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.

随着像Amazon这样的电子商务购物网站变得越来越受欢迎，互联网上涌现出大量产品，给产品搜索带来了很大困难。但是，传统的基于文本的搜索仅限于检索与查询有关的产品，个性化产品搜索仍然是电子商务中的一个难题。因此，在本文中，我们提出了一种个性化的产品搜索方法，该方法将个性化的多媒体推荐与搜索结合在一起。首先，我们基于产品的描述和用户的交易历史记录构造一个超图。然后基于两种纹理特征提取方法，计算出产品与用户之间的相似度。之后，引入迭代过程以获得每个产品与用户的最终相关性得分。在我们收集的Amazon数据集上的实验结果表明了该方法的有效性。我们的方法的MAP @ 5可以达到0.48，而MAP @ 10可以达到0.44。我们提出了一种用于个性化产品搜索的新排名方法，其中我们利用用户的交易历史记录来选择更接近用户偏好的产品到更高的位置。在收集的数据集上的实验结果表明，我们的方法比比较方法要好得多。