融合主题模型和协同过滤的多样化移动应用推荐

发布时间：2019-07-10 12:21

【摘要】：随着移动应用的急速增长,手机助手等移动应用获取平台也面临着信息过载的问题.面对大量的移动应用,用户很难找到最适合的;而另一方面,长尾应用淹没在资源池中不易被人所知.已有推荐方法多注重推荐准确率,忽视了多样性,推荐结果中多是下载量高的应用,使得推荐系统的数据积累越来越偏向于热门应用,导致长期的推荐效果越来越差.针对这一问题,首先改进了两种推荐方法,提出了将用户的主题模型和应用的主题模型与MF相结合的LDA_MF模型,以及将应用的标签信息和用户行为数据同时加以考虑的LDA_CF算法.为了结合不同算法的优点,在保证推荐准确率的条件下提升推荐结果的多样性,提出了融合LDA_MF,LDA_CF以及经典的基于物品的协同过滤模型的混合推荐算法.使用真实的大数据评测所提推荐算法,结果显示,所提推荐方法能够得到推荐多样性更好且准确率更高的结果.
[Abstract]:With the rapid growth of mobile applications, mobile application acquisition platforms such as mobile assistants are also facing the problem of information overload. In the face of a large number of mobile applications, it is difficult for users to find the most suitable; on the other hand, long-tailed applications are not easily known in the resource pool. The existing recommendation methods pay more attention to the recommendation accuracy and ignore the diversity. Most of the recommendation results are the applications with high downloads, which makes the data accumulation of the recommendation system more and more inclined to hot applications, resulting in the long-term recommendation effect getting worse and worse. In order to solve this problem, two recommendation methods are improved, and a LDA_MF model which combines the user's topic model and the applied topic model with MF is proposed, and the LDA_CF algorithm which considers the applied label information and user behavior data at the same time is proposed. In order to combine the advantages of different algorithms and improve the diversity of recommendation results under the condition of ensuring the accuracy of recommendation, a hybrid recommendation algorithm combining LDA_MF,LDA_CF and classical object-based collaborative filtering model is proposed. Using the real big data to evaluate the proposed recommendation algorithm, the results show that the proposed recommendation method can get better recommendation diversity and higher accuracy.
【作者单位】：数据工程与知识工程教育部重点实验室(中国人民大学信息学院);清华大学经济管理学院;
【基金】：国家自然科学基金(71272029,71490724,61472426) 国家高技术研究发展计划(863)(2014AA015204) 北京市自然科学基金(4152026)~~
【分类号】：TP391.3

【相似文献】