Enhanced visualizations with Plotly - 修复了类名"""
|
548 |
|
549 |
@staticmethod
|
550 |
-
def
|
551 |
-
"""Create
|
552 |
-
|
553 |
-
fig = go.Figure()
|
554 |
-
fig.add_annotation(
|
555 |
-
text=f"LIME Error: {lime_result['error']}",
|
556 |
-
x=0.5, y=0.5,
|
557 |
-
xref="paper", yref="paper",
|
558 |
-
showarrow=False,
|
559 |
-
font=dict(size=14)
|
560 |
-
)
|
561 |
-
fig.update_layout(height=400, title="LIME Analysis Error")
|
562 |
-
return fig
|
563 |
-
|
564 |
-
if not lime_result.get('feature_importance'):
|
565 |
-
fig = go.Figure()
|
566 |
-
fig.add_annotation(
|
567 |
-
text="No LIME features available",
|
568 |
-
x=0.5, y=0.5,
|
569 |
-
xref="paper", yref="paper",
|
570 |
-
showarrow=False
|
571 |
-
)
|
572 |
-
fig.update_layout(height=400, title="No LIME Data")
|
573 |
-
return fig
|
574 |
|
575 |
-
|
576 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
577 |
|
578 |
fig = go.Figure(data=[
|
579 |
-
go.Bar(
|
580 |
-
y=features,
|
581 |
-
x=scores,
|
582 |
-
orientation='h',
|
583 |
-
marker_color=colors,
|
584 |
-
text=[f'{score:.3f}' for score in scores],
|
585 |
-
textposition='auto',
|
586 |
-
hovertemplate='<b>%{y}</b><br>Importance: %{x:.3f}<extra></extra>'
|
587 |
-
)
|
588 |
])
|
589 |
|
|
|
590 |
fig.update_layout(
|
591 |
-
title="
|
592 |
-
|
593 |
-
yaxis_title="Features",
|
594 |
height=400,
|
595 |
showlegend=False
|
596 |
)
|
597 |
|
598 |
return fig
|
599 |
-
|
600 |
@staticmethod
|
601 |
-
def
|
602 |
-
"""Create
|
603 |
-
|
604 |
-
fig = go.Figure()
|
605 |
-
fig.add_annotation(
|
606 |
-
text=f"Attention Error: {attention_result['error']}",
|
607 |
-
x=0.5, y=0.5,
|
608 |
-
xref="paper", yref="paper",
|
609 |
-
showarrow=False,
|
610 |
-
font=dict(size=14)
|
611 |
-
)
|
612 |
-
fig.update_layout(height=400, title="Attention Analysis Error")
|
613 |
-
return fig
|
614 |
-
|
615 |
-
if not attention_result.get('attention_weights'):
|
616 |
-
fig = go.Figure()
|
617 |
-
fig.add_annotation(
|
618 |
-
text="No attention weights available",
|
619 |
-
x=0.5, y=0.5,
|
620 |
-
xref="paper", yref="paper",
|
621 |
-
showarrow=False
|
622 |
-
)
|
623 |
-
fig.update_layout(height=400, title="No Attention Data")
|
624 |
-
return fig
|
625 |
|
626 |
-
|
|
|
|
|
627 |
|
628 |
-
#
|
629 |
-
|
630 |
-
|
631 |
-
|
632 |
-
|
633 |
-
|
|
|
|
|
634 |
|
635 |
-
|
636 |
-
|
637 |
-
|
638 |
-
|
639 |
-
tokens = [tokens[i] for i in top_indices]
|
640 |
-
normalized_weights = normalized_weights[top_indices]
|
641 |
-
weights = weights[top_indices]
|
642 |
|
643 |
-
fig
|
644 |
-
|
645 |
-
|
646 |
-
|
647 |
-
|
648 |
-
|
649 |
-
|
650 |
-
|
651 |
-
|
652 |
-
|
653 |
-
|
654 |
-
|
|
|
|
|
|
|
|
|
655 |
|
656 |
fig.update_layout(
|
657 |
-
title="
|
658 |
-
xaxis_title="
|
659 |
-
yaxis_title="
|
660 |
-
height=400
|
661 |
-
showlegend=False,
|
662 |
-
xaxis=dict(tickmode='array', tickvals=list(range(len(tokens))), ticktext=tokens, tickangle=45)
|
663 |
)
|
664 |
|
665 |
return fig
|
666 |
|
667 |
@staticmethod
|
668 |
-
def
|
669 |
-
"""Create
|
670 |
-
|
|
|
671 |
|
672 |
-
|
673 |
-
|
674 |
-
|
675 |
-
|
676 |
-
|
677 |
-
|
678 |
-
|
679 |
-
|
680 |
-
delta={'reference': 50},
|
681 |
-
gauge={
|
682 |
-
'axis': {'range': [None, 100]},
|
683 |
-
'bar': {'color': colors['pos'] if result['sentiment'] == 'Positive' else colors['neg']},
|
684 |
-
'steps': [
|
685 |
-
{'range': [0, 33], 'color': colors['neg']},
|
686 |
-
{'range': [33, 67], 'color': colors['neu']},
|
687 |
-
{'range': [67, 100], 'color': colors['pos']}
|
688 |
-
],
|
689 |
-
'threshold': {
|
690 |
-
'line': {'color': "red", 'width': 4},
|
691 |
-
'thickness': 0.75,
|
692 |
-
'value': 90
|
693 |
-
}
|
694 |
-
}
|
695 |
-
))
|
696 |
-
else:
|
697 |
-
# Two-way gauge
|
698 |
-
fig = go.Figure(go.Indicator(
|
699 |
-
mode="gauge+number",
|
700 |
-
value=result['confidence'] * 100,
|
701 |
-
domain={'x': [0, 1], 'y': [0, 1]},
|
702 |
-
title={'text': f"Confidence: {result['sentiment']}"},
|
703 |
-
gauge={
|
704 |
-
'axis': {'range': [None, 100]},
|
705 |
-
'bar': {'color': colors['pos'] if result['sentiment'] == 'Positive' else colors['neg']},
|
706 |
-
'steps': [
|
707 |
-
{'range': [0, 50], 'color': "lightgray"},
|
708 |
-
{'range': [50, 100], 'color': "gray"}
|
709 |
-
]
|
710 |
-
}
|
711 |
-
))
|
712 |
-
|
713 |
-
fig.update_layout(height=400, font={'size': 16})
|
714 |
-
return fig
|
715 |
-
|
716 |
-
except Exception as e:
|
717 |
-
logger.error(f"Failed to create gauge: {e}")
|
718 |
-
# 返回错误图表
|
719 |
-
fig = go.Figure()
|
720 |
-
fig.add_annotation(
|
721 |
-
text=f"Visualization Error: {str(e)}",
|
722 |
-
x=0.5, y=0.5,
|
723 |
-
xref="paper", yref="paper",
|
724 |
-
showarrow=False,
|
725 |
-
font=dict(size=14)
|
726 |
-
)
|
727 |
-
fig.update_layout(height=400)
|
728 |
-
return fig
|
729 |
-
|
730 |
-
@staticmethod
|
731 |
-
def create_probability_bars(result: Dict, theme: str = 'default') -> go.Figure:
|
732 |
-
"""Create probability bar chart"""
|
733 |
-
colors = config.THEMES.get(theme, config.THEMES['default'])
|
734 |
|
735 |
-
|
736 |
-
|
737 |
-
|
738 |
-
|
739 |
-
|
740 |
-
|
741 |
-
labels = ['Negative', 'Positive']
|
742 |
-
values = [result['neg_prob'], result['pos_prob']]
|
743 |
-
bar_colors = [colors['neg'], colors['pos']]
|
744 |
-
|
745 |
-
fig = go.Figure(data=[
|
746 |
-
go.Bar(x=labels, y=values, marker_color=bar_colors,
|
747 |
-
text=[f'{v:.3f}' for v in values])
|
748 |
-
])
|
749 |
-
|
750 |
-
fig.update_traces(texttemplate='%{text}', textposition='outside')
|
751 |
-
fig.update_layout(
|
752 |
-
title="Sentiment Probabilities",
|
753 |
-
yaxis_title="Probability",
|
754 |
-
height=400,
|
755 |
-
showlegend=False
|
756 |
-
)
|
757 |
-
|
758 |
-
return fig
|
759 |
-
|
760 |
-
except Exception as e:
|
761 |
-
logger.error(f"Failed to create bars: {e}")
|
762 |
-
fig = go.Figure()
|
763 |
-
fig.add_annotation(
|
764 |
-
text=f"Visualization Error: {str(e)}",
|
765 |
-
x=0.5, y=0.5,
|
766 |
-
xref="paper", yref="paper",
|
767 |
-
showarrow=False
|
768 |
-
)
|
769 |
-
fig.update_layout(height=400)
|
770 |
-
return fig
|
771 |
-
|
772 |
-
@staticmethod
|
773 |
-
def create_batch_summary(results: List[Dict], theme: str = 'default') -> go.Figure:
|
774 |
-
"""Create batch analysis summary"""
|
775 |
-
colors = config.THEMES.get(theme, config.THEMES['default'])
|
776 |
|
777 |
-
|
778 |
-
|
779 |
-
|
780 |
-
|
781 |
-
|
782 |
-
|
783 |
-
|
784 |
-
|
785 |
-
|
786 |
-
|
787 |
-
|
788 |
-
|
789 |
-
|
790 |
-
|
791 |
-
|
792 |
-
|
793 |
-
|
794 |
-
|
795 |
-
|
796 |
-
|
797 |
-
|
798 |
-
|
799 |
-
|
800 |
-
|
801 |
-
|
802 |
-
|
803 |
-
|
804 |
-
|
805 |
-
|
806 |
-
|
807 |
-
|
808 |
-
|
809 |
-
|
810 |
-
fig.update_layout(height=400)
|
811 |
-
return fig
|
812 |
-
|
813 |
-
@staticmethod
|
814 |
-
def create_confidence_distribution(results: List[Dict]) -> go.Figure:
|
815 |
-
