Research On Personalized Recommendation Algorithm Integrating Image Similarity And Collaborative Filtering

With the development of Internet, the scale of e-commerce market expand quickly. Asa very important shopping assistant, e-commerce recommendation system can help usersfind the desired products in the mass merchandise, and also can help enterprises retainusers effectively, increase sales and enhance the competitiveness.At present, the recommendation system has been widely used in large-scalee-commerce websites, such as Amazon, Taobao, JD. However, there are still multipleproblems existing in traditional collaborative filtering technology, such as cold starts,sparsity problem and scalability problem, and also the problem that the majority of currente-commerce shopping websites attract customers through the image information of product.Focusing on the research on the recommendation algorithm how to take advantage ofproduct image information, this thesis proposed a recommendation algorithm integratingimage similarity and collaborative filtering. The main research work is as follows:(1) Taking into account the sparsity problem, the new user problem as well as lowcredibility problem in looking for nearest neighbors, this thesis proposed a similarity factorwhich can adaptively adjust based on the ratings data and a preference factor which candistinguish user’s positive score from negative score. Taking the two factors intoconsideration, this thesis proposed a collaborative filtering recommendation algorithmusing optimizing neighbors selection.(2) The key problem of image similarity recommendation is image matching. Throughthe in-depth study of local feature image matching algorithm, this thesis proposed a newlocal feature description algorithm called CGCI-SIFT, and through experiments proved itsperformance which compared with SIFT, ASIFT, SURF and PCA-SIFT on six aspectsincluding scaling, rotational, brightness, affine invariant and time consumption.(3) This thesis proposed an interest model and an adjustable weight balance factorbased on the integrating strategy which integrating image similarity and the collaborativefiltering recommendation of neighbor. This thesis proposed a personalizedrecommendation system integrating image similarity and collaborative filtering. Accordingto the characteristics of E-commerce’s massive images, using the visual vocabulary tree toreduce the dimension of image feature.