Aspect Based Sentiment Analysis Using Modified Latent Dirichlet Allocation and Optimized BERT with LSTM Classification

Aspect-Based Sentiment Analysis (ABSA) is essential in industries such as healthcare, automotive, and finance, where understanding customer feedback is crucial for business strategies and product development.Traditional sentiment analysis methods often fail to account for specific aspects, providing only generalized sentiment scores (polarities).This study aims to develop a robust ABSA architecture to accurately classify aspect-based sentiment polarity.We propose a Modified Latent Dirichlet Allocation (M-LDA) model that integrates Term Frequency-Inverse Document Frequency (TF-IDF) and Latent Dirichlet Allocation (LDA) to extract relevant aspects.Additionally, an Optimized Bidirectional Encoder Representations from Transformers (O-BERT) model is used to determine aspect-based sentiment polarity, categorizing sentiments as positive, negative, or neutral.A Long Short-Term Memory (LSTM) network is then employed to enhance classification accuracy.The M-LDA O-BERTLSTM for ABSA (MOL-ABSA) performance, evaluated using accuracy and macro-F-score metrics on the SemEval 2014 and a generalized dataset, demonstrates its effectiveness, achieving accuracies of 81.19%, 85.26%, and 87.53%, respectively.This integrated approach combines traditional NLP techniques with advanced machine learning models to ensure high accuracy.