推荐算法的研究及易物网的实现

发布时间：2018-01-28 09:26

本文关键词： 协同过滤相似度平均偏差推荐 Slope One算法易物网平台　出处：《北京交通大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着互联网的快速发展,人们可以获取到海量的数据,这极大地促进了人类的进步。然而,随着数据量的不断增长,如何在海量的数据中捕获自己感兴趣的信息已成为大数据分析的研究热点之一,在这种情况下,个性化推荐技术应运而生。个性化推荐不仅可以提高用户在有效的时间内发现自己感兴趣信息的效率,同时可以使商户及时主动的将有用信息提供给用户。因此,研究个性化推荐算法具有极大的商业价值和意义,已经引起了学术界和商业界的广泛关注。因此,本文针对基于用户的个性化协同过滤算法进行了研究并将算法应用于构建的易物网交易平台,主要研究成果包括:(1)针对协同过滤算法的数据稀疏这一问题,本文提出了一种基于项目活跃度的填充算法。该算法对用户的评分数据进行slope one预填充,有效地解决了单一使用用户评分的个数来计算用户相似度数据稀疏的问题,填充的方式简单合理有效。与传统填充方法相比,所提算法能够增强数据的稀疏性和提高用户相似度计算的精度。(2)针对协同过滤算法数据稀疏的相似度计算精确度的问题,本文提出了一种基于距离惩罚因子的协同过滤算法。该算法将用户间共同评分交集的所有评分距离作为惩罚因子来修正传统皮尔森相似度,通过对相似度增加距离惩罚因子自适应的调整用户相似度,改善协同过滤优化算法中用户间相似度精确度。实验验证了所提算法的有效性。(3)针对儿童绘本交易的实际问题,构建了易物网平台并且将上述提出的协同过滤算法集成在实际平台中。该平台主要包括四个模块:推荐图书模块、图书交易模块、会员管理模块、图书维护模块。本文提出的算法成功地应用于该易物网平台,并实现了推荐模块具有的基本功能。
[Abstract]:With the rapid development of the Internet, people can obtain a large amount of data, which greatly promotes the progress of human beings. However, with the continuous growth of data. How to capture the interesting information in the massive data has become one of the research hotspots in big data's analysis, in this case. Personalized recommendation technology emerges as the times require. Personalized recommendation can not only improve the efficiency of users to find their interesting information in an effective time. At the same time, it can make merchants provide useful information to users in time. Therefore, the study of personalized recommendation algorithm has great commercial value and significance, and has attracted extensive attention from academia and business circles. In this paper, the personalized collaborative filtering algorithm based on users is studied and applied to the tradeoff platform of barter net. The main research results include: 1) sparse data for collaborative filtering algorithm. In this paper, a filling algorithm based on item activity is proposed, which prepopulates the user's rating data with slope one. It effectively solves the problem of using the number of user scores to calculate the sparse data of user similarity, and the filling method is simple, reasonable and effective, compared with the traditional filling method. The proposed algorithm can enhance the sparsity of data and improve the accuracy of user similarity calculation. In this paper, a cooperative filtering algorithm based on distance penalty factor is proposed, which uses all the scoring distances of the common score intersection among users as penalty factors to modify the traditional Pearson similarity. The user similarity is adjusted adaptively by increasing the distance penalty factor to the similarity. Improve the accuracy of user similarity in collaborative filtering optimization algorithm. Experimental results show that the proposed algorithm is effective. The platform is composed of four modules: recommended book module, book transaction module and member management module. The algorithm proposed in this paper has been successfully applied to the platform and realized the basic functions of the recommendation module.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.3

【参考文献】