基于用户体验度和长尾理论的推荐算法

发布时间：2018-05-29 20:49

本文选题：推荐系统 + 用户偏好　；参考：《北方民族大学》2017年硕士论文

【摘要】：在信息数据骤增的时代,网络使用者们面临“信息繁多但有价值信息难以排查和筛取”的窘境。过滤技术是信息推荐系统中最为成功的技术,它诞生于20世纪九十年代,目前因为网络用户提供可靠性推荐、节省时间和方便快捷的体验已成为科研焦点。它能快速并主动预测用户的属性和兴趣方向,推送用户可能需要但又难以筛选、判断的相关信息资源。但日益增长的数据信息给传统的推荐算法带来了新的问题,庞大或稀疏的数据使得相似度计算不再精准,用户对推荐信息的满意程度下降,冷启动等问题相继产生。许多学者开始针对这些问题对传统算法进行研究与改进。本文在传统推荐算法基础上,提出一种基于用户体验度和长尾理论的推荐算法。首先阐述了用户体验度对用户产生的潜在影响,简单说明传统算法的计算公式并选择皮尔森相似性计算公式进行进一步计算。针对仅凭单一评分查找相似邻居判断有偏差这一问题,通过用户对项目的行为优评占总行为记录数及用户历史行为评分占总项目的比重产生权重,重点研究“用户评分矩阵和项目类别矩阵”,凭借长尾理论来平衡热门、冷门项目推荐概率。经实验验证得出算法有一定的改进。
[Abstract]:In the era of rapid increase of information data, network users are faced with the dilemma of "information is numerous but valuable information is difficult to search and sift". Filtering technology is the most successful technology in information recommendation system. It was born in the 1990s. At present, because network users provide reliable recommendation, save time and experience quickly has become the focus of scientific research. It can quickly and actively predict the user's properties and interest direction, push the user may need but difficult to filter, judge the relevant information resources. However, the growing data information brings new problems to the traditional recommendation algorithms. The huge or sparse data makes the similarity calculation not accurate, the users' satisfaction with the recommendation information drops, and the cold start up problems arise one after another. Many scholars begin to study and improve the traditional algorithms for these problems. Based on the traditional recommendation algorithm, this paper presents a recommendation algorithm based on user experience and long tail theory. In this paper, the potential influence of user experience on user is discussed, and the calculation formula of traditional algorithm is simply explained, and Pearson similarity formula is selected for further calculation. In order to solve the problem that there is a deviation in finding similar neighbors only by a single score, the weight is generated by the proportion of the users' behavior evaluation to the total behavior records and the proportion of the users' historical behavior scores to the total items. Focus on the "user rating matrix and item category matrix", with the long tail theory to balance the hot, cold project recommendation probability. Experimental results show that the algorithm has some improvement.
【学位授予单位】：北方民族大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.3

【参考文献】