Research On Collaborative Filtering Recommendation Algorithm Based On Cloud Model And User Clustering

With the rapid development of internet, data speed growth, the problem of information overload is being paid attention. Although, the technique of collaborative filtering has achieved good results in terms of solving the problem of information overload, it suffers from the problems of data sparse and scalable with the increase of users and item in practical application. These problems become hot research topic in the field, and have very good research value. Therefore, how to effectively alleviate the data sparse problem and other issues of collaborative filtering recommendation system to further improve the forecast accuracy of the recommendation system is the main objective of this research.Clustering techniques are often used to group similar users in recommendation system, and then it can effectively find a reasonable set of similar neighbors to improve forecast accuracy. Therefore, a kind of improved fuzzy clustering algorithm is proposed(SoMKfcm). In order to improve the traditional Fuzzy C-Means algorithm which is sensitivity to initial point and easily to falls into local optimal solution, some improvements are proposed in the SoMKfcm. Firstly, a new initial point selection strategy is proposed, which effectively avoid the influence of noise data points; Secondly, the weighted value of sample and the sample distance of cluster center are integrated into the objective function; Finally, the simulated annealing algorithm is combined with to optimize the iterative solution process, and adding solving stochastic jump to avoid falling into local optimum results. The experimental results on MATLAB platform based on real data sets show that compared with the traditional fuzzy clustering algorithm, the proposed algorithm has a higher accuracy of clustering results, and effectively improves the original defects of the traditional algorithm.In the basis of the aforementioned work, based on score data and personal information data, we present a novel approach that combines the clustering-based recommendation model with personal information and cloud model(CCCF algorithm). Firstly, we use personal information and backward algorithm to reconstruct ratings to generate user fusion behavior Preference Vectors. Secondly, based on the integration of behavioral preference matrices, clusters is generated by utilizing SoMKfcm algorithm, and the strategy of importance group selection is given to provide data smoothing and neighborhood selection, and then a new multi-dimensional similarity method is proposed. Finally, based on the above process prediction model is made.In order to verify the effectiveness of the CCCF algorithm, the experiments utilized Moveilens 1m and Moveilens 100 k dataset to make comparative tests with several other related algorithms. The results show that under different sparse circumstances, CCCF algorithm can effectively alleviate the impact of the data sparse sex on recommendation algorithm, and the algorithm prediction accuracy has been significantly improved.