Enhancement of Sentiment Analysis in Hotel Reviews through Latent Semantic Indexing and Convolutional Neural Networks

Abstract

Sentiment Analysis (SA) is a prominent field of study concerned with the classification of sentences within a document as either positive or negative. The extraction of features from a document plays a crucial role in SA for achieving precise text classification. This study employs the Latent Semantic Indexing (LSI) algorithm for feature extraction, designed to address the limitations inherent in the Term Frequency-Inverse Document Frequency (TF-IDF) technique. The features extracted are then utilized in the Convolutional Neural Network (CNN) classification algorithm, which encompasses two convolutional layers, a single polling layer, a fully-connected layer, and two output nodes. This is done to evaluate the efficacy of the proposed model. Experimental results indicate that the combination of LSI and CNN significantly improves text classification. Customer reviews exert considerable influence on individuals' travel plans, with a preference typically shown towards hotels with a preponderance of positive reviews. Consequently, these reviews serve as crucial resources for managers seeking to enhance their services. In this study, a dataset of hotel reviews is employed, and the resulting data is evaluated using standard metrics such as precision, recall, f-score, and accuracy, yielding results of 89%, 77%, 80.5%, and 87% respectively.