A product review writing recommender system based on LDA and TF-IDF

Abstract

Twitter is a micro-blogging platform where people broadcast their views and opinions to fellow users in crisp messages called Tweets. However, the platform's format of restricted character limit makes it challenging for many users to express their views exhaustively. The paper proposes a recommender system to help in writing effective product review Tweets within the restricted character limit of Twitter. The approach is divided into two phases where, the first phase uses the Latent Dirichlet Allocation (LDA) algorithm to find pivotal features from the training corpus and suggests them to the users while writing new Tweets. In the second phase, the approach suggests the most appropriate opinion words to describe the respective features by using an approach based on the occurrence frequency of opinion words and TF-IDF. The evaluation results show significant improvement in the quality of product review Tweets. The percentage of good reviews corresponding to a parameter such as correct usage of feature words is found to be 17.85 percent higher, whereas an improvement of 23.22 percent is reported with regard to the correct use of opinion words using the generated recommendations.