基于隐式反馈的图书推荐系统设计与实现

发布时间：2018-04-27 05:25

本文选题：推荐系统 + 隐式反馈　；参考：《吉林大学》2017年硕士论文

【摘要】：在“互联网+”的时代背景下,个性化推荐系统为用户提供“私人订制”式的推荐服务以满足不同用户的消费需求。因个性化推荐系统具有互动性的特点,使其成为实现“互联网+”新型消费模式的重要手段。然而随着推荐服务规模越来越大,评分数据不足、用户-项目矩阵稀疏等问题愈发凸显,传统推荐算法面临难以突破的瓶颈。为了解决这一问题,不少研究学者开始更多地关注用户隐式行为的分析和研究,尝试从中挖掘用户兴趣偏好以弥补显式评分数据带来的不足。本文针对图书推荐系统的评分数据稀疏和单类协同过滤等问题展开研究,分析系统中数据的特点,将隐式反馈作为建立和更新用户兴趣模型的数据来源,创造性地提出一种改进算法——基于分层隐式反馈的贝叶斯排序算法,通过预测用户对项目的相对喜好来得到推荐列表。本文的研究工作可概括如下:首先,文章从系统中存在的显式和隐式反馈数据的特点与区别出发,分析本系统选取隐式用户反馈的原因以及隐式反馈带来的问题。然后从基于评分和基于排序两个方面介绍了几种经典的推荐算法。针对当下排序推荐算法往往只考虑用户有过操作行为的项目与未操作过的项目之间的偏好差别,缺乏对用户偏好程度深度理解和区分,本文探索用户隐式反馈所表示的用户偏好,将隐式反馈类型按所代表用户兴趣程度进行划分,使用分层隐式反馈模型来表示用户偏好。在贝叶斯个性化排序算法的基础上设计了“强正反馈-弱正反馈-负反馈”的三层隐式反馈表示模型,为用户生成符合用户兴趣的项目推荐列表。然后,经过大量实验研究分析潜因子矩阵维度k值大小、推荐列表长度等参数对算法性能的影响,并从AUC、平均精度均值、平均百分比排序、归一化折损累积增益等角度与经典的排序推荐算法作比较,对改进的算法进行评估,得出本文算法所生成的推荐列表可以更准确地把握用户当前兴趣。最后,在改进算法基础上,从分析个性化图书系统的功能需求入手,对系统进行功能模块划分与设计,实现了基于隐式反馈的个性化图书推荐系统。
[Abstract]:Under the background of "Internet", personalized recommendation system provides users with "private customized" recommendation services to meet the consumer needs of different users. Because of the interactive characteristic of personalized recommendation system, it becomes an important means to realize the new consumption mode of "Internet". However, with the increasing scale of recommendation service, insufficient score data and sparse user-item matrix, the traditional recommendation algorithm is facing the bottleneck that is difficult to break through. In order to solve this problem, many researchers begin to pay more attention to the analysis and research of implicit behavior of users, and try to mine user interest preference from it in order to make up for the deficiency caused by explicit rating data. In this paper, the sparse scoring data and single class collaborative filtering of book recommendation system are studied, the characteristics of the data in the system are analyzed, and the implicit feedback is used as the data source to establish and update the user interest model. An improved algorithm, Bayesian sorting algorithm based on hierarchical implicit feedback, is proposed in this paper. The list of recommendations is obtained by predicting the relative preferences of users to items. The research work of this paper can be summarized as follows: firstly, from the characteristics and differences of explicit and implicit feedback data in the system, this paper analyzes the reasons for selecting implicit user feedback and the problems caused by implicit feedback in the system. Then, several classical recommendation algorithms are introduced from two aspects: based on score and based on sorting. In order to solve the problem, the current ranking recommendation algorithm only considers the preference difference between the items that the user has operated on and the items that have not been operated, and lacks the deep understanding and distinction of the degree of user preference. In this paper, the user preferences expressed by implicit feedback are explored. The types of implicit feedback are divided according to the degree of interest represented by the users, and the hierarchical implicit feedback model is used to express user preferences. Based on Bayesian personalized sorting algorithm, a three-layer implicit feedback representation model of "strong positive feedback, weak positive feedback and negative feedback" is designed. Then, after a lot of experiments, we analyze the effect of the parameters of the latent factor matrix dimension k value, the recommended list length and other parameters on the performance of the algorithm, and rank it from AUC, average precision mean value, average percentage, etc. The normalized loss cumulative gain is compared with the classical ranking recommendation algorithm, and the improved algorithm is evaluated. It is concluded that the list of recommendations generated by this algorithm can more accurately grasp the current interests of users. Finally, on the basis of the improved algorithm, starting with the analysis of the functional requirements of the personalized book system, the functional modules of the system are divided and designed, and the personalized book recommendation system based on implicit feedback is realized.
【学位授予单位】：吉林大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.3

【参考文献】