基于社交网络的协同过滤推荐算法研究

发布时间：2018-02-16 17:14

本文关键词： 推荐系统协同过滤相似度计算社交网络　出处：《华南理工大学》2013年硕士论文　论文类型：学位论文

【摘要】：随着互联网技术和社交网站的迅猛发展，信息过载问题越来越严峻。个性化推荐系统可以主动分析用户的行为，挖掘用户的兴趣，从而为用户推荐满足他需求的信息资源，成为了解决信息过载问题的一个重要工具。其中协同过滤推荐算法由于其简单且普适性强，在各个领域的推荐系统中得到了广泛的应用。本文论述了推荐系统的研究背景、现状和意义，详细介绍了协同过滤推荐算法的基本思想和实现过程，分析其采用的相似度计算方法存在的不足，以及算法单独依赖评分数据的局限。对此，我们从改善相似度计算方法和如何借助社交网络改善协同过滤推荐算法两个方面展开研究。本文的主要工作和贡献如下： 1、提出一种融合Jaccard系数规范化欧氏距离方法（JNED）来度量用户之间的相似度。该方法使得处于不同评分维度空间用户的相似度具有可比性且可靠度更高。通过实例情景验证和具体实验统计表明，JNED方法得到的相似度结果和实际情况比较吻合，同时具有较高的信息利用率。 2、提出一种基于社交网络的协同过滤推荐算法(SNCF)。SNCF通过社交网络为目标用户搜索候选邻居集，，结合用户熟识度和评分相似度作为最终的相似度，以此生成最近邻居并预测评分，产生推荐。该方法避免了单独依赖评分相似度的局限，可以更加准确的获得更多的邻居用户，从而提高推荐算法的准确度和覆盖率。 3、利用爬虫从豆瓣网采集真实社交网络数据和电影评分数据进行算法验证实验。利用这些数据对本文提出的JNED相似度计算方法和SNCF推荐算法进行验证。实验结果表明，相对于传统的相似度计算方法和基于用户的协同过滤推荐算法，JNED相似度计算方法和SNCF算法在预测评分准确度、覆盖率和新颖度都有提升。
[Abstract]:With the rapid development of Internet technology and social network, the problem of information overload is becoming more and more serious. It has become an important tool to solve the problem of information overload. Because of its simplicity and universality, collaborative filtering recommendation algorithm has been widely used in recommendation systems in various fields. This paper discusses the research background, current situation and significance of recommendation system, introduces in detail the basic idea and implementation process of collaborative filtering recommendation algorithm, and analyzes the shortcomings of the similarity calculation method. As well as the limitation that the algorithm depends on the score data alone. In this paper, we study the two aspects of improving similarity calculation method and how to improve collaborative filtering recommendation algorithm with the help of social network. The main work and contributions of this paper are as follows:. 1. A Jaccard coefficient normalized Euclidean distance (Euclidean distance method) is proposed to measure the similarity between users. This method makes the similarity of users in different dimensions of score more comparable and more reliable. The results of scene verification and experimental statistics show that the similarity obtained by JNED method is in good agreement with the actual situation. At the same time, it has high information utilization ratio. 2. A collaborative filtering recommendation algorithm based on social network (SNS) is proposed to search candidate neighbor set for target users through social network, which combines user familiarity and score similarity as the final similarity, so as to generate nearest neighbor and predict score. This method avoids the limitation of relying solely on the similarity of score, and can obtain more neighbor users more accurately, thus improving the accuracy and coverage of the recommendation algorithm. 3. The crawler is used to collect real social network data and movie score data from Douban net for algorithm verification. The JNED similarity calculation method and the SNCF recommendation algorithm proposed in this paper are verified by these data. The experimental results show that, Compared with the traditional similarity calculation method, the user based collaborative filtering recommendation algorithm and the SNCF algorithm, the prediction accuracy, coverage and novelty are improved.
【学位授予单位】：华南理工大学
【学位级别】：硕士
【学位授予年份】：2013
【分类号】：TP391.3

【参考文献】