Research On E-commerce Recommender System Based On Clustering Algorithm

The growing popularity and in-depth development of the Internet and mobile Internet has creates the current era of Big Data. The rapid increase of data volume in the field of e-commerce is particularly evident, which provides more opportunities for future development as well as the problem of information overload, i.e. Users can not access to the information they are really interested in from the mass merchandise information. In this context recommender system came into being.Collaborative filtering algorithm is the most widely used technology in recommender system, but also shows some shortages such as poor real-time responsiveness, low accuracy, cold-start and sparse-data problems in practical applications. Among the recent researches on recommender system, some scholars have proposed the introduction of clustering technology to solve some of the problems above. Clustering-based recommendation algorithm divides users or projects into different clusters, inside which elements having high similarity, to simplify the nearest-neighbor searching process and reduce the overall computational complexity as well as time consuming. In addition, since the clustering process can be done off-line, the overall real-time responsiveness of recommender system can be greatly enhanced.In this paper, some theoretical research of clustering algorithm and recommendation algorithms have been investigated thoroughly. Some improved algorithms are proposed and their effectiveness are verified experimentally. The main work and contents of this paper are as follows:(1) Bringing the Entropy-weight into the calculation of traditional Euclidean distance to improve K-means algorithm, so that elements in the same cluster can have more accurate similarity. Using a logarithmic variation to optimize the way how the particle of particle swarm optimization algorithm (PSO) update its velocity, in order to improve the overall search ability and convergence of the algorithm. And also get the learning parameters of particle optimized. By combining the improved K-means algorithm with the improved PSO algorithm, an improved hybrid clustering algorithm and named as Ajusted PSO-Kmeans algorithm has been proposed.(2) Optimizing user-based collaborative filtering algorithm by introducing the Entropy-weight into the calculation of Pearson similarity, in order to eliminate dimensionless difference between user-rating vectors in the process of finding the nearest neighbor and improves the recommendation accuracy. By combining the improved PSO-Kmeans algorithm with the improved UCF algorithm, an improved Clustering-based recommendation algorithm and named as A-UCCF algorithm has been proposed.(3) A few experiments have been conducted to verify the effectiveness of the algorithms proposed above. Firstly, experiments for the proposed Ajusted PSO-Kmeans algorithm, K-means and PSO-Kmeans algorithm have been done on the data set Iris and Wine, and result shows that the new algorithm has better clustering quality and accuracy. Secondly, compare the proposed A-UCCF algorithm with UCF and UCCF algorithm based on the data set MovieLens, and result shows that the new algorithm has both better recommendation accuracy and higher real-time responsiveness.