融合用户信任度和相似度的基于核心用户抽取的鲁棒性推荐算法

Robust Recommendation Algorithm Based on Core User Extraction with User Trust and Similarity

推荐系统可以方便地帮助人们做出决策,然而,目前很少有研究考虑到剔除不相关噪声用户的影响,保留少量核心用户做推荐。该文提出基于信任关系和兴趣相似度的核心用户抽取的新方法。首先计算所有用户对之间的信任度和兴趣相似度并且排序,然后根据用户在最近邻列表中出现的频率和位置权重两种策略选择候选核心用户集合,最后利用用户的推荐能力筛选出最终的核心用户并且做推荐。实验表明利用核心用户做推荐的有效性,并且证明了利用20%的核心用户做推荐,可以达到超过90%的准确性,而且利用核心用户做推荐能很好地抵御托攻击对推荐系统造成的负面影响。

Recommendation systems can help people make decisions conveniently.However,few studies consider the effect of removing irrelevant noise users and retaining a small number of core users to make recommendations.A new method of core user extraction is proposed based on trust relationship and interest similarity.First,all users trust and interest similarity between pairs are calculated and sorted,then according to the frequency and position weight users travel in the nearest neighbor in the list of two kinds of strategies for the selection of candidate core collection of users.Finally,according to the user’s ability the core users are sieved out.Experimental results show that the core user recommendation effectiveness,and verify that the core of user 20%can reach more than recommended accuracy of 90%,and through the use of core user recommendation the negative effects caused by the attacks on the recommendation system can be resisted.