Research On Commodity Recommendation Methods And Some Key Issues

Online shopping is becoming more and more popular in people’s daily life.With the rapid growth of the user in e-commerce, the number of commodities in online business platform has been greatly enriched. However, faced with massive product choices, for the consumers it is very difficult to pick out what they really need. Commodity recommendation service is a main approach for third-party provider to solve this problem, but currently most of the existing applications are still remaining in the manual stage, depending much on the artificial discovery and recommendation from web editors and lacking of customization and personalization.We designed an intelligent commodity recommendation system zuikemai.com, aiming at the recommendation of commodity, commodity set, commodity classification and commodity-related articles, and we had a deep exploration and research on the problems encountered in the system:(1) Recommendation of item set is introduced to the recommender system, which expands the traditional user-item model. We proposed a new collaborative filtering algorithm based on composite score to solve the set recommendation.(2) Introduced a new perspective, item recommendation for user is regarded as a special case of classification recommendation, while each item is a smallest class, as a leaf node on the multiple level classification tree. We proposed a classification granularity dynamic selection algorithm based on sparsity to select the appropriate classification granularity, which can be combined with commodity sort within classification granularity to get the commodity recommendation for users.(3) We proposed a new user-based collaborative filtering algorithm based on open platforms such as Sina Weibo, using external data to solve the cold start problem.(4) And we proposed a reciprocal recommendation algorithm which based on an improved combination of correlation between commodities and articles.