基于协同过滤的个性化推荐算法研究及系统实现

发布时间：2018-01-01 12:07

本文关键词：基于协同过滤的个性化推荐算法研究及系统实现　出处：《西南交通大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着互联网的飞速发展,信息呈现几何级的增长,海量数据问题日趋严重,用户快速获得自己感兴趣的信息变得相当困难。加之移动设备的普及,用户更愿意在移动设备上获取信息。在这样的情况下,个性化推荐系统应运而生。本文在实际需求的驱动下,首先了解推荐系统的理论知识,学习相关推荐算法,然后重点研究了协同过滤推荐算法和k-means聚类算法,并对协同过滤算法进行改进。最后设计和实现了文章推荐子系统。本文做的工作主要有以下几个方面:1.分析了热门对象可能对推荐算法中用户相似度的影响,提出在计算用户相似度时加入惩罚因子,以降低热门对象对用户相似度的影响。通过实验验证,提出的算法的准确率和召回率都有所提高。2.对于协同过滤算法的时间瓶颈及扩展问题,提出了采用聚类算法对协同过滤算法进行改进。在聚类时不单单使用用户的评分信息,而且挖掘了用户评分对象的特征信息,通过这两部分信息为用户偏好进行建模,进而聚类。通过实验验证,改进后的算法的推荐效率与预测准确率都有所提升。在此基础上,综合惩罚因子与用户聚类两种策略来改进基于用户的协同过滤算法。通过实验验证,综合改进比任意的单一改进在推荐效率与推荐质量都要优越。3.根据某公司业务的特定手机应用场景和具体需求,应用提出的协同过滤算法设计并实现了文章推荐子系统。该子系统主要包含用户日志收集与处理模块、用户与推荐对象建模模块、推荐模块和推荐列表展示模块,并且对这四个模块进行设计与实现。
[Abstract]:With the rapid development of Internet, information is growing exponentially, massive data is becoming a serious problem, users quickly get the interesting information becomes quite difficult. Coupled with the popularity of mobile devices, users prefer to get information on a mobile device. In this case, the personalized recommendation system came into being. Based on the actual needs of the driver first, understand the theory of knowledge recommendation system, learning recommendation algorithm, and then focus on the collaborative filtering algorithm and K-means clustering algorithm and the collaborative filtering algorithm was improved. The design and implementation of the recommended subsystem. The main work of this paper is as follows: 1. analysis of hot objects may affect the user recommendation algorithm in similarity, is proposed by adding a penalty factor in calculating the similarity of users, in order to reduce the degree of similarity of users hot. Ring. Through the experiment, the accuracy of the proposed algorithm and the recall rate has increased to.2. and extended the time bottleneck problem in collaborative filtering algorithm, the clustering algorithm of collaborative filtering algorithm is improved. Not only the user clustering score information and mining feature information of user rating objects. Through this two part information for user preference modeling and clustering. Through the experiment, the recommended efficiency and improved prediction accuracy are improved. On this basis, the comprehensive penalty factor and user clustering two strategies to improve user based collaborative filtering algorithm. Through the experiment, comprehensive improvement than single improvement in any recommendation efficiency and quality of recommendation can be superior to.3. according to specific application scenarios of a mobile phone business and the specific needs, using the proposed collaborative filtering algorithm. The sub recommendation system is implemented and implemented. The subsystem mainly includes user log collection and processing module, user and recommendation object modeling module, recommendation module and recommendation list display module, and designs and implements these four modules.

【学位授予单位】：西南交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.3

【参考文献】