A Collaborative Filtering Recommendation Algorithm Based On User Characteristic Attribute And Cloud Model

Along with the emergence of WEB2.0technology, internet user get rid of the age ofaccepting information passively, they become more and more active in the internet world.User can interactive and share life with friends through WEB2.0. However, with the rapidincrease of users, a new generation of internet information model based on producing userinformation leads to the explosive growth of internet information, which also brings trouble touser in searching the information they’re interested in.“Information Overload” makes itdifficult for user to find the needed information quickly and accurately. In the twotechnologies, Information Filtering technology represented by recommender systems caninitiatively recommend information to user for what they need, which makes recommendersystems become research priorities for the current scholars.As one of the most successful technologies, collaborative filtering can mine the potentialneeds of users and make recommendation accordingly. But there are still many issues exist inapplication, such as the problems of data sparsity and the cold start and so on, needing to besolved. Aiming to the key, this thesis launches a study mainly from the following severalaspects.Firstly, through learning of the cloud model and research of the similarity calculation incollaborative filtering, it is known that taking the method of computing the cloud similaritybased on cloud model instead of the tradition method of similarity calculation can computeuser similarity more accurately, and improve the recommend quality. The experimental resultsmanifest that the similarity calculation method based on cloud model can obtain usersimilarity more effectively, and improve the quality of algorithm.Secondly, aiming at the problem that the traditional user-based collaborative algorithm donot take into user feature properties when calculating user similarity, this thesis presents animproved collaborative filtering algorithm based on user feature property. It adds user featureproperty similarity when calculating user similarity, and gets the final similarity result throughweighted factor. It can obtain similar user more accurately. It also can avoid the cold startproblem that a new user has no record when adding in the recommend system.Thirdly, in order to solve the problem of data sparsity, this thesis takes advantage of thecloud model combining with data filling, presents an improved collaborative algorithm. Thisalgorithm initially predicts the user item rating through cloud model, and fills the result into the initial matrix, then calculates the user similarity based on the new matrix, gets the finalsimilarity combining with the user feature property. At last, it gets the final prediction for userto make final recommendation.Finally, this thesis takes experiments using the data sets provided by MovieLens site. Theexperimental results show that the collaborative filtering algorithm based on use feature andcloud model can make more effective recommendation to user. By adding user featureproperty, it can provide user with accurate and personalized recommendation, it also can solvethe user cold start problem. Combining with the filling data method based on cloud model, thenew collaborative filtering algorithm not only has the above advantages, but also can solvethe problem of data sparsity in the tradition collaborative filtering algorithm.