推荐系统的研究及其在移动电子商务中的应用

发布时间：2018-06-29 05:46

本文选题：协同过滤 + 时间上下文信息　；参考：《电子科技大学》2015年硕士论文

【摘要】：随着社会的计算机化,人们生产和收集数据的能力显著增强。海量信息在给这个时代带来机遇的同时也带来了许多棘手的难题,那就是信息的筛选和过滤。推荐系统是一种智能的信息筛选工具:系统自动记录并分析用户的历史行为数据,然后将用户最感兴趣的信息呈现在用户面前。自推荐系统问世以来,被广泛地运用于在线电子商务系统中,同时也面临许多问题,例如以往算法未将时间这一动态信息建模到推荐模型中,从而不能捕获到与时间相关的各种重要规律,这称为静态推荐。为了解决这一问题,本文首先对推荐领域各类算法进行深入分析;然后,重点探讨了时间这一动态上下文信息对推荐结果的影响,并以协同过滤推荐算法为基础,将时间信息建模到该算法中,并给出具体的算法流程;最后结合移动设备特点,提出融合了移动设备特点的推荐方案并应用于在线图书销售系统中。本文的主要研究工作如下:(1)对目前主流的推荐算法进行深入研究,分析了各个算法的推荐机制以及优缺点,并通过简单实例对各算法进行了描述。重点分析了协同过滤推荐,详细描述了该类算法的执行流程;(2)通过分析现实生活推荐机制中的时间规律以及Netflix数据集中的时间现象,论证了时间信息对推荐系统的重要性;(3)对时间信息进行建模,提出了基于用户影响度的算法(IOE-User CF)以及基于物品耦合度及流行度的算法(PC-Item CF),并给出具体的流程;然后利用Netflix数据集从MAE值、推荐的准确率、召回率以及F1值等指标对两类算法的推荐质量进行评测。通过实验得出:本文融合了时间信息的算法比以往算法具有更好的推荐准确度;(4)最后,利用本文的算法思路,同时结合移动设备能方便地获取用户通讯录这一优点,建立融合移动设备特点的动态推荐模型。(5)为了论证本文的推荐模型具有一定的使用价值,最后设计一个简单的在线图书销售系统,并将上述模型应用该系统中。
[Abstract]:With the computerization of society, people's ability to produce and collect data has increased significantly. Mass information not only brings opportunities to this era, but also brings a lot of difficult problems, that is, information screening and filtering. Recommendation system is an intelligent information filtering tool: the system automatically records and analyzes the user's historical behavior data, and then presents the most interesting information to the user. Since the advent of the recommendation system, it has been widely used in the online e-commerce system, but also faces many problems, such as the previous