Study On Online Review Helpfulness Classification Based On Product Feature Mining

With the rapid development of e-commerce, more and more consumers are accustomed to shopping online. When online-shopping behavior occurs, consumers can comment on purchased goods, which provides not only feedback to sellers, but also advice and guidance to other consumers. An increase in product sales always brings large amount of online reviews, and some hot products receive ten thousands of reviews which make consumers hard to handle with. It requires both sides to quickly filter out helpful ones from the mass of product reviews and extract useful information from a large number of redundant information to guide the sale and purchase.The urgent needs of useful information indicated in massive online reviews have raised domestic and foreign researchers'concern on a specific field of review mining-review helpfulness classification.In this study, we take into account the reality that most e-commerce sites generally don't provide comprehensive review information, and starting from the product feature information in review contents, we provide a reference for review helpfulness classification by product feature mining. To take full advantage of the massive reviews, we adopt a semi-supervised learning approach to train the classification model. Finally, we obtain a review helpfulness classification model with excellent performance.We analyze the deficiencies of existing product features mining method first, and improve word segmentation, feature selection and pruning to get an optimized product feature mining results. On this basis, we study the influential factors of review helpfulness, and add product feature information as an important reference to the feature set of the review helpfulness classification. Finally, we utilize an important extension of Support Vector Machine Transductive Support Vector Machine to deploy semi-supervised learning and train the semi-supervised classification model of online review helpfulness, which use both labeled and unlabeled reviews. The result shows that the classification model outperforms the traditional supervised learning model, especially when considering review content information only. The result indicates that product feature information is an important influential factor of review helpfulness, and semi-supervised learning can effectively improve the classification performance.