Sentiment Analysis of Restaurant Reviews Using Moth-Flame Optimized Fuzzy Logic and XGBoost on LDA-Weighted TF-IDF Features

As mobile technologies and networks become more common, online review websites have seen a big growth in user-generated content.The current rise in posts requires new, adaptive methods of performing sentiment analysis with restaurant reviews and feedback in service industries.The model we introduce uses Latent Dirichlet Allocation (LDA), Term Frequency-Inverse Document Frequency (TF-IDF), fuzzy logic and extreme gradient boosting (XGBoost) to perform hybrid sentiment analysis.Moth-Flame Optimization (MFO) algorithm is applied to our model to tune both the thresholds and settings of XGBoost.First, the pipeline does input processing using standard natural language processing (NLP) methods, next it performs feature extraction with TF-IDF and LDA.Fuzzy logic rates features according to how much emotion is in a review and how frequently the features are mentioned.The results are improved further using MFO which seeks out the right classifier settings and fuzzy rules for a better performance.To find both linguistic patterns and trends in sentiment, text data is analyzed at the levels of unigrams, bigrams and trigrams.Classification is done by using the well-optimized XGBoost model which provides both the sentiment of texts along with confidence scores.The results indicate that the hybrid way of building a model is more effective than the standard approaches in several evaluation areas.The method reaches 96.07% in accuracy, 95.43% in sensitivity, and 97.12% in specificity and F1-score of 96.16% for Kaggle restaurant review dataset.These results prove that combining advanced feature extraction, fuzzy logic reasoning and evolutionary optimization securely handles sentiment analysis on digital media.