""" Sentiment AI Analysis - Core sentiment/emotion detection engine Analyzes text and detects emotions using keyword matching Focuses on the LAST SENTENCE for real-time accuracy """ import re from typing import Dict, Any from .sentiment_keyword_map import KeywordMap class SentimentAnalyzer: """ Ultra-fast emotion analyzer optimized for real-time detection Detects 50+ emotions from text as it's being typed or streamed Focuses on the LAST SENTENCE/PHRASE for accurate real-time updates """ SENTENCE_SEPARATORS = re.compile(r'[.!?;:\n]+') def __init__(self, custom_keywords: Dict[str, str] = None): """Initialize with keyword mappings""" self.keyword_map = KeywordMap(custom_keywords) self._word_pattern = re.compile(r'\b\w+\b') def _get_last_sentence(self, text: str) -> str: """ Extract the last sentence/phrase from text This ensures emoji reflects CURRENT sentiment, not cumulative """ parts = self.SENTENCE_SEPARATORS.split(text) parts = [p.strip() for p in parts if p.strip()] if not parts: return text.strip() return parts[-1] def _analyze_segment(self, text: str, position_weight: float = 1.0) -> Dict[str, float]: """ Analyze a text segment and return emotion scores Args: text: Text to analyze position_weight: Multiplier for recency (higher = more recent) Returns: Dict mapping emotions to scores """ text_lower = text.lower() words = self._word_pattern.findall(text_lower) if not words: return {} emotion_scores: Dict[str, float] = {} negation_active = False intensifier_active = False # Process words with position weighting word_count = len(words) for idx, word in enumerate(words): # Position weight: last word gets full weight, first gets 0.5 word_position_weight = 0.5 + (0.5 * (idx / max(word_count - 1, 1))) # Check for negation if self.keyword_map.is_negation(word): negation_active = True continue # Check for intensifier if self.keyword_map.is_intensifier(word): intensifier_active = True continue # Check if word maps to emotion emotion = self.keyword_map.get_emotion_for_word(word) if emotion: # Handle negation if negation_active: emotion = self.keyword_map.get_opposite_emotion(emotion) negation_active = False # Calculate score with position weighting base_score = 1.5 if intensifier_active else 1.0 final_score = base_score * word_position_weight * position_weight intensifier_active = False # Accumulate emotion_scores[emotion] = emotion_scores.get(emotion, 0) + final_score else: negation_active = False intensifier_active = False return emotion_scores def analyze(self, text: str) -> Dict[str, Any]: """ Analyze text and return detected emotion FOCUSES ON LAST SENTENCE for real-time accuracy Args: text: Text to analyze Returns: Dict with 'label' (emotion), 'score', and 'details' """ if not text or not text.strip(): return {"label": "neutral", "score": 0.0, "details": {}} # Get the last sentence (what user is currently typing) last_sentence = self._get_last_sentence(text) # Analyze last sentence with full weight last_sentence_scores = self._analyze_segment(last_sentence, position_weight=2.0) # If found emotions in last sentence, use those if last_sentence_scores: primary_emotion = max(last_sentence_scores, key=lambda k: last_sentence_scores[k]) max_score = last_sentence_scores[primary_emotion] normalized_score = min(max_score / 3.0, 1.0) return { "label": primary_emotion, "score": normalized_score, "details": {"segment": "last_sentence", "text": last_sentence[:50]} } # Fallback: analyze full text if last sentence has no emotion words full_text_scores = self._analyze_segment(text, position_weight=1.0) if not full_text_scores: return {"label": "neutral", "score": 0.0, "details": {"segment": "none"}} primary_emotion = max(full_text_scores, key=lambda k: full_text_scores[k]) max_score = full_text_scores[primary_emotion] normalized_score = min(max_score / 3.0, 1.0) return { "label": primary_emotion, "score": normalized_score, "details": {"segment": "full_text", "all_emotions": full_text_scores} } if __name__ == "__main__": print("=" * 60) print("Testing Sentiment Analyzer") print("=" * 60) analyzer = SentimentAnalyzer() test_cases = [ ("I am so happy today!", "happiness"), ("I love this!", "love"), ("This is exciting!", "excitement"), ("Thank you so much!", "gratitude"), ("I am really sad", "sadness"), ("This makes me angry", "anger"), ("I am scared", "fear"), ("So frustrating!", "frustration"), ("I am curious about this", "curiosity"), ("Wow, that is amazing!", "amazement"), ("I am so confused", "confused"), ("I miss you", "longing"), ] print("\nBasic Emotion Detection:") passed = 0 for text, expected in test_cases: result = analyzer.analyze(text) match = expected in result["label"] or result["label"] == expected status = "✓" if match else "✗" if match: passed += 1 print(f"{status} '{text}' → {result['label']}") print(f"\n{passed}/{len(test_cases)} tests passed") print("\n" + "=" * 60) print("Last Sentence Focus Tests (Real-Time Updates)") print("=" * 60) multi_tests = [ ("I love this! But now I am angry", "anger"), ("Happy day! Wait, I'm confused", "confused"), ("Great work! This is frustrating", "frustration"), ("Sad news. But I'm grateful now", "gratitude"), ("I was scared. Now I'm excited!", "excitement"), ] print("\nLast Sentence Detection:") passed_multi = 0 for text, expected in multi_tests: result = analyzer.analyze(text) match = expected in result["label"] or result["label"] == expected status = "✓" if match else "✗" if match: passed_multi += 1 print(f"{status} '{text[:50]}...' → {result['label']}") print(f"\n{passed_multi}/{len(multi_tests)} tests passed") print("\n✅ Sentiment Analyzer ready!")