A Study On Collaborative Recommendation Model Based On Rough Set

With rapid development of e-commerce, online consumers are able to obtain a very large number of products and messages. Compared with previous, consumers can get various products on line much easier and more directly. However, there is no better structure for consumers to manage resulting from the huge, increasing, updating and everywhere distributed products on line. Thus we get too much related information, and get little when we search them some time. It is a requirement for e-commerce websites and a tendency of future service mode to provide personalized recommendations for online consumers. The personalized recommending technology, which is the key part of a recommender, is the recommending methods and modes that can comprehend users'interests and recommend related products for them. Meanwhile, it assists websites to attract users and their loyalties.As one of the widely applicated and state-of-the-art method in personalized recommending technology, collaborative filtering(CF) recommendation is able to provide the opinion of the others who have the similar interests with the active users online. This method takes advantage of the previous information to calculate the similarity between users, the ratings that neighbor users with high similarity rated on other items to predict the preference content of active users show on the target items. According to the preference, the system recommends for the active users. The merit of CF is no special demand on recommending items and is able to deal with the items difficult to structure such as books, music, movies, etc. With the increasing amount of users and items in recommender, the ratings on items are becoming extremely sparseness. The scalability and lower precision lead it to be limitedly applicated in real recommender.This paper proposes a new collaborative recommending model called the collaborative recommending model based on rough set. It attributes the lower precision in personalized recommendation to the hybrid interests and insufficient use of indirect ratings, which results in the worse reflection of calculated similarity towards the actual interests. Therefore, this paper proposes the incorporation of rough sets in CF recommendation, and form the new model. Firstly, the model depicts the hybrid interests via rough users clustering on the attention content from the user-item ratings, which can avoid the imprecise similarity when the user affiliates to the boundary of the user clustering. Secondly, based on the rough user clustering, the model pads the unrated items to make full use of indirect ratings, according to the improved method on ROSTIDA. Thirdly, calculate the similarity using preference assertion factor to predict and recommend items. The whole recommending procedure is formed as the collaborative recommending model based on rough set and related recommending algorithms are improved and depicted. After the analysis on real data, the experiment suggests that the proposed model in this paper is effective on the improvement of precision in CF, and the scalability is in control as well. The research in this paper is concluded with the shortcomings and the outlook given at last.