Multi-Dimensional Aspect Level Helpfulness Prediction of Online Customer Reviews

Estimating and predicting the reviews? helpfulness become essential for consumers and e-commerce systems that help access the proper reference through massive product reviews. Reviews? helpfulness is usually calculated based on perceived helpful votes. This study extends the prior review helpfulness studies. It aims to identify review characteristics that best represent the review helpfulness and then use them to predict accurate new helpfulness scores at the minimum possible error. Several natural language processing tools are configured and used to extract review characteristics from the Amazon.com dataset. Six review characteristics (i.e., review age, review aspects, review length, review polarity, review rating, and review subjectivity) that span the three main categories of the review elements were identified as the most influential for review helpfulness and proposed a multiple linear regression (MLR) model that makes use of such characteristics to predict review?s helpfulness. The ability of the proposed model to predict the review helpfulness at minimum error was tested and compared with related prediction methods under various scenarios. The results show that combining the characteristics associated with the linguistic, content, and peripheral review elements improves the accuracy of helpfulness prediction, and the proposed MLR model predicts the most accurate helpfulness score at minimum error. The MLR model outperforms the SVM and DT methods by 17.68% and 1.74% in reducing MAE error and by 9.3% and 0.91% in reducing RMSE error, respectively. This study offers a novel contribution to the literature by illustrating the importance of incorporating the most influential review characteristics in the review helpfulness prediction and how it affects the predictive performance. This study extends ongoing studies on helpfulness prediction and provides notable implications for research and practice; e-commerce systems can have better organization and ranking of their reviews, and customers can efficiently access knowledge to make better purchase decisions.