"""Create confidence distribution plot"""
|
816 |
-
try:
|
817 |
-
confidences = [r['confidence'] for r in results
|
818 |
-
if 'confidence' in r and r['sentiment'] != 'Error']
|
819 |
-
|
820 |
-
if not confidences:
|
821 |
-
fig = go.Figure()
|
822 |
-
fig.add_annotation(text="No confidence data to display", x=0.5, y=0.5,
|
823 |
-
xref="paper", yref="paper", showarrow=False)
|
824 |
-
fig.update_layout(height=400, title="Confidence Distribution")
|
825 |
-
return fig
|
826 |
-
|
827 |
-
fig = go.Figure(data=[go.Histogram(
|
828 |
-
x=confidences,
|
829 |
-
nbinsx=20,
|
830 |
-
marker_color='skyblue',
|
831 |
-
opacity=0.7
|
832 |
-
)])
|
833 |
-
|
834 |
-
fig.update_layout(
|
835 |
-
title="Confidence Distribution",
|
836 |
-
xaxis_title="Confidence Score",
|
837 |
-
yaxis_title="Frequency",
|
838 |
-
height=400
|
839 |
-
)
|
840 |
-
|
841 |
-
return fig
|
842 |
-
|
843 |
-
except Exception as e:
|
844 |
-
logger.error(f"Failed to create confidence distribution: {e}")
|
845 |
-
fig = go.Figure()
|
846 |
-
fig.add_annotation(text=f"Error: {str(e)}", x=0.5, y=0.5,
|
847 |
-
xref="paper", yref="paper", showarrow=False)
|
848 |
-
fig.update_layout(height=400)
|
849 |
-
return fig
|
850 |
|
851 |
-
# Main application functions
|
852 |
def analyze_single_text(text: str, language: str, theme: str, clean_text: bool,
|
853 |
remove_punct: bool, remove_nums: bool):
|
854 |
-
"""Enhanced single text analysis
|
855 |
try:
|
856 |
if not text.strip():
|
857 |
-
return "
|
858 |
-
|
859 |
-
# 初始化检查
|
860 |
-
if not initialize_models():
|
861 |
-
return "❌ Failed to load sentiment analysis models. Please check your internet connection and try again.", None, None
|
862 |
|
863 |
# Map display names back to language codes
|
864 |
language_map = {
|
@@ -878,7 +540,6 @@ def analyze_single_text(text: str, language: str, theme: str, clean_text: bool,
|
|
878 |
'remove_numbers': remove_nums
|
879 |
}
|
880 |
|
881 |
-
# 分析文本
|
882 |
result = SentimentAnalyzer.analyze_text(text, language_code, preprocessing_options)
|
883 |
|
884 |
# Add to history
|
@@ -897,49 +558,39 @@ def analyze_single_text(text: str, language: str, theme: str, clean_text: bool,
|
|
897 |
history_manager.add_entry(history_entry)
|
898 |
|
899 |
# Create visualizations
|
900 |
-
gauge_fig =
|
901 |
-
bars_fig =
|
902 |
|
903 |
# Create info text
|
904 |
info_text = f"""
|
905 |
-
|
906 |
-
- **Sentiment:** {result['sentiment']} (
|
907 |
- **Language:** {result['language'].upper()}
|
908 |
-
- **Keywords:** {', '.join(result['keywords'])
|
909 |
-
- **
|
910 |
-
|
911 |
-
📊 **Probability Scores:**
|
912 |
-
- Positive: {result['pos_prob']:.3f}
|
913 |
-
- Negative: {result['neg_prob']:.3f}
|
914 |
-
- Neutral: {result.get('neu_prob', 0):.3f}
|
915 |
"""
|
916 |
|
917 |
return info_text, gauge_fig, bars_fig
|
918 |
|
919 |
except Exception as e:
|
920 |
-
logger.error(f"
|
921 |
-
|
922 |
-
return error_msg, None, None
|
923 |
|
924 |
def analyze_batch_texts(batch_text: str, language: str, theme: str,
|
925 |
clean_text: bool, remove_punct: bool, remove_nums: bool):
|
926 |
-
"""Batch text analysis
|
927 |
try:
|
928 |
if not batch_text.strip():
|
929 |
-
return "
|
930 |
-
|
931 |
-
# 初始化检查
|
932 |
-
if not initialize_models():
|
933 |
-
return "❌ Failed to load sentiment analysis models", None, None, None
|
934 |
|
935 |
# Parse batch input
|
936 |
texts = TextProcessor.parse_batch_input(batch_text)
|
937 |
|
938 |
if len(texts) > config.BATCH_SIZE_LIMIT:
|
939 |
-
return f"
|
940 |
|
941 |
if not texts:
|
942 |
-
return "
|
943 |
|
944 |
# Map display names back to language codes
|
945 |
language_map = {
|
@@ -984,8 +635,8 @@ def analyze_batch_texts(batch_text: str, language: str, theme: str,
|
|
984 |
history_manager.add_batch_entries(batch_entries)
|
985 |
|
986 |
# Create visualizations
|
987 |
-
summary_fig =
|
988 |
-
confidence_fig =
|
989 |
|
990 |
# Create results table
|
991 |
df_data = []
|
@@ -995,9 +646,9 @@ def analyze_batch_texts(batch_text: str, language: str, theme: str,
|
|
995 |
'Index': i+1,
|
996 |
'Text': text[:50] + '...' if len(text) > 50 else text,
|
997 |
'Sentiment': 'Error',
|
998 |
-
'Confidence':
|
999 |
'Language': 'Unknown',
|
1000 |
-
'Error': result
|
1001 |
})
|
1002 |
else:
|
1003 |
df_data.append({
|
@@ -1006,7 +657,7 @@ def analyze_batch_texts(batch_text: str, language: str, theme: str,
|
|
1006 |
'Sentiment': result['sentiment'],
|
1007 |
'Confidence': f"{result['confidence']:.3f}",
|
1008 |
'Language': result['language'].upper(),
|
1009 |
-
'Keywords': ', '.join(result
|
1010 |
})
|
1011 |
|
1012 |
df = pd.DataFrame(df_data)
|
@@ -1020,235 +671,214 @@ def analyze_batch_texts(batch_text: str, language: str, theme: str,
|
|
1020 |
avg_confidence = np.mean([r['confidence'] for r in successful_results])
|
1021 |
|
1022 |
summary_text = f"""
|
1023 |
-
|
1024 |
- **Total Texts:** {len(texts)}
|
1025 |
- **Successful:** {len(successful_results)}
|
1026 |
- **Errors:** {error_count}
|
1027 |
- **Average Confidence:** {avg_confidence:.3f}
|
1028 |
-
- **
|
1029 |
-
- Positive: {sentiment_counts.get('Positive', 0)}
|
1030 |
-
- Negative: {sentiment_counts.get('Negative', 0)}
|
1031 |
-
- Neutral: {sentiment_counts.get('Neutral', 0)}
|
1032 |
"""
|
1033 |
else:
|
1034 |
-
summary_text = f"
|
1035 |
|
1036 |
return summary_text, df, summary_fig, confidence_fig
|
1037 |
|
1038 |
except Exception as e:
|
1039 |
logger.error(f"Batch analysis failed: {e}")
|
1040 |
-
return f"
|
1041 |
-
|
1042 |
-
def get_history_stats():
|
1043 |
|
1044 |
-
|
1045 |
-
|
1046 |
-
|
1047 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1048 |
"""
|
1049 |
|
1050 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
1051 |
|
1052 |
except Exception as e:
|
1053 |
logger.error(f"Advanced analysis failed: {e}")
|
1054 |
-
|
1055 |
-
|
1056 |
-
|
1057 |
-
xref="paper", yref="paper", showarrow=False)
|
1058 |
-
empty_fig.update_layout(height=400)
|
1059 |
-
|
1060 |
-
return f"❌ Error: {str(e)}", empty_fig, empty_fig, empty_fig, empty_fig
|
1061 |
"""Get enhanced history statistics"""
|
1062 |
-
|
1063 |
-
|
1064 |
-
|
1065 |
-
|
1066 |
-
|
1067 |
-
|
1068 |
-
📊 **Comprehensive History Statistics:**
|
1069 |
|
1070 |
-
|
1071 |
- Total Analyses: {stats['total_analyses']}
|
1072 |
-
-
|
1073 |
-
-
|
1074 |
-
-
|
1075 |
|
1076 |
-
|
1077 |
- Average Confidence: {stats['avg_confidence']:.3f}
|
1078 |
- Highest Confidence: {stats['max_confidence']:.3f}
|
1079 |
- Lowest Confidence: {stats['min_confidence']:.3f}
|
1080 |
|
1081 |
-
|
1082 |
- Languages Detected: {stats['languages_detected']}
|
1083 |
- Most Common Language: {stats['most_common_language'].upper()}
|
1084 |
|
1085 |
-
|
1086 |
- Average Text Length: {stats['avg_text_length']:.1f} characters
|
1087 |
-
|
1088 |
-
except Exception as e:
|
1089 |
-
logger.error(f"Failed to get history stats: {e}")
