Personalized product search based on user transaction history and hypergraph learning

As the e-commerce shopping websites like Amazon become more and more popular, amounts of products spring up on the internet and bring great difficulties to product search. However, the conventional text-based search is confined to retrieving products relevant to query and personalized product search is still a challenging problem in e-commerce. Consequently, in this paper, we propose a personalized product search approach, which combines personalized multimedia recommendation into searching. First, we construct a hypergraph based on products' descriptions and user's transaction history. Then the similarity between products and the user is calculated based on two kind of textural feature extraction methods. After that, iterative procedure is introduced to obtain the final relevance score of each product to the user. Experimental results on our collected Amazon dataset show the effectiveness of the proposed approach. The MAP@5 of our method can reach 0.48 and the MAP@10 can reach 0.44. We propose a new re-ranking method for personalized product search, in which we utilize user's transaction history to choose products which is closer to the user's preference into the higher positions. Experimental results on our collected dataset show that our method is much better than the comparison methods.