基于分布式SVD和社交影响力的APP推荐算法

发布时间：2018-05-18 11:48

本文选题：推荐算法 + 协同过滤　；参考：《华东师范大学》2017年硕士论文

【摘要】：随着移动互联网的飞速发展,手机应用(简称APP)的种类和数量都呈现高速增长趋势。据统计截止2017年3月,安卓和苹果应用市场一共拥有约为500万款的APP,如何更好地向用户推荐APP是提升用户体验和增加企业营收的关键性问题。我们发现现有APP推荐策略主要基于用户个人信息,并没有考虑社交网络中用户所处群体对其产生的影响。然而与现实中购物相似,用户在下载或购买APP时必然会受到社交网络好友的影响。因此,向传统APP个性化推荐算法中融入网络用户间的社交影响力是否会带来正向收益,以及如何融入是一个值得深入探索的课题。本文对个性化推荐算法和社交影响力进行深入研究,提出了基于分布式SVD和社交影响力的APP推荐算法,简称SVD-WSI。该算法的核心目标是向用户推荐预测使用次数较高的APP,主要包含两个部分:第一部分根据用户个人信息,从个性化角度预测用户使用APP次数,提出基于分布式SVD的APP使用次数预测算法—D-SVD;第二部分根据用户所处社交群体信息,从社交影响角度进行预测,提出基于社交影响力的APP使用次数预测算法—D-SI。在实验部分,我们首先基于MovieLens公开数据,验证了 D-SVD算法相比基于用户、物品的协同过滤以及SVD++算法有更高的准确率。其次我们基于腾讯提供的应用宝数据,验证了 SVD-WSI算法引入社交影响力后对推荐效果带来的正向收益。最后针对求解社交网络中影响力传播最广的用户集合—影响力最大化问题,我们基于腾讯数据对比了 D-SI算法影响力计算模型与基准WC算法的差异,实验发现本文所提算法能够挖掘到影响力传播更广的用户集合。
[Abstract]:With the rapid development of mobile internet, the category and quantity of mobile phone application (app) are increasing rapidly. Android and Apple have about 5 million apps in the app market as of March 2017, and how to better recommend the APP to users is a key issue for improving the user experience and increasing corporate revenue. We find that the current APP recommendation strategy is mainly based on the user's personal information and does not take into account the influence of the user's group on the social network. However, similar to real shopping, users are bound to be influenced by social network friends when downloading or buying APP. Therefore, whether it will bring positive benefits to the traditional APP personalized recommendation algorithm, and how to integrate it, is a topic worth further exploring. In this paper, the personalized recommendation algorithm and social influence are deeply studied, and a APP recommendation algorithm based on distributed SVD and social influence is proposed, which is called SVD-WSI. The core goal of the algorithm is to recommend the app with high prediction times to the user, which includes two parts: the first part predicts the number of users using APP from the angle of personalization according to the personal information of the user. Based on distributed SVD, a APP usage prediction algorithm-D-SVD is proposed. In the second part, based on the information of the users' social groups, the APP usage prediction algorithm-D-SI based on the social influence is proposed from the point of view of social influence. In the experimental part, we first based on the MovieLens open data, verify that the D-SVD algorithm has higher accuracy than the user-based collaborative filtering and SVD algorithm. Secondly, based on the YingYongBao data provided by Tencent, we verify that the SVD-WSI algorithm brings positive benefits to the recommendation effect by introducing social influence. Finally, in order to solve the problem of maximizing the influence of the most widely used users in social networks, we compare the difference between the D-SI algorithm and the benchmark WC algorithm based on Tencent data. Experimental results show that the proposed algorithm can mine the user set with wider influence spread.
【学位授予单位】：华东师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.3

【相似文献】