Implicit aspect based sentiment analysis for restaurant review using LDA topic modeling and ensemble approach

Abstract

Technological advancements in e-commerce and Web 2.0 have revolutionized the way customers express their opinions about services and features through reviews on various websites. This trend is especially prevalent in the travel industry, where online sources provide valuable insights into the food and shelter aspects of destinations. However, the sheer volume of reviews available online makes it challenging for travellers to filter relevant information. To address this issue, aspect-based sentiment analysis (ABSA) is proposed as a technique to extract feature-based opinions. Topic modelling and sentiment analysis are two significant techniques used to assist with this analysis. Topic modelling is the process of identifying thematic relationships among documents, while sentiment analysis identifies the opinions expressed in the text. In this study, one of the leading travel websites, Tripadvisor, is used to collect customer reviews of various restaurants, which are then subjected to aspect-based sentiment analysis using latent dirichlet allocation (LDA) and ensemble bagging support vector machine (EBSVM) classifier techniques. The aim is to identify the most relevant aspect for the restaurant domain and improve sentiment analysis performance. To address the issue of class imbalances in the datasets, synthetic minority over-sampling technique (SMOTE) is implemented. The performance of the LDA is evaluated using the coherence score, which produces quality topics for restaurant reviews. The effectiveness of the EBSVM classifier is measured using accuracy, precision, recall, and F1 score. The proposed model achieves an accuracy of 96.1%, which outperforms other techniques discussed in the existing literature. Overall, this study demonstrates the effectiveness of aspect-based sentiment analysis in extracting relevant opinions from a large volume of reviews and highlights the potential of machine learning techniques in improving sentiment analysis performance. The suggested approach enhances overall performance when compared to other techniques discussed in the existing literature.