Collaborative Filtering Recommendation Algorithm Based On Item Classification And Cloud Model

As the development of personalized recommendation service in e-commerce,current online shopping mode has switched from traditional information retrieval fromlarge-scale data source to personalized recommendation service with more diversity andpersonality. Collaborative filtering technique (CF), as the cornerstone for personalizedrecommendation service, thus plays an important role. Two fundamental problems,however, should be tackled to achieve a better performance for this technique. On onehand, the outcome of recommendation is adversely affected by data sparseness andcold-start. On the other hand, the repeated computation similarity significantly reducesthe scalability of CF.This paper proposes an improved CF algorithm based on the classification of itemsto handle the two problems mentioned above. And then, based on this algorithm and thecloud model, another algorithm has also been proposed to overcome the problem ofwrong recommendation in existing collaborative filtering algorithm which based oncloud model. The main contents of this paper are as follows:First, an improved CF algorithm based on the classification of items is introducedto overcome the problems caused by the data sparseness and inaccuracy of the userneighbors. The new algorithm first rates the unrated items by applying the itemclassification, and then calculates the user similarity within classes fornearest-neighbors, after which it could recommend the items based on the finalprediction.The research on the cloud model which focuses on the process of transforming aqualitative concept to a set of numerical values, shows that by replacing user ratingvectors with the cloud feature vectors to compute similarity, cloud model could get amore accurate set of the nearest neighbors, and thus avoid the drawback of strictattribute matching in traditional method to calculate the similarity. This model, however,has its own limitation. A conclusive recommendation set is generated for the user,whereas a classified recommendation set is more suitable for different interestingaspects from the user.In order to solve this problem of cloud model, this paper presents another newcollaborative filtering recommendation algorithm by combining the item classificationand cloud model. Firstly the algorithm utilizes the item classification information and cloud model to compute items inner-similarity, and then gets the scores from neighboritems which have the highest similarity and uses their scores to forecast the unratedinner-class items. Secondly, the neighbors of user are obtained by computing theinner-class user similarities in the cloud model, providing the final forecast grade andcarrying out the recommendation.Finally, the Mat lab experiment results indicate the efficiency and scalability bothalgorithm in terms of recommendation quality and accuracy compared to traditional CFalgorithms.