|
1090 |
-
return f"❌ Error getting statistics: {str(e)}"
|
1091 |
|
1092 |
def filter_history_display(sentiment_filter: str, language_filter: str, min_confidence: float):
|
1093 |
"""Display filtered history"""
|
1094 |
-
|
1095 |
-
|
1096 |
-
|
1097 |
-
|
1098 |
-
|
1099 |
-
|
1100 |
-
|
1101 |
-
|
1102 |
-
|
1103 |
-
|
1104 |
-
|
1105 |
-
|
1106 |
-
|
1107 |
-
|
1108 |
-
|
1109 |
-
|
1110 |
-
|
1111 |
-
|
1112 |
-
|
1113 |
-
|
1114 |
-
|
1115 |
-
|
1116 |
-
|
1117 |
-
|
1118 |
-
|
1119 |
-
|
1120 |
-
|
1121 |
-
|
1122 |
-
|
1123 |
-
🔍 **Filtered Results:**
|
1124 |
- Found {len(filtered_history)} entries matching criteria
|
1125 |
- Showing most recent {min(20, len(filtered_history))} entries
|
1126 |
-
|
1127 |
-
|
1128 |
-
|
1129 |
-
|
1130 |
-
except Exception as e:
|
1131 |
-
logger.error(f"Failed to filter history: {e}")
|
1132 |
-
return f"❌ Error filtering history: {str(e)}", None
|
1133 |
|
1134 |
def plot_history_dashboard():
|
1135 |
-
"""Create history dashboard
|
1136 |
-
|
1137 |
-
|
1138 |
-
|
1139 |
-
|
1140 |
-
|
1141 |
-
|
1142 |
-
fig = make_subplots(
|
1143 |
-
rows=2, cols=2,
|
1144 |
-
subplot_titles=['Sentiment Timeline', 'Confidence Distribution',
|
1145 |
-
'Language Distribution', 'Sentiment Summary'],
|
1146 |
-
specs=[[{"secondary_y": False}, {"secondary_y": False}],
|
1147 |
-
[{"type": "pie"}, {"type": "bar"}]]
|
1148 |
-
)
|
1149 |
-
|
1150 |
-
# Extract data
|
1151 |
-
indices = list(range(len(history)))
|
1152 |
-
pos_probs = [item.get('pos_prob', 0) for item in history]
|
1153 |
-
confidences = [item['confidence'] for item in history]
|
1154 |
-
sentiments = [item['sentiment'] for item in history]
|
1155 |
-
languages = [item.get('language', 'en') for item in history]
|
1156 |
-
|
1157 |
-
# Sentiment timeline
|
1158 |
-
colors = ['#4CAF50' if s == 'Positive' else '#F44336' if s == 'Negative' else '#FF9800'
|
1159 |
-
for s in sentiments]
|
1160 |
-
fig.add_trace(
|
1161 |
-
go.Scatter(x=indices, y=pos_probs, mode='lines+markers',
|
1162 |
-
marker=dict(color=colors, size=6),
|
1163 |
-
name='Positive Probability'),
|
1164 |
-
row=1, col=1
|
1165 |
-
)
|
1166 |
-
|
1167 |
-
# Confidence distribution
|
1168 |
-
fig.add_trace(
|
1169 |
-
go.Histogram(x=confidences, nbinsx=10, name='Confidence'),
|
1170 |
-
row=1, col=2
|
1171 |
-
)
|
1172 |
-
|
1173 |
-
# Language distribution
|
1174 |
-
lang_counts = Counter(languages)
|
1175 |
-
fig.add_trace(
|
1176 |
-
go.Pie(labels=list(lang_counts.keys()), values=list(lang_counts.values()),
|
1177 |
-
name="Languages"),
|
1178 |
-
row=2, col=1
|
1179 |
-
)
|
1180 |
-
|
1181 |
-
# Sentiment summary
|
1182 |
-
sent_counts = Counter(sentiments)
|
1183 |
-
colors_dict = {'Positive': '#4CAF50', 'Negative': '#F44336', 'Neutral': '#FF9800'}
|
1184 |
-
fig.add_trace(
|
1185 |
-
go.Bar(x=list(sent_counts.keys()), y=list(sent_counts.values()),
|
1186 |
-
marker_color=[colors_dict.get(k, '#999999') for k in sent_counts.keys()]),
|
1187 |
-
row=2, col=2
|
1188 |
-
)
|
1189 |
-
|
1190 |
-
fig.update_layout(height=800, showlegend=False, title_text="Analysis Dashboard")
|
1191 |
-
return fig, f"📊 Dashboard showing {len(history)} analyses"
|
1192 |
-
|
1193 |
-
except Exception as e:
|
1194 |
-
logger.error(f"Failed to create dashboard: {e}")
|
1195 |
-
return None, f"❌ Error creating dashboard: {str(e)}"
|
1196 |
|
1197 |
def export_history_csv():
|
1198 |
"""Export history to CSV"""
|
|
|
|
|
|
|
|
|
1199 |
try:
|
1200 |
-
history = history_manager.get_history()
|
1201 |
-
if not history:
|
1202 |
-
return None, "📊 No history to export"
|
1203 |
-
|
1204 |
df = pd.DataFrame(history)
|
1205 |
temp_file = tempfile.NamedTemporaryFile(delete=False, suffix='.csv', mode='w')
|
1206 |
df.to_csv(temp_file.name, index=False)
|
1207 |
-
return temp_file.name, f"
|
1208 |
except Exception as e:
|
1209 |
-
|
1210 |
-
return None, f"❌ Export failed: {str(e)}"
|
1211 |
|
1212 |
def export_history_excel():
|
1213 |
"""Export history to Excel"""
|
|
|
|
|
|
|
|
|
1214 |
try:
|
1215 |
-
history = history_manager.get_history()
|
1216 |
-
if not history:
|
1217 |
-
return None, "📊 No history to export"
|
1218 |
-
|
1219 |
df = pd.DataFrame(history)
|
1220 |
temp_file = tempfile.NamedTemporaryFile(delete=False, suffix='.xlsx')
|
1221 |
df.to_excel(temp_file.name, index=False)
|
1222 |
-
return temp_file.name, f"
|
1223 |
except Exception as e:
|
1224 |
-
|
1225 |
-
return None, f"❌ Export failed: {str(e)}"
|
1226 |
|
1227 |
def clear_all_history():
|
1228 |
"""Clear analysis history"""
|
1229 |
-
|
1230 |
-
|
1231 |
-
return f"🗑️ Cleared {count} entries from history"
|
1232 |
-
except Exception as e:
|
1233 |
-
logger.error(f"Failed to clear history: {e}")
|
1234 |
-
return f"❌ Error clearing history: {str(e)}"
|
1235 |
|
1236 |
def get_recent_analyses():
|
1237 |
"""Get recent analysis summary"""
|
1238 |
-
|
1239 |
-
|
1240 |
-
|
1241 |
-
|
1242 |
-
|
1243 |
-
|
1244 |
-
|
1245 |
-
|
1246 |
-
|
1247 |
-
|
1248 |
-
return summary_text
|
1249 |
-
except Exception as e:
|
1250 |
-
logger.error(f"Failed to get recent analyses: {e}")
|
1251 |
-
return f"❌ Error getting recent analyses: {str(e)}"
|
1252 |
|
1253 |
# Sample data
|
1254 |
SAMPLE_TEXTS = [
|
@@ -1280,10 +910,10 @@ Not sure if I like it or not.
|
|
1280 |
Amazing quality and fast delivery!
|
1281 |
Could be better, but it's okay."""
|
1282 |
|
1283 |
-
# Gradio Interface
|
1284 |
-
with gr.Blocks(theme=gr.themes.Soft(), title="Multilingual Sentiment Analyzer") as demo:
|
1285 |
-
gr.Markdown("# 🎭 Multilingual Sentiment Analyzer")
|
1286 |
-
gr.Markdown("Comprehensive sentiment analysis with batch processing and multilingual support")
|
1287 |
|
1288 |
with gr.Tab("📝 Single Analysis"):
|
1289 |
with gr.Row():
|
@@ -1320,140 +950,106 @@ with gr.Blocks(theme=gr.themes.Soft(), title="Multilingual Sentiment Analyzer")
|
|
1320 |
)
|
1321 |
|
1322 |
with gr.Column(scale=1):
|
1323 |
-
result_info = gr.Markdown("Enter text and click Analyze
|
1324 |
|
1325 |
with gr.Row():
|
1326 |
gauge_plot = gr.Plot(label="Sentiment Gauge")
|
1327 |
bars_plot = gr.Plot(label="Probability Distribution")
|
1328 |
-
|
1329 |
-
with gr.Tab("
|
1330 |
with gr.Row():
|
1331 |
with gr.Column(scale=2):
|
1332 |
-
|
1333 |
-
label="Text
|
1334 |
-
placeholder="Enter
|
1335 |
-
lines=
|
1336 |
)
|
1337 |
|
1338 |
with gr.Row():
|
1339 |
-
|
1340 |
choices=['Auto Detect', 'English', 'Chinese', 'Spanish', 'French', 'German', 'Swedish'],
|
1341 |
value='Auto Detect',
|
1342 |
label="Language"
|
1343 |
)
|
1344 |
-
|
1345 |
choices=list(config.THEMES.keys()),
|
1346 |
value='default',
|
1347 |
label="Theme"
|
1348 |
)
|
1349 |
|
1350 |
-
gr.Markdown("### 🔍 Explainability Options")
|
1351 |
-
gr.Markdown("**LIME** shows which words influence sentiment most. **Attention** shows which tokens the model focuses on.")
