Comparing the effect of sentiment analysis and user ratings on the performance of recommender systems

In recent years, the emergence of social networks has led to an increasing attention to recommender systems based on user reviews. The purpose of developing such systems is to use valuable information from users' textual comments in the process of modeling and recommending. User comments reflect the actual opinions on the products and services, so they are a valuable resource for recommending. In social networking environments, collaborative filtering systems are used to provide advice to users. The basis of this approach is the experience and opinion of the other people to buy items and products. This approach is based on the assumption that users who have the same interest have a similar rank. In this research, a system is proposed to provide recommend for users to buy books by combining the collaborative filtering and sentiment analysis. For sentiment analysis, ensemble methods based on weighted voting have been used to extract user’s opinions. In the weighting method, a greater weight is assigned to a classifier which has higher accuracy. The selected model has been evaluated on the 7210 user’s comment which extracted from the Amazon website by the web crawler. The results show that the sentiment analysis of the feeling of the users' comments systems has a positive effect on the performance of recommender systems.