Research Of Collaborative Filtering Recommendation Algorithm Based On Clustering Methods

With the rapid development and popularization of the Internet and Mobile Terminals, Internet information content render exponentially surge, which will also lead to the information overload problem. How to find users really interested in huge amounts of data information and recommend it to the user has become a hot research in the industry. Recommended system can help users to quickly dig the potential and deep-level information what is need by users. It also can help users extract information content faster and better, so as to be widely used. Collaborative filtering is one of the successful recommendation algorithm, and its simple model concepts and implementation process much favored by large enterprises. However, there are four problems of cold start, data sparseness, computational efficiency and system expandability about collaborative filtering.Because Cold start and data sparseness problem for collaborative filtering, this paper introduced a concept of cloud model put forward by the Academician of Li De-yi. Cloud model belongs to the field of uncertain artificial intelligence, mainly multi-dimensional vector mapped to three-dimensional vectors, which use excepted, entropy and hyper entropy to quantify. So this concept makes the problem from the local to the global which can reduce the negative influence from the data sparse and cold start. In addition, this paper presents an improved similarity calculation model basic on cloud model, and the model can better describe the similarity between users (Project). Finally, When computing cloud characteristics, this paper also consider the user preference factors, such as time factors and score factors.Because computational efficiency and system expandability for collaborative filtering, this paper proposed a clustering algorithm based on binaiy tree. This is the reason why low computational efficiency. So using clustering algorithm the number of comparisons neighbor become a research hotspot. K-Means algorithm is not only limited to the center and set the initial value of K, but there is a problem you can’t belong to multiple categories. So an improved clustering algorithm based on binary tree is presented. The binary tree structure and K-Means clustering algorithm combine to form a new hierarchical clustering algorithm.Finally, this paper will fuse the improved clustering algorithm and cloud model, which is from an improved recommendation algorithm basic on clustering. In the calculation process because there is no correlation between nodes, this algorithm introduces a distributed computing framework Spark based memory, and the algorithm in this paper parallel implementation on the platform. Through the experimental proof, the algorithm can not only enhance accuracy, but also parallelism to improve system scalability.In this paper, the collaborative filtering algorithm based on clustering is put forward by the test In the simulation reality cluster environment. It is mainly used to evaluate the stability, accuracy and the response time of recommendation algorithm, which is scientific and effective evaluation for it. Contrasting experimental results, the proposed collaborative filtering algorithm based on clustering has outstanding performance and meet the requirements of the algorithm in stability, accuracy, efficiency and scalability.