|
1352 |
-
|
1353 |
with gr.Row():
|
1354 |
-
|
1355 |
-
|
1356 |
-
|
1357 |
-
info="Explains feature importance (requires: pip install lime)"
|
1358 |
-
)
|
1359 |
-
use_attention = gr.Checkbox(
|
1360 |
-
label="👁️ Use Attention Weights",
|
1361 |
-
value=True,
|
1362 |
-
info="Shows token-level attention patterns"
|
1363 |
-
)
|
1364 |
-
|
1365 |
-
lime_features = gr.Slider(
|
1366 |
-
minimum=5,
|
1367 |
-
maximum=20,
|
1368 |
-
value=10,
|
1369 |
-
step=1,
|
1370 |
-
label="LIME Features Count",
|
1371 |
-
info="Number of top features to analyze"
|
1372 |
-
)
|
1373 |
|
1374 |
-
|
1375 |
|
1376 |
gr.Examples(
|
1377 |
-
examples=[
|
1378 |
-
|
1379 |
-
|
1380 |
-
["The service was terrible and the staff was very rude. I will never come back here again."]
|
1381 |
-
],
|
1382 |
-
inputs=advanced_input,
|
1383 |
-
label="Sample Texts for Advanced Analysis"
|
1384 |
)
|
1385 |
|
1386 |
with gr.Column(scale=1):
|
1387 |
-
|
1388 |
-
**Advanced Analysis Features:**
|
1389 |
-
|
1390 |
-
🔍 **LIME (Local Interpretable Model-agnostic Explanations)**
|
1391 |
-
- Shows which words contribute most to the sentiment prediction
|
1392 |
-
- Red bars = pushes toward negative sentiment
|
1393 |
-
- Green bars = pushes toward positive sentiment
|
1394 |
-
|
1395 |
-
👁️ **Attention Weights**
|
1396 |
-
- Visualizes which tokens the model pays attention to
|
1397 |
-
- Darker/higher bars = more attention from the model
|
1398 |
-
- Helps understand model focus patterns
|
1399 |
-
|
1400 |
-
Configure explainability settings and click **Advanced Analyze** to start.
|
1401 |
-
""")
|
1402 |
|
1403 |
with gr.Row():
|
1404 |
-
|
1405 |
-
|
|
|
|
|
1406 |
|
1407 |
with gr.Row():
|
1408 |
-
|
1409 |
-
|
1410 |
-
|
1411 |
-
with gr.Tab("
|
1412 |
with gr.Row():
|
1413 |
with gr.Column(scale=2):
|
1414 |
-
|
1415 |
-
label="
|
1416 |
-
placeholder="Enter
|
1417 |
-
lines=
|
1418 |
)
|
1419 |
|
1420 |
with gr.Row():
|
1421 |
-
|
1422 |
choices=['Auto Detect', 'English', 'Chinese', 'Spanish', 'French', 'German', 'Swedish'],
|
1423 |
value='Auto Detect',
|
1424 |
label="Language"
|
1425 |
)
|
1426 |
-
|
1427 |
choices=list(config.THEMES.keys()),
|
1428 |
value='default',
|
1429 |
label="Theme"
|
1430 |
)
|
1431 |
|
1432 |
with gr.Row():
|
1433 |
-
|
1434 |
-
|
1435 |
-
|
1436 |
-
|
1437 |
-
|
|
|
|
|
|
|
1438 |
|
1439 |
-
gr.
|
1440 |
-
|
1441 |
-
|
1442 |
-
|
|
|
|
|
1443 |
)
|
|
|
|
|
1444 |
|
1445 |
with gr.Column(scale=1):
|
1446 |
-
|
1447 |
|
1448 |
with gr.Row():
|
1449 |
-
|
1450 |
-
|
1451 |
-
interactive=False
|
1452 |
-
)
|
1453 |
-
|
1454 |
-
with gr.Row():
|
1455 |
-
batch_summary_plot = gr.Plot(label="Sentiment Summary")
|
1456 |
-
batch_confidence_plot = gr.Plot(label="Confidence Distribution")
|
1457 |
|
1458 |
with gr.Tab("📈 History & Analytics"):
|
1459 |
with gr.Row():
|
@@ -1514,13 +1110,6 @@ Configure explainability settings and click **Advanced Analyze** to start.
|
|
1514 |
outputs=[result_info, gauge_plot, bars_plot]
|
1515 |
)
|
1516 |
|
1517 |
-
# Advanced Analysis
|
1518 |
-
advanced_analyze_btn.click(
|
1519 |
-
analyze_advanced_text,
|
1520 |
-
inputs=[advanced_input, advanced_language, advanced_theme, use_lime, use_attention, lime_features],
|
1521 |
-
outputs=[advanced_result_info, advanced_gauge_plot, advanced_bars_plot, lime_plot, attention_plot]
|
1522 |
-
)
|
1523 |
-
|
1524 |
# Batch Analysis
|
1525 |
batch_analyze_btn.click(
|
1526 |
analyze_batch_texts,
|
@@ -1528,6 +1117,13 @@ Configure explainability settings and click **Advanced Analyze** to start.
|
|
1528 |
outputs=[batch_summary, batch_results_table, batch_summary_plot, batch_confidence_plot]
|
1529 |
)
|
1530 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1531 |
# History & Analytics
|
1532 |
stats_btn.click(
|
1533 |
get_history_stats,
|
@@ -1565,17 +1161,5 @@ Configure explainability settings and click **Advanced Analyze** to start.
|
|
1565 |
outputs=history_status
|
1566 |
)
|
1567 |
|
1568 |
-
# 启动应用
|
1569 |
if __name__ == "__main__":
|
1570 |
-
|
1571 |
-
logger.info("Starting Multilingual Sentiment Analyzer...")
|
1572 |
-
demo.launch(
|
1573 |
-
share=True,
|
1574 |
-
server_name="0.0.0.0",
|
1575 |
-
server_port=7860,
|
1576 |
-
show_error=True
|
1577 |
-
)
|
1578 |
-
except Exception as e:
|
1579 |
-
logger.error(f"Failed to launch application: {e}")
|
1580 |
-
print(f"❌ Application failed to start: {e}")
|
1581 |
-
print("Please check your dependencies and try again.")
