Project Characteristics Of Model-based Collaborative Filtering Recommendation Algorithm

Due to the changing information technology and quickly developing network,we encounter the problem of information overload.A giant amount of information make user to feel very difficult to satisfy their real requirement when manually accomplish information retrieval work, so different personalized recommendation systems come out today.The recommendation systems accomplish information filter according to user's preference to items,and with the utility of KD technology we carry out personalized recommendation. Collaborative filtration is the technology widely used for reducing information overload.When caculating similarity in nearest neighbor searching among the target items, the tradional Item-based collaborative filtering recommendation algorithm only consider users'evaluation,but not take the attributes of items themselvs into consideration, so when the users' evaluation data is not enough,it is difficult to accurately attain nearest neighbor sets, and it lead to recommendation accuracy decreasing since the loss of information while forming the nearest neighbor sets of the target items.As to the problems existing with the tradional Item-based collaborative filtering recommendation algorithm, this paper put forward the collaborative filtering algorithm based on item characteristic model (Abbreviated as ICM-based Algorithm) .First,we construct item characteristic model on the offline context,then with utility of a new method to select unevaluated items we evaluate those items which have not been evaluated by users, and calculate out the similarity among them ,at last we obtain nearest neighbor of those items and give out recommendation to the user.We construct experimental database from the public MovieLens datasets,with the database we accomplish the experiment of collaborative filtering algorithm based on item characteristic model,and compare the result with the traditonal Item-based collaborative filtering algorithm,which take relation simlarity as measuring standard of simlarity.The experimental results demonstrate that the recommendation accuracy of ICM-based collaborative filtering recommendation algorithm is better than the accuracy of traditional Item-based collaborative filtering recommendation algorithm.Through test and analysis of the experimental data,it is verified that the algorithm take the attributes of item itself and user's evaluation into consideraton,so the algorithm increase the measuring accuracy of items similarity,and calculate the nearest neighbor more accurately.Even if user evaluation data is not enough, the algorithm also can give out better accuracy of recommendation, and efficiently upgrade recommendation quality.In addition, since the preliminary evaluation of unevaluated items is obtained according to evaluation of its similar items, also it works for new items, so to some extent, and it can ease the cold-start problem in traditional Item-based collaborative filtering recommendation algorithm.