Research On Item Comprehensive Similarity And Factor Analysis For Collaborative Filtering Recommendation Algorithm

With the rapid growth of the informatization progress, Internet has become gradually penetrated into various areas of people’s work and life. The rapid growth of information lead to problems of " information overload " and " resource isotropic " have appeared. Be faced with huge amounts of data, how to find the information that people are interested in quickly in a short time become a focus. In this context, the personalized recommendation system came into being. The core technology of personalized recommendation system is recommendation algorithm, because the quality of recommendation algorithm determines the performance of the recommendation system. In a large number of recommendation algorithms, collaborative filtering algorithm is one of the most successful.This paper first introduces the collaborative filtering technology and its research status at home and abroad. Then it introduces the classification of collaborative filtering technology: memory-based collaborative filtering and model-based collaborative filtering, and analyze its advantages and disadvantages. Next we make a detailed analysis comparing between item-based collaborative filtering and user-based collaborative filtering, and point out that item-based collaborative filtering is better than user-based collaborative filtering in general. Finally, on the basis of in-depth study of the existing problems of collaborative filtering, this paper do the following work:With the problem of item’s cold start and the accuracy of similarity calculation between items is low, this paper proposes a new similarity calculation method--item-integrated similarity. The item-integrated similarity is the combination of the item-rated similarity and item-classified similarity, for the item-rated similarity is the similarity that we get from the traditional collaborative filtering algorithm and item-classified similarity is the similarity of the item’s classification characteristics between items. This similarity-calculated method not only can effectively alleviate the problem of cold start, but also can measure the similarity between items more accurate.With the problem of poor scalability and high-dimension of user-item rating matrix, this paper puts forward a clustering algorithm based on item-integrated similarity. Through clustering we can reduce the matrix dimensionality and decrease the search space of the nearest neighbor, in this way we can improve the problem of poor scalability for recommendation system.With the problem of low precision of recommendation, this paper proposes that using factor analysis method to ensure the minimum lost. Then we propose a new hybrid recommendation technology-a collaborative filtering algorithm based on item comprehensive similarity and factor analysis. The algorithm combines the advantages of item-based collaborative filtering algorithm and model-based collaborative filtering algorithms, it can effectively improve the recommendation quality of collaborative filtering algorithm.Finally we make a series of experiments to prove that the algorithm we propose has a better recommendation quality.