|
|
|
5 |
import plotly.express as px
|
6 |
from plotly.subplots import make_subplots
|
7 |
import numpy as np
|
8 |
+
from wordcloud import WordCloud
|
9 |
from collections import Counter, defaultdict
|
10 |
import re
|
11 |
import json
|
|
|
17 |
from functools import lru_cache
|
18 |
from dataclasses import dataclass
|
19 |
from typing import List, Dict, Optional, Tuple
|
20 |
+
import nltk
|
21 |
+
from nltk.corpus import stopwords
|
22 |
+
import langdetect
|
23 |
import pandas as pd
|
24 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
25 |
# Configuration
|
26 |
@dataclass
|
27 |
class Config:
|
|
|
41 |
'sv': 'Swedish'
|
42 |
}
|
43 |
|
|
|
44 |
MODELS = {
|
45 |
'en': "cardiffnlp/twitter-roberta-base-sentiment-latest",
|
46 |
'multilingual': "cardiffnlp/twitter-xlm-roberta-base-sentiment",
|
47 |
+
'zh': "uer/roberta-base-finetuned-dianping-chinese"
|
|
|
|
|
48 |
}
|
49 |
|
50 |
# Color themes
|
|
|
57 |
|
58 |
config = Config()
|
59 |
|
60 |
+
# Logging setup
|
61 |
+
logging.basicConfig(level=logging.INFO)
|
62 |
+
logger = logging.getLogger(__name__)
|
63 |
+
|
64 |
+
# Initialize NLTK
|
65 |
+
try:
|
66 |
+
nltk.download('stopwords', quiet=True)
|
67 |
+
nltk.download('punkt', quiet=True)
|
68 |
+
STOP_WORDS = set(stopwords.words('english'))
|
69 |
+
except:
|
70 |
+
STOP_WORDS = {'the', 'a', 'an', 'and', 'or', 'but', 'in', 'on', 'at', 'to', 'for', 'of', 'with', 'by'}
|
71 |
+
|
72 |
class ModelManager:
|
73 |
+
"""Manages multiple language models"""
|
74 |
def __init__(self):
|
75 |
self.models = {}
|
76 |
self.tokenizers = {}
|
77 |
+
self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
78 |
+
self._load_default_model()
|
|
|
79 |
|
80 |
+
def _load_default_model(self):
|
81 |
+
"""Load the default models"""
|
82 |
try:
|
83 |
+
# Load multilingual model as default
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
84 |
model_name = config.MODELS['multilingual']
|
|
|
|
|
85 |
self.tokenizers['default'] = AutoTokenizer.from_pretrained(model_name)
|
86 |
self.models['default'] = AutoModelForSequenceClassification.from_pretrained(model_name)
|
87 |
self.models['default'].to(self.device)
|
88 |
+
logger.info(f"Default model loaded: {model_name}")
|
89 |
|
90 |
+
# Load Chinese model
|
91 |
+
zh_model_name = config.MODELS['zh']
|
92 |
+
self.tokenizers['zh'] = AutoTokenizer.from_pretrained(zh_model_name)
|
93 |
+
self.models['zh'] = AutoModelForSequenceClassification.from_pretrained(zh_model_name)
|
94 |
+
self.models['zh'].to(self.device)
|
95 |
+
logger.info(f"Chinese model loaded: {zh_model_name}")
|
96 |
|
97 |
except Exception as e:
|
98 |
+
logger.error(f"Failed to load models: {e}")
|
99 |
+
raise
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
100 |
|
101 |
def get_model(self, language='en'):
|
102 |
"""Get model for specific language"""
|
103 |
+
if language == 'zh':
|
104 |
+
return self.models['zh'], self.tokenizers['zh']
|
105 |
+
elif language in ['en', 'auto'] or language not in config.SUPPORTED_LANGUAGES:
|
106 |
+
return self.models['default'], self.tokenizers['default']
|
107 |
+
return self.models['default'], self.tokenizers['default'] # Use multilingual for other languages
|
108 |
|
109 |
@staticmethod
|
110 |
def detect_language(text: str) -> str:
|
111 |
+
"""Detect text language properly"""
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
112 |
try:
|
113 |
+
# Use langdetect for all languages
|
114 |
detected = langdetect.detect(text)
|
115 |
+
# Map some common langdetect codes to our supported languages
|
116 |
language_mapping = {
|
117 |
'zh-cn': 'zh',
|
118 |
'zh-tw': 'zh'
|
|
|
122 |
except:
|
123 |
return 'en'
|
124 |
|
125 |
+
model_manager = ModelManager()
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
126 |
|
127 |
class HistoryManager:
|
128 |
+
"""Enhanced history manager with more features"""
|
129 |
def __init__(self):
|
130 |
self._history = []
|
131 |
|
|
|
135 |
self._history = self._history[-config.MAX_HISTORY_SIZE:]
|
136 |
|
137 |
def add_batch_entries(self, entries: List[Dict]):
|
138 |
+
"""Add multiple entries at once"""
|
139 |
for entry in entries:
|
140 |
self.add_entry(entry)
|
141 |
|
|
|
143 |
return self._history.copy()
|
144 |
|
145 |
def get_recent_history(self, n: int = 10) -> List[Dict]:
|
146 |
+
"""Get n most recent entries"""
|
147 |
return self._history[-n:] if self._history else []
|
148 |
|
149 |
def filter_history(self, sentiment: str = None, language: str = None,
|
150 |
min_confidence: float = None) -> List[Dict]:
|
151 |
+
"""Filter history by criteria"""
|
152 |
filtered = self._history
|
153 |
|
154 |
if sentiment:
|
|
|
178 |
'positive_count': sentiments.count('Positive'),
|
179 |
'negative_count': sentiments.count('Negative'),
|
180 |
'neutral_count': sentiments.count('Neutral'),
|
181 |
+
'avg_confidence': np.mean(confidences),
|
182 |
+
'max_confidence': np.max(confidences),
|
183 |
+
'min_confidence': np.min(confidences),
|
184 |
'languages_detected': len(set(languages)),
|
185 |
'most_common_language': Counter(languages).most_common(1)[0][0] if languages else 'en',
|
186 |
+
'avg_text_length': np.mean([len(item.get('full_text', '')) for item in self._history])
|
187 |
}
|
188 |
|
189 |
history_manager = HistoryManager()
|
|
|
210 |
@staticmethod
|
211 |
def extract_keywords(text: str, top_k: int = 5) -> List[str]:
|
212 |
"""Extract key words from text"""
|
213 |
+
# For Chinese text, extract characters
|
214 |
if re.search(r'[\u4e00-\u9fff]', text):
|
215 |
words = re.findall(r'[\u4e00-\u9fff]+', text)
|
216 |
all_chars = ''.join(words)
|
217 |
char_freq = Counter(all_chars)
|
218 |
return [char for char, _ in char_freq.most_common(top_k)]
|
219 |
else:
|
220 |
+
# For other languages, use word-based extraction
|
221 |
cleaned = TextProcessor.clean_text(text)
|
222 |
words = cleaned.split()
|
223 |
word_freq = Counter(words)
|
|
|
230 |
return [line.strip() for line in lines if line.strip()]
|
231 |
|
232 |
class SentimentAnalyzer:
|
233 |
+
"""Enhanced sentiment analysis"""
|
234 |
|
235 |
@staticmethod
|
236 |
def analyze_text(text: str, language: str = 'auto', preprocessing_options: Dict = None) -> Dict:
|
|
|
238 |
if not text.strip():
|
239 |
raise ValueError("Empty text provided")
|
240 |
|
|
|
|
|
|
|
|
|
241 |
# Detect language if auto
|
242 |
if language == 'auto':
|
243 |
detected_lang = model_manager.detect_language(text)
|
|
|
245 |
detected_lang = language
|
246 |
|
247 |
# Get appropriate model
|
248 |
+
model, tokenizer = model_manager.get_model(detected_lang)
|
|
|
|
|
|
|
|
|
249 |
|
250 |
+
# Preprocessing options - don't clean Chinese text
|
251 |
options = preprocessing_options or {}
|
252 |
processed_text = text
|
253 |
if options.get('clean_text', False) and not re.search(r'[\u4e00-\u9fff]', text):
|
|
|
259 |
|
260 |
try:
|
261 |
# Tokenize and analyze
|
262 |
+
inputs = tokenizer(processed_text, return_tensors="pt", padding=True,
|
263 |
+
truncation=True, max_length=config.MAX_TEXT_LENGTH).to(model_manager.device)
|
|
|
|
|
|
|
|
|
|
|
264 |
|
265 |
with torch.no_grad():
|
266 |
outputs = model(**inputs)
|
|
|
307 |
|
308 |
except Exception as e:
|
309 |
logger.error(f"Analysis failed: {e}")
|
310 |
+
raise
|
311 |
|
312 |
@staticmethod
|
313 |
def analyze_batch(texts: List[str], language: str = 'auto',
|
|
|
320 |
result['batch_index'] = i
|
321 |
results.append(result)
|
322 |
except Exception as e:
|
323 |
+
# Add error result
|
324 |
results.append({
|
325 |
'sentiment': 'Error',
|
326 |
'confidence': 0.0,
|
327 |
'error': str(e),
|
328 |
'batch_index': i,
|
329 |
+
'text': text
|
330 |
})
|
331 |
return results
|
332 |
|
333 |
+
class PlotlyVisualizer:
|
334 |
+
"""Enhanced visualizations with Plotly"""
|
335 |
|
336 |
@staticmethod
|
337 |
+
def create_sentiment_gauge(result: Dict, theme: str = 'default') -> go.Figure:
|
338 |
+
"""Create an animated sentiment gauge"""
|
339 |
+
colors = config.THEMES[theme]
|
340 |
+
|
341 |
+
if result['has_neutral']:
|
342 |
+
# Three-way gauge
|
343 |
+
fig = go.Figure(go.Indicator(
|
344 |
+
mode = "gauge+number+delta",
|
345 |
+
value = result['pos_prob'] * 100,
|
346 |
+
domain = {'x': [0, 1], 'y': [0, 1]},
|
347 |
+
title = {'text': f"Sentiment: {result['sentiment']}"},
|
348 |
+
delta = {'reference': 50},
|
349 |
+
gauge = {
|
350 |
+
'axis': {'range': [None, 100]},
|
351 |
+
'bar': {'color': colors['pos'] if result['sentiment'] == 'Positive' else colors['neg']},
|
352 |
+
'steps': [
|
353 |
+
{'range': [0, 33], 'color': colors['neg']},
|
354 |
+
{'range': [33, 67], 'color': colors['neu']},
|
355 |
+
{'range': [67, 100], 'color': colors['pos']}
|
356 |
+
],
|
357 |
+
'threshold': {
|
358 |
+
'line': {'color': "red", 'width': 4},
|
359 |
+
'thickness': 0.75,
|
360 |
+
'value': 90
|
361 |
+
}
|
362 |
+
}
|
363 |
+
))
|
364 |
+
else:
|
365 |
+
# Two-way gauge
|
366 |
+
fig = go.Figure(go.Indicator(
|
367 |
+
mode = "gauge+number",
|
368 |
+
value = result['confidence'] * 100,
|
369 |
+
domain = {'x': [0, 1], 'y': [0, 1]},
|
370 |
+
title = {'text': f"Confidence: {result['sentiment']}"},
|
371 |
+
gauge = {
