面向在线交友领域的互惠推荐算法研究

发布时间：2018-05-23 13:43

本文选题：互惠推荐 + 在线交友　；参考：《山东师范大学》2017年硕士论文

【摘要】：随着电子商务的发展,出现了越来越多的用户到用户模式的双向推荐,传统个性化推荐的项目到用户的推荐已经无法满足用户的需求,这就催生了时下以在线交友网站为代表的互惠推荐的蓬勃发展。在线交友网站每天为进行交友的异性双方提供大量的交友信息,求职网站每天不仅为求职者提供大量的职位信息,而且为企业的人员招聘提供大量的求职简历信息,但是这些信息繁杂庞大,缺乏有效的系统分类,无论是对交友双方还是对参与到求职招聘环节中的求职人员和招聘人员,都难以精确定位自己感兴趣的对象。因此,改进互惠推荐算法并且最大程度上提高推荐质量是迫切需要解决的。本文提出面向在线交友领域的互惠推荐算法,从矩阵补全、互惠相似度两个方面来提高互惠推荐算法的效率,充分考虑两个方面之间的关系,对互惠推荐算法进行了较为深入的研究工作。主要工作及贡献如下:(1)针对以在线交友为代表的互惠推荐系统中存在的数据稀疏问题,提出一种缓解数据稀疏性的矩阵补全算法。首先,深入分析了常用的两种矩阵补全方法;其次,在此基础上将两种方法进行混合加权,提出了基于LMaFit和K-Means混合加权的矩阵补全算法,将低秩矩阵补全LMaFit算法和K-Means聚类算法的优势进行互补;最后,通过实验验证了基于LMaFit和K-Means混合加权的矩阵补全算法比任何单独一种在平均绝对误差MAE上都表现得要好。(2)提出一种基于互惠相似度的互惠推荐算法。首先,对男女用户的显式偏好和隐式偏好做了相关定义,并在此基础上给出了男女用户之间显式偏好相似度和隐式偏好相似度的计算方法;其次,针对显式偏好相似度和隐式偏好相似度在男女用户互惠推荐中作用的大小,分别赋予不同的权重因子,形成改进后的互惠相似度;最后,与当前互惠推荐中常用的两种推荐算法进行了实验对比,该算法在准确率、召回率和调和平均数方面,比另外两种算法都有明显地改善。
[Abstract]:With the development of e-commerce, more and more users to the user mode of two-way recommendation, the traditional personalized recommendation from the item to the user recommendation has been unable to meet the needs of users. This gave birth to the rapid development of reciprocal recommendations, as represented by online dating sites. Online dating sites provide a lot of dating information for the opposite sex who make friends every day. Job search websites not only provide a lot of job information for job seekers every day, but also provide a lot of resume information for the recruitment of people in enterprises. However, these information are complicated and lack of effective systematic classification. It is difficult to accurately locate the object of interest to both friends and job seekers and recruiters who are involved in the job recruitment process. Therefore, it is urgent to improve the reciprocal recommendation algorithm and improve the recommendation quality to the greatest extent. In this paper, a reciprocal recommendation algorithm for online dating is proposed, which improves the efficiency of the reciprocal recommendation algorithm from two aspects of matrix complement and reciprocal similarity, and fully considers the relationship between the two aspects. The reciprocal recommendation algorithm is studied deeply. The main work and contributions are as follows: (1) aiming at the problem of data sparsity in the reciprocal recommendation system represented by online dating, a matrix complement algorithm is proposed to reduce the data sparsity. Firstly, two kinds of matrix complement methods are deeply analyzed, secondly, the two methods are mixed weighted, and a matrix complement algorithm based on LMaFit and K-Means is proposed. The advantages of the low rank matrix complement LMaFit algorithm and the K-Means clustering algorithm are complementary. Finally, Experimental results show that the matrix complement algorithm based on LMaFit and K-Means is better than any single one on the average absolute error (MAE).) A reciprocal recommendation algorithm based on reciprocal similarity is proposed. Firstly, the explicit preference and implicit preference of male and female users are defined, and on this basis, the explicit preference similarity and implicit preference similarity between male and female users are calculated. According to the magnitude of explicit preference similarity and implicit preference similarity in the reciprocal recommendation of male and female users, different weight factors are assigned to form the improved reciprocal similarity. Finally, The experimental results are compared with the two commonly used recommendation algorithms in the reciprocal recommendation system. The proposed algorithm is much better than the other two algorithms in terms of accuracy recall and harmonic average.
【学位授予单位】：山东师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.3

【参考文献】