融合用户兴趣度和信任度的协同过滤推荐算法

Collaborative filtering recommendation algorithm based on user interestingness and believability

针对传统协同过滤推荐算法推荐精度较低等问题,提出一种个性化社交活动推荐算法。采用文件主题模型求取用户与其参加过的所有社交活动的主题分布,利用隐含主题概率分布表征用户的兴趣度;利用信任传递机制求取用户间的直接信任值和间接信任值;综合用户对社交活动的兴趣度和评分,构建新的兴趣度相似矩阵得到用户间的综合相似度,将综合用户相似度与用户信任度相融合,得到个性化推荐权值,以不同的权值配比获得最终的社交化活动推荐,利用豆瓣同城数据集确定各模块的参数设置值。对比实验结果表明,在保证较高覆盖率的基础上,算法相较于其它两种推荐算法准确率至少提高了5.26%,召回率至少提高了12.5%。

To solve the low recommendation accuracy of traditional collaborative filtering recommendation algorithm,a persona-lized social activity recommendation algorithm was proposed.The topic distribution of all social activities that the user participated in was extracted using the LDA file theme model,and the user’s interest degree was characterized by the implicit topic pro-bability distribution.The trust transfer mechanism was used to obtain the direct trust value and the indirect trust value between the users.The interest degree and rating of users on social activities were integrated,the new interest degree similarity matrix was constructed to obtain the comprehensive similarity between users,the integrated user similarity and user trust were integrated to obtain personalized recommendation weights.The parameter setting values of each module were determined using the watercress dataset.Compared with two other recommendation algorithms,the accuracy of the proposed algorithm is improved by at least 5.26%and the recall rate by at least 12.5%on the basis of guaranteeing high coverage.