|
372 |
+
'axis': {'range': [None, 100]},
|
373 |
+
'bar': {'color': colors['pos'] if result['sentiment'] == 'Positive' else colors['neg']},
|
374 |
+
'steps': [
|
375 |
+
{'range': [0, 50], 'color': "lightgray"},
|
376 |
+
{'range': [50, 100], 'color': "gray"}
|
377 |
+
]
|
378 |
+
}
|
379 |
+
))
|
380 |
|
381 |
+
fig.update_layout(height=400, font={'size': 16})
|
382 |
+
return fig
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
383 |
|
384 |
@staticmethod
|
385 |
+
def create_probability_bars(result: Dict, theme: str = 'default') -> go.Figure:
|
386 |
+
"""Create probability bar chart"""
|
387 |
+
colors = config.THEMES[theme]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
388 |
|
389 |
+
if result['has_neutral']:
|
390 |
+
labels = ['Negative', 'Neutral', 'Positive']
|
391 |
+
values = [result['neg_prob'], result['neu_prob'], result['pos_prob']]
|
392 |
+
bar_colors = [colors['neg'], colors['neu'], colors['pos']]
|
393 |
+
else:
|
394 |
+
labels = ['Negative', 'Positive']
|
395 |
+
values = [result['neg_prob'], result['pos_prob']]
|
396 |
+
bar_colors = [colors['neg'], colors['pos']]
|
397 |
|
398 |
fig = go.Figure(data=[
|
399 |
+
go.Bar(x=labels, y=values, marker_color=bar_colors, text=[f'{v:.3f}' for v in values])
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
400 |
])
|
401 |
|
402 |
+
fig.update_traces(texttemplate='%{text}', textposition='outside')
|
403 |
fig.update_layout(
|
404 |
+
title="Sentiment Probabilities",
|
405 |
+
yaxis_title="Probability",
|
|
|
406 |
height=400,
|
407 |
showlegend=False
|
408 |
)
|
409 |
|
410 |
return fig
|
411 |
+
|
412 |
@staticmethod
|
413 |
+
def create_batch_summary(results: List[Dict], theme: str = 'default') -> go.Figure:
|
414 |
+
"""Create batch analysis summary"""
|
415 |
+
colors = config.THEMES[theme]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
416 |
|
417 |
+
# Count sentiments
|
418 |
+
sentiments = [r['sentiment'] for r in results if 'sentiment' in r]
|
419 |
+
sentiment_counts = Counter(sentiments)
|
420 |
|
421 |
+
# Create pie chart
|
422 |
+
fig = go.Figure(data=[go.Pie(
|
423 |
+
labels=list(sentiment_counts.keys()),
|
424 |
+
values=list(sentiment_counts.values()),
|
425 |
+
marker_colors=[colors.get(s.lower()[:3], '#999999') for s in sentiment_counts.keys()],
|
426 |
+
textinfo='label+percent',
|
427 |
+
hole=0.3
|
428 |
+
)])
|
429 |
|
430 |
+
fig.update_layout(
|
431 |
+
title=f"Batch Analysis Summary ({len(results)} texts)",
|
432 |
+
height=400
|
433 |
+
)
|
|
|
|
|
|
|
434 |
|
435 |
+
return fig
|
436 |
+
|
437 |
+
@staticmethod
|
438 |
+
def create_confidence_distribution(results: List[Dict]) -> go.Figure:
|
439 |
+
"""Create confidence distribution plot"""
|
440 |
+
confidences = [r['confidence'] for r in results if 'confidence' in r and r['sentiment'] != 'Error']
|
441 |
+
|
442 |
+
if not confidences:
|
443 |
+
return go.Figure()
|
444 |
+
|
445 |
+
fig = go.Figure(data=[go.Histogram(
|
446 |
+
x=confidences,
|
447 |
+
nbinsx=20,
|
448 |
+
marker_color='skyblue',
|
449 |
+
opacity=0.7
|
450 |
+
)])
|
451 |
|
452 |
fig.update_layout(
|
453 |
+
title="Confidence Distribution",
|
454 |
+
xaxis_title="Confidence Score",
|
455 |
+
yaxis_title="Frequency",
|
456 |
+
height=400
|
|
|
|
|
457 |
)
|
458 |
|
459 |
return fig
|
460 |
|
461 |
@staticmethod
|
462 |
+
def create_history_dashboard(history: List[Dict]) -> go.Figure:
|
463 |
+
"""Create comprehensive history dashboard"""
|
464 |
+
if len(history) < 2:
|
465 |
+
return go.Figure()
|
466 |
|
467 |
+
# Create subplots
|
468 |
+
fig = make_subplots(
|
469 |
+
rows=2, cols=2,
|
470 |
+
subplot_titles=['Sentiment Timeline', 'Confidence Distribution',
|
471 |
+
'Language Distribution', 'Sentiment Summary'],
|
472 |
+
specs=[[{"secondary_y": False}, {"secondary_y": False}],
|
473 |
+
[{"type": "pie"}, {"type": "bar"}]]
|
474 |
+
)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
475 |
|
476 |
+
# Extract data
|
477 |
+
indices = list(range(len(history)))
|
478 |
+
pos_probs = [item['pos_prob'] for item in history]
|
479 |
+
confidences = [item['confidence'] for item in history]
|
480 |
+
sentiments = [item['sentiment'] for item in history]
|
481 |
+
languages = [item.get('language', 'en') for item in history]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
482 |
|
483 |
+
# Sentiment timeline
|
484 |
+
colors = ['#4CAF50' if s == 'Positive' else '#F44336' for s in sentiments]
|
485 |
+
fig.add_trace(
|
486 |
+
go.Scatter(x=indices, y=pos_probs, mode='lines+markers',
|
487 |
+
marker=dict(color=colors, size=8),
|
488 |
+
name='Positive Probability'),
|
489 |
+
row=1, col=1
|
490 |
+
)
|
491 |
+
|
492 |
+
# Confidence distribution
|
493 |
+
fig.add_trace(
|
494 |
+
go.Histogram(x=confidences, nbinsx=10, name='Confidence'),
|
495 |
+
row=1, col=2
|
496 |
+
)
|
497 |
+
|
498 |
+
# Language distribution
|
499 |
+
lang_counts = Counter(languages)
|
500 |
+
fig.add_trace(
|
501 |
+
go.Pie(labels=list(lang_counts.keys()), values=list(lang_counts.values()),
|
502 |
+
name="Languages"),
|
503 |
+
row=2, col=1
|
504 |
+
)
|
505 |
+
|
506 |
+
# Sentiment summary
|
507 |
+
sent_counts = Counter(sentiments)
|
508 |
+
fig.add_trace(
|
509 |
+
go.Bar(x=list(sent_counts.keys()), y=list(sent_counts.values()),
|
510 |
+
marker_color=['#4CAF50' if k == 'Positive' else '#F44336' for k in sent_counts.keys()]),
|
511 |
+
row=2, col=2
|
512 |
+
)
|
513 |
+
|
514 |
+
fig.update_layout(height=800, showlegend=False)
|
515 |
+
return fig
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
516 |
|
517 |
+
# Main application functions
|
518 |
def analyze_single_text(text: str, language: str, theme: str, clean_text: bool,
|
519 |
remove_punct: bool, remove_nums: bool):
|
520 |
+
"""Enhanced single text analysis"""
|
521 |
try:
|
522 |
if not text.strip():
|
523 |
+
return "Please enter text", None, None
|
|
|
|
|
|
|
|
|
524 |
|
525 |
# Map display names back to language codes
|
526 |
language_map = {
|
|
|
540 |
'remove_numbers': remove_nums
|
541 |
}
|
542 |
|
|
|
543 |
result = SentimentAnalyzer.analyze_text(text, language_code, preprocessing_options)
|
544 |
|
545 |
# Add to history
|
|
|
558 |
history_manager.add_entry(history_entry)
|
559 |
|
560 |
# Create visualizations
|
561 |
+
gauge_fig = PlotlyVisualizer.create_sentiment_gauge(result, theme)
|
562 |
+
bars_fig = PlotlyVisualizer.create_probability_bars(result, theme)
|
563 |
|
564 |
# Create info text
|
565 |
info_text = f"""
|
566 |
+
**Analysis Results:**
|
567 |
+
- **Sentiment:** {result['sentiment']} ({result['confidence']:.3f} confidence)
|
568 |
- **Language:** {result['language'].upper()}
|
569 |
+
- **Keywords:** {', '.join(result['keywords'])}
|
570 |
+
- **Stats:** {result['word_count']} words, {result['char_count']} characters
|
|
|
|
|
|
|
|
|
|
|
571 |
"""
|
572 |
|
573 |
return info_text, gauge_fig, bars_fig
|
574 |
|
575 |
except Exception as e:
|
576 |
+
logger.error(f"Analysis failed: {e}")
|
577 |
+
return f"Error: {str(e)}", None, None
|
|
|
578 |
|
579 |
def analyze_batch_texts(batch_text: str, language: str, theme: str,
|
580 |
clean_text: bool, remove_punct: bool, remove_nums: bool):
|
581 |
+
"""Batch text analysis"""
|
582 |
try:
|
583 |
if not batch_text.strip():
|
584 |
+
return "Please enter texts (one per line)", None, None, None
|
|
|
|
|
|
|
|
|
585 |
|
586 |
# Parse batch input
|
587 |
texts = TextProcessor.parse_batch_input(batch_text)
|
588 |
|
589 |
if len(texts) > config.BATCH_SIZE_LIMIT:
|
590 |
+
return f"Too many texts. Maximum {config.BATCH_SIZE_LIMIT} allowed.", None, None, None
|
591 |
|
592 |
if not texts:
|
593 |
+
return "No valid texts found", None, None, None
|
594 |
|
595 |
# Map display names back to language codes
|
596 |
language_map = {
|
|
|
635 |
history_manager.add_batch_entries(batch_entries)
|
636 |
|
637 |
# Create visualizations
|
638 |
+
summary_fig = PlotlyVisualizer.create_batch_summary(results, theme)
|
639 |
+
confidence_fig = PlotlyVisualizer.create_confidence_distribution(results)
|
640 |
|
641 |
# Create results table
|
642 |
df_data = []
|
|
|
646 |
'Index': i+1,
|
647 |
'Text': text[:50] + '...' if len(text) > 50 else text,
|
648 |
'Sentiment': 'Error',
|
649 |
+
'Confidence': 0.0,
|
650 |
'Language': 'Unknown',
|
651 |
+
'Error': result['error']
|
652 |
})
|
653 |
else:
|
654 |
df_data.append({
|
|
|
657 |
'Sentiment': result['sentiment'],
|
658 |
'Confidence': f"{result['confidence']:.3f}",
|
659 |
'Language': result['language'].upper(),
|
660 |
+
'Keywords': ', '.join(result['keywords'][:3])
|
661 |
})
|
662 |
|
663 |
df = pd.DataFrame(df_data)
|
|
|
671 |
avg_confidence = np.mean([r['confidence'] for r in successful_results])
|
672 |
|
673 |
summary_text = f"""
|
674 |
+
**Batch Analysis Summary:**
|
675 |
- **Total Texts:** {len(texts)}
|
676 |
- **Successful:** {len(successful_results)}
|
677 |
- **Errors:** {error_count}
|
678 |
- **Average Confidence:** {avg_confidence:.3f}
|
679 |
+
- **Sentiments:** {dict(sentiment_counts)}
|
|
|
|
|
|
|
680 |
"""
|
681 |
else:
|
682 |
+
summary_text = f"All {len(texts)} texts failed to analyze."
