社会网络中基于用户评分数据的协同推荐研究

发布时间：2018-03-31 06:20

本文选题：推荐系统　切入点：相似度计算　出处：《河南大学》2016年硕士论文

【摘要】：近些年来,随着信息技术的发展,电子商务领域积累了大量的数据,海量数据在为客户提供多样化商品选择的同时也带来了选择难的问题。推荐系统作为一种基于海量信息进行检索和筛选的个性化推送技术越来越受到人们的重视。它以客户的实际需求为中心,根据客户的历史行为数据预测用户的行为偏好,从而主动为客户提供个性化的数据信息服务。然而,在基于用户评分进行用户兴趣偏好建模的推荐系统中常常面临着一些不可避免的问题,如用户相似度计算精度问题、评分矩阵稀疏性问题。针对推荐系统中这两个常见的问题本文分别进行了探索,提出了一种新的计算用户相似度的计算方法,改善了传统相似度计算方法的不足;利用了物品的标签信息在一定程度上缓解了评分矩阵存在大量缺失值时不能很好的找出与目标用户具有相似偏好的用户的问题。具体来说,文中的贡献主要包含以下三个方面:(1)针对皮尔逊相似度在计算时遇到分母为0而相似度无法计算的情况,提出了一种改进方法,该方法能够根据用户评分情况适应选择合理的相似度计算方法。(2)基于mahout中已有的未知评分技术,结合评分项的标签信息提出了一种新的未知评分预测方法,该方法首先计算了用户对标签的兴趣分布,然后根据评分项包含标签的情况实现未知评分预测。(3)针对新提出的相似度计算方法和未知评分预测方法,结合mahout原有的方法进行了较好的集成。
[Abstract]:In recent years, with the development of information technology, the field of electronic commerce has accumulated a lot of data. Mass data not only provides customers with a variety of goods to choose, but also brings the problem of difficult choice. As a personalized push technology based on mass information retrieval and screening, recommendation system has been paid more and more attention to. It is centered on the actual needs of the customer, Based on the customer's historical behavior data, the user's behavior preference is predicted, thus providing personalized data information service for the customer. However, In the recommendation system of user interest preference modeling based on user score, there are some unavoidable problems, such as the accuracy of user similarity calculation. This paper explores the two common problems in recommendation system, and proposes a new method to calculate user similarity, which improves the deficiency of traditional similarity calculation method. The use of the label information of the items to some extent alleviates the problem of finding users with similar preferences to the target users when there are a large number of missing values in the scoring matrix. The contributions in this paper mainly include the following three aspects: 1) an improved method is proposed to solve the problem that Pearson's similarity can not be calculated when its denominator is zero. Based on the unknown scoring technology in mahout and the label information of scoring items, a new unknown score prediction method is proposed. Firstly, the distribution of users' interest in labels is calculated, and then unknown score prediction is realized according to the condition that labels are included in the scoring items. (3) for the new similarity calculation method and unknown score prediction method, Combined with the original method of mahout, better integration is carried out.
【学位授予单位】：河南大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.3

【相似文献】