Group Recommender Systems with Weighted Matrix Factorization and Sentiment Analysis

Given the rapid and continuous expansion of the Internet and social networks in recent years, leading to massive growth of data, the use of group recommender systems (GRSs) has experienced dramatic growth. This is because a GRS works much better, compared to personal recommender systems, in an area where a group of people participates in a single activity. Additionally, the principles of GRSs can be used to overcome some shortcomings of personal recommender systems, such as cold start. So far, many group recommenders have been presented in a variety of areas such as tourism, e-commerce, social networks, entertainment and travel, and they are used in more diversified areas every day. Thus, today, given the large volume of data, we know as Big Data, as well as the involvement of groups of people in various systems, such as social networks, GRSs have become a vital requirement for online services. In this paper, the architecture of the proposed system includes three modules - constructing groups, rewriting ranks, and modeling group interests - and creating recommendations. Assuming that the users are not grouped in advance, the proposed system in the construct module divides the users by K-means clustering method into groups with the same features, which is done randomly in most of these systems. Moreover, the new task performed using sentiment analysis to modify the difference in ratings and user comments for entries in rewriting ratings module are for improving system accuracy. Finally, the matrix factorization method has been used to create recommendations for the group. The new task performed is the use of the matrix factorization method, especially the weighted matrix analysis in group-recommender field because this method was used more in personal recommender. In doing so, three approaches - AF, BF and WBF - were designed in the group-interest modeling and creating recommendations modules and each approach was tested, which resulted in an increase in system performance compared to previous samples.

کلیدواژه ها:

Group Recommender Systems ، Weighted Matrix Factorization ، Sentiment Analysis ، GRS ، Big Data ، K-means Clustering ، Matrix Factorization