|
683 |
|
684 |
return summary_text, df, summary_fig, confidence_fig
|
685 |
|
686 |
except Exception as e:
|
687 |
logger.error(f"Batch analysis failed: {e}")
|
688 |
+
return f"Error: {str(e)}", None, None, None
|
|
|
|
|
689 |
|
690 |
+
def analyze_advanced_text(text: str, language: str, theme: str, include_keywords: bool,
|
691 |
+
keyword_count: int, min_confidence: float):
|
692 |
+
"""Advanced analysis with additional features"""
|
693 |
+
try:
|
694 |
+
if not text.strip():
|
695 |
+
return "Please enter text", None, None
|
696 |
+
|
697 |
+
# Map display names back to language codes
|
698 |
+
language_map = {
|
699 |
+
'Auto Detect': 'auto',
|
700 |
+
'English': 'en',
|
701 |
+
'Chinese': 'zh',
|
702 |
+
'Spanish': 'es',
|
703 |
+
'French': 'fr',
|
704 |
+
'German': 'de',
|
705 |
+
'Swedish': 'sv'
|
706 |
+
}
|
707 |
+
language_code = language_map.get(language, 'auto')
|
708 |
+
|
709 |
+
result = SentimentAnalyzer.analyze_text(text, language_code)
|
710 |
+
|
711 |
+
# Advanced keyword extraction
|
712 |
+
if include_keywords:
|
713 |
+
result['keywords'] = TextProcessor.extract_keywords(text, keyword_count)
|
714 |
+
|
715 |
+
# Confidence filtering
|
716 |
+
meets_confidence = result['confidence'] >= min_confidence
|
717 |
+
|
718 |
+
# Add to history
|
719 |
+
history_entry = {
|
720 |
+
'text': text[:100] + '...' if len(text) > 100 else text,
|
721 |
+
'full_text': text,
|
722 |
+
'sentiment': result['sentiment'],
|
723 |
+
'confidence': result['confidence'],
|
724 |
+
'pos_prob': result['pos_prob'],
|
725 |
+
'neg_prob': result['neg_prob'],
|
726 |
+
'neu_prob': result.get('neu_prob', 0),
|
727 |
+
'language': result['language'],
|
728 |
+
'timestamp': datetime.now().isoformat(),
|
729 |
+
'analysis_type': 'advanced',
|
730 |
+
'meets_confidence_threshold': meets_confidence
|
731 |
+
}
|
732 |
+
history_manager.add_entry(history_entry)
|
733 |
+
|
734 |
+
# Create visualizations
|
735 |
+
gauge_fig = PlotlyVisualizer.create_sentiment_gauge(result, theme)
|
736 |
+
bars_fig = PlotlyVisualizer.create_probability_bars(result, theme)
|
737 |
+
|
738 |
+
# Create detailed info text
|
739 |
+
confidence_status = "✅ High Confidence" if meets_confidence else "⚠️ Low Confidence"
|
740 |
+
|
741 |
+
info_text = f"""
|
742 |
+
**Advanced Analysis Results:**
|
743 |
+
- **Sentiment:** {result['sentiment']} ({result['confidence']:.3f} confidence)
|
744 |
+
- **Confidence Status:** {confidence_status}
|
745 |
+
- **Language:** {result['language'].upper()}
|
746 |
+
- **Text Statistics:**
|
747 |
+
- Words: {result['word_count']}
|
748 |
+
- Characters: {result['char_count']}
|
749 |
+
- Average word length: {result['char_count']/max(result['word_count'], 1):.1f}
|
750 |
"""
|
751 |
|
752 |
+
if include_keywords:
|
753 |
+
info_text += f"\n- **Top Keywords:** {', '.join(result['keywords'])}"
|
754 |
+
|
755 |
+
if not meets_confidence:
|
756 |
+
info_text += f"\n\n⚠️ **Note:** Confidence ({result['confidence']:.3f}) is below threshold ({min_confidence})"
|
757 |
+
|
758 |
+
return info_text, gauge_fig, bars_fig
|
759 |
|
760 |
except Exception as e:
|
761 |
logger.error(f"Advanced analysis failed: {e}")
|
762 |
+
return f"Error: {str(e)}", None, None
|
763 |
+
|
764 |
+
def get_history_stats():
|
|
|
|
|
|
|
|
|
765 |
"""Get enhanced history statistics"""
|
766 |
+
stats = history_manager.get_stats()
|
767 |
+
if not stats:
|
768 |
+
return "No analysis history available"
|
769 |
+
|
770 |
+
return f"""
|
771 |
+
**Comprehensive History Statistics:**
|
|
|
772 |
|
773 |
+
**Analysis Counts:**
|
774 |
- Total Analyses: {stats['total_analyses']}
|
775 |
+
- Positive: {stats['positive_count']}
|
776 |
+
- Negative: {stats['negative_count']}
|
777 |
+
- Neutral: {stats['neutral_count']}
|
778 |
|
779 |
+
**Confidence Metrics:**
|
780 |
- Average Confidence: {stats['avg_confidence']:.3f}
|
781 |
- Highest Confidence: {stats['max_confidence']:.3f}
|
782 |
- Lowest Confidence: {stats['min_confidence']:.3f}
|
783 |
|
784 |
+
**Language Statistics:**
|
785 |
- Languages Detected: {stats['languages_detected']}
|
786 |
- Most Common Language: {stats['most_common_language'].upper()}
|
787 |
|
788 |
+
**Text Statistics:**
|
789 |
- Average Text Length: {stats['avg_text_length']:.1f} characters
|
790 |
+
"""
|
|
|
|
|
|
|
791 |
|
792 |
def filter_history_display(sentiment_filter: str, language_filter: str, min_confidence: float):
|
793 |
"""Display filtered history"""
|
794 |
+
# Convert filters
|
795 |
+
sentiment = sentiment_filter if sentiment_filter != "All" else None
|
796 |
+
language = language_filter.lower() if language_filter != "All" else None
|
797 |
+
|
798 |
+
filtered_history = history_manager.filter_history(
|
799 |
+
sentiment=sentiment,
|
800 |
+
language=language,
|
801 |
+
min_confidence=min_confidence if min_confidence > 0 else None
|
802 |
+
)
|
803 |
+
|
804 |
+
if not filtered_history:
|
805 |
+
return "No entries match the filter criteria", None
|
806 |
+
|
807 |
+
# Create DataFrame for display
|
808 |
+
df_data = []
|
809 |
+
for entry in filtered_history[-20:]: # Show last 20 entries
|
810 |
+
df_data.append({
|
811 |
+
'Timestamp': entry['timestamp'][:16], # YYYY-MM-DD HH:MM
|
812 |
+
'Text': entry['text'],
|
813 |
+
'Sentiment': entry['sentiment'],
|
814 |
+
'Confidence': f"{entry['confidence']:.3f}",
|
815 |
+
'Language': entry['language'].upper(),
|
816 |
+
'Type': entry.get('analysis_type', 'single')
|
817 |
+
})
|
818 |
+
|
819 |
+
df = pd.DataFrame(df_data)
|
820 |
+
|
821 |
+
summary = f"""
|
822 |
+
**Filtered Results:**
|
|
|
823 |
- Found {len(filtered_history)} entries matching criteria
|
824 |
- Showing most recent {min(20, len(filtered_history))} entries
|
825 |
+
"""
|
826 |
+
|
827 |
+
return summary, df
|
|
|
|
|
|
|
|
|
828 |
|
829 |
def plot_history_dashboard():
|
830 |
+
"""Create history dashboard"""
|
831 |
+
history = history_manager.get_history()
|
832 |
+
if len(history) < 2:
|
833 |
+
return None, "Need at least 2 analyses for dashboard"
|
834 |
+
|
835 |
+
fig = PlotlyVisualizer.create_history_dashboard(history)
|
836 |
+
return fig, f"Dashboard showing {len(history)} analyses"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
837 |
|
838 |
def export_history_csv():
|
839 |
"""Export history to CSV"""
|
840 |
+
history = history_manager.get_history()
|
841 |
+
if not history:
|
842 |
+
return None, "No history to export"
|
843 |
+
|
844 |
try:
|
|
|
|
|
|
|
|
|
845 |
df = pd.DataFrame(history)
|
846 |
temp_file = tempfile.NamedTemporaryFile(delete=False, suffix='.csv', mode='w')
|
847 |
df.to_csv(temp_file.name, index=False)
|
848 |
+
return temp_file.name, f"Exported {len(history)} entries to CSV"
|
849 |
except Exception as e:
|
850 |
+
return None, f"Export failed: {str(e)}"
|
|
|
851 |
|
852 |
def export_history_excel():
|
853 |
"""Export history to Excel"""
|
854 |
+
history = history_manager.get_history()
|
855 |
+
if not history:
|
856 |
+
return None, "No history to export"
|
857 |
+
|
858 |
try:
|
|
|
|
|
|
|
|
|
859 |
df = pd.DataFrame(history)
|
860 |
temp_file = tempfile.NamedTemporaryFile(delete=False, suffix='.xlsx')
|
861 |
df.to_excel(temp_file.name, index=False)
|
862 |
+
return temp_file.name, f"Exported {len(history)} entries to Excel"
|
863 |
except Exception as e:
|
864 |
+
return None, f"Export failed: {str(e)}"
|
|
|
865 |
|
866 |
def clear_all_history():
|
867 |
"""Clear analysis history"""
|
868 |
+
count = history_manager.clear()
|
869 |
+
return f"Cleared {count} entries from history"
|
|
|
|
|
|
|
|
|
870 |
|
871 |
def get_recent_analyses():
|
872 |
"""Get recent analysis summary"""
|
873 |
+
recent = history_manager.get_recent_history(10)
|
874 |
+
if not recent:
|
875 |
+
return "No recent analyses available"
|
876 |
+
|
877 |
+
summary_text = "**Recent Analyses (Last 10):**\n\n"
|
878 |
+
for i, entry in enumerate(recent, 1):
|
879 |
+
summary_text += f"{i}. **{entry['sentiment']}** ({entry['confidence']:.3f}) - {entry['text']}\n"
|
880 |
+
|
881 |
+
return summary_text
|
|
|
|
|
|
|
|
|
|
|
882 |
|
883 |
# Sample data
|
884 |
SAMPLE_TEXTS = [
|
|
|
910 |
Amazing quality and fast delivery!
