Topic Modeling Using LDA-Based and Machine Learning for Aspect Sentiment Analysis

Abstract

E-commerce platforms facilitate the buying and selling of goods. Customers can also provide feedback on their shopping experiences. Analyzing consumers' feedback is essential for the e-commerce industry, particularly for shop owners. However, a detailed analysis of reviews individually is impractical. Fortunately, machine learning models allow for identifying the factors that drive consumer decisions when purchasing products and analyzing their opinions expressed in reviews. As the number of product transactions rises, it is also increasingly essential to automatically identify the factors influencing consumer choices and their resulting positive or negative reactions. This paper proposes topic modeling using Latent Dirichlet Allocation (LDA) for identifying influential topics from customer text reviews. The PRDECT-ID dataset was utilized to build the machine-learning models. The XGBoost Classifier achieved the highest accuracy of 87% in topic modeling through hyperparameter tuning. Furthermore, the Random Forest model achieves the highest accuracy of 93% for sentiment analysis. During the experiment, the authors discovered that topic modeling accuracy improves with hyperparameter tuning of LDA and machine learning models. Moreover, utilizing TF-IDF as a technique for extracting features during sentiment analysis boosts sentiment classification accuracy.