import string import torch import numpy as np def extract_text_features(review_text, sentiment_analyzer): """ Extracts 6 review-level metadata features to match inputs expected by the model: 1. Review Length (character count) 2. Sentiment Compound Score (VADER Compound value) 3. Punctuation count 4. Average word length 5. Capitalization ratio 6. Days since first review (static placeholder value = 0) Returns: torch.Tensor of shape [1, 6] containing the normalized metadata vector. """ # 1. Review character length review_length = len(review_text) # 2. VADER Sentiment Compound Score if sentiment_analyzer is not None: sentiment = sentiment_analyzer.polarity_scores(review_text)['compound'] else: sentiment = 0.0 # 3. Punctuation count punctuation_count = sum(1 for char in review_text if char in string.punctuation) # 4. Average word length words = review_text.split() avg_word_length = np.mean([len(word) for word in words]) if words else 0.0 # 5. Capitalization ratio capital_letters = sum(1 for char in review_text if char.isupper()) capitalization_ratio = capital_letters / len(review_text) if review_text else 0.0 # 6. Days since first review placeholder days_since_first_review = 0.0 features = [ float(review_length), float(sentiment), float(punctuation_count), float(avg_word_length), float(capitalization_ratio), float(days_since_first_review) ] return torch.tensor(features, dtype=torch.float32).unsqueeze(0)