|
911 |
Could be better, but it's okay."""
|
912 |
|
913 |
+
# Gradio Interface
|
914 |
+
with gr.Blocks(theme=gr.themes.Soft(), title="Advanced Multilingual Sentiment Analyzer") as demo:
|
915 |
+
gr.Markdown("# 🎭 Advanced Multilingual Sentiment Analyzer")
|
916 |
+
gr.Markdown("Comprehensive sentiment analysis with batch processing, advanced analytics, and multilingual support")
|
917 |
|
918 |
with gr.Tab("📝 Single Analysis"):
|
919 |
with gr.Row():
|
|
|
950 |
)
|
951 |
|
952 |
with gr.Column(scale=1):
|
953 |
+
result_info = gr.Markdown("Enter text and click Analyze")
|
954 |
|
955 |
with gr.Row():
|
956 |
gauge_plot = gr.Plot(label="Sentiment Gauge")
|
957 |
bars_plot = gr.Plot(label="Probability Distribution")
|
958 |
+
|
959 |
+
with gr.Tab("📊 Batch Analysis"):
|
960 |
with gr.Row():
|
961 |
with gr.Column(scale=2):
|
962 |
+
batch_input = gr.Textbox(
|
963 |
+
label="Batch Text Input (One text per line)",
|
964 |
+
placeholder="Enter multiple texts, one per line...",
|
965 |
+
lines=8
|
966 |
)
|
967 |
|
968 |
with gr.Row():
|
969 |
+
batch_language = gr.Dropdown(
|
970 |
choices=['Auto Detect', 'English', 'Chinese', 'Spanish', 'French', 'German', 'Swedish'],
|
971 |
value='Auto Detect',
|
972 |
label="Language"
|
973 |
)
|
974 |
+
batch_theme = gr.Dropdown(
|
975 |
choices=list(config.THEMES.keys()),
|
976 |
value='default',
|
977 |
label="Theme"
|
978 |
)
|
979 |
|
|
|
|
|
|
|
980 |
with gr.Row():
|
981 |
+
batch_clean = gr.Checkbox(label="Clean Text", value=False)
|
982 |
+
batch_remove_punct = gr.Checkbox(label="Remove Punctuation", value=True)
|
983 |
+
batch_remove_nums = gr.Checkbox(label="Remove Numbers", value=False)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
984 |
|
985 |
+
batch_analyze_btn = gr.Button("🔍 Analyze Batch", variant="primary", size="lg")
|
986 |
|
987 |
gr.Examples(
|
988 |
+
examples=[[BATCH_SAMPLE]],
|
989 |
+
inputs=batch_input,
|
990 |
+
label="Sample Batch Input"
|
|
|
|
|
|
|
|
|
991 |
)
|
992 |
|
993 |
with gr.Column(scale=1):
|
994 |
+
batch_summary = gr.Markdown("Enter texts and click Analyze Batch")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
995 |
|
996 |
with gr.Row():
|
997 |
+
batch_results_table = gr.DataFrame(
|
998 |
+
label="Detailed Results",
|
999 |
+
interactive=False
|
1000 |
+
)
|
1001 |
|
1002 |
with gr.Row():
|
1003 |
+
batch_summary_plot = gr.Plot(label="Sentiment Summary")
|
1004 |
+
batch_confidence_plot = gr.Plot(label="Confidence Distribution")
|
1005 |
+
|
1006 |
+
with gr.Tab("🔬 Advanced Analysis"):
|
1007 |
with gr.Row():
|
1008 |
with gr.Column(scale=2):
|
1009 |
+
advanced_input = gr.Textbox(
|
1010 |
+
label="Text for Advanced Analysis",
|
1011 |
+
placeholder="Enter text for detailed analysis...",
|
1012 |
+
lines=4
|
1013 |
)
|
1014 |
|
1015 |
with gr.Row():
|
1016 |
+
advanced_language = gr.Dropdown(
|
1017 |
choices=['Auto Detect', 'English', 'Chinese', 'Spanish', 'French', 'German', 'Swedish'],
|
1018 |
value='Auto Detect',
|
1019 |
label="Language"
|
1020 |
)
|
1021 |
+
advanced_theme = gr.Dropdown(
|
1022 |
choices=list(config.THEMES.keys()),
|
1023 |
value='default',
|
1024 |
label="Theme"
|
1025 |
)
|
1026 |
|
1027 |
with gr.Row():
|
1028 |
+
include_keywords = gr.Checkbox(label="Extract Keywords", value=True)
|
1029 |
+
keyword_count = gr.Slider(
|
1030 |
+
minimum=3,
|
1031 |
+
maximum=10,
|
1032 |
+
value=5,
|
1033 |
+
step=1,
|
1034 |
+
label="Number of Keywords"
|
1035 |
+
)
|
1036 |
|
1037 |
+
min_confidence_slider = gr.Slider(
|
1038 |
+
minimum=0.0,
|
1039 |
+
maximum=1.0,
|
1040 |
+
value=0.7,
|
1041 |
+
step=0.1,
|
1042 |
+
label="Minimum Confidence Threshold"
|
1043 |
)
|
1044 |
+
|
1045 |
+
advanced_analyze_btn = gr.Button("🔬 Advanced Analyze", variant="primary", size="lg")
|
1046 |
|
1047 |
with gr.Column(scale=1):
|
1048 |
+
advanced_result_info = gr.Markdown("Configure settings and click Advanced Analyze")
|
1049 |
|
1050 |
with gr.Row():
|
1051 |
+
advanced_gauge_plot = gr.Plot(label="Sentiment Gauge")
|
1052 |
+
advanced_bars_plot = gr.Plot(label="Probability Distribution")
|
|
|
|
|
|
|
|
|
|
|
|
|
1053 |
|
1054 |
with gr.Tab("📈 History & Analytics"):
|
1055 |
with gr.Row():
|
|
|
1110 |
outputs=[result_info, gauge_plot, bars_plot]
|
1111 |
)
|
1112 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1113 |
# Batch Analysis
|
1114 |
batch_analyze_btn.click(
|
1115 |
analyze_batch_texts,
|
|
|
1117 |
outputs=[batch_summary, batch_results_table, batch_summary_plot, batch_confidence_plot]
|
1118 |
)
|
1119 |
|
1120 |
+
# Advanced Analysis
|
1121 |
+
advanced_analyze_btn.click(
|
1122 |
+
analyze_advanced_text,
|
1123 |
+
inputs=[advanced_input, advanced_language, advanced_theme, include_keywords, keyword_count, min_confidence_slider],
|
1124 |
+
outputs=[advanced_result_info, advanced_gauge_plot, advanced_bars_plot]
|
1125 |
+
)
|
1126 |
+
|
1127 |
# History & Analytics
|
1128 |
stats_btn.click(
|
1129 |
get_history_stats,
|
|
|
1161 |
outputs=history_status
|
1162 |
)
|
1163 |
|
|
|
1164 |
if __name__ == "__main__":
|
1165 |
+
demo.launch(share=True)
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|