Abstract
Companies having different products rely on their customers' reviews of their products. After purchasing a product, a customer will post some reviews on the website. Before purchasing the product, another customer will read the feedback from these reviews before making a decision. It is very important for companies to analyse such reviews, whether they are negative or positive, to enhance the quality of their product. Researchers are now working on separating the negative or positive comments by means of sentiment scores. A sentiment analysis can be performed through supervised learning or unsupervised learning, where each method requires a lot of pre-processing work for the analysis. This paper presents a strategy for sentiment classification using Latent Semantic Indexing (LSI). The purpose of LSI is to rank documents with respect to a given query. However, in this study, a mechanism was provided to generate positive and negative queries automatically. These queries were then used to obtain negative and positive scores so that a decision could be made on the basis of these scores. This method was not only aimed at separating the positive and negative reviews, but also at providing ranked lists of positive or negative comments. These lists are very important for companies to carry out significant reviews from the top of the negative list, and shining reviews from the top of the positive list. The sorted lists of positive/negative reviews based on the LSI scores generated by the positive/negative queries were checked manually, and were proved to be highly satisfactory, while the precision of the sentiment analysis was 0.67, which could be increased by a little bit of tuning of the given reviews. The MCC value also showed that this method was acceptable.