Matrix Factorization Techniques for Analysis of Online Rating Data

• Main Author: 張良卉
• Other Authors: Ruby Chui-Hsing Weng
• Format: Others
• Language: zh-TW
• Online Access: http://ndltd.ncl.edu.tw/handle/21309354969407723185

碩士 === 國立政治大學 === 統計研究所 === 101 === The explosive growth of the internet has led to information overload. Electronic retailers and content providers use recommender systems to meet a variety of special needs and tastes. The retailers use the internet as a marketing method, and the consumers use the internet to find the products they want. Recommender systems then appear. Such systems are particularly useful for entertainment products such as movies, music, and TV shows. Recommender systems recommend the products or the information users may like to them by their characteristic and preference. Recommender systems can be divided to two strategies. One is content filtering approach, which creates a profile for each user or product to characterize its nature. Another is collaborative filtering approach, which relies only on past user behavior without requiring the creation of explicit profiles. Collaborative filtering analyzes relationships between users and interdependencies among products to identify new user-item associations. The two primary areas of collaborative filtering are the neighborhood methods and latent factor models. Neighborhood methods are centered on computing the relationships between items or, alternatively, between users. Latent factor models are an alternative approach that tries to explain the ratings by characterizing both items and users on factors inferred from the ratings patterns. Matrix factorization techniques are some of the most successful realizations of latent factor models. One benefit of the matrix factorization approach to collaborative filtering is its flexibility in dealing with various data aspects and other application-specific requirements. It tries to capture the interactions between users and items that produce the different rating values. However, much of the observed variation in rating values is due to effects associated with either users or items, known as biases or intercepts, independent of any interactions. This research try to find out whether putting the biases into matrix factorization models makes the prediction more accurate. This research analyzed the MovieLens data from GroupLens Research Project of Minnesota University. We found that adding biasterms to matrix factorization can improve the accuracy of prediction, though it requires a bit more computing time.