Analyzing category correlations for recommendation system

Since the late 20th century, the Internet users have noticeably increased and these users have provided lots of information on the Web and searched for information from the Web. Now there are huge amount of new information on the Web everyday. However, not all data are reliable and valuable. This implies that it becomes more and more difficult to find a satisfactory result from the Web. We often iterate searching several times to find what we are looking for. Researcher suggests a recommendation system to solve this problem. Instead of searching several times, a recommendation system proposes relevant information. In the Web 2.0 era, a recommendation system often relies on the collaborative filtering from users. In general, the collaborative filtering approach works based on user information such as gender, location or preference. However, it may cause the cold-star problem or the sparsity problem since it requires initial user information. Recently, there are several attempts to tackle these collaborative filtering problems. One of such attempts is to use category correlation of contents. For instance, a movie has genre information given by movie experts and directors. We notice that these category information are more reliable compared with user ratings. Moreover, a newly created content always has category information; namely, we can avoid the cold-start problem. We consider a movie recommendation system. We revisit the previous algorithm using genre correlation and improve the algorithm. We also test the modified algorithm and analyze the results with respect to a characteristic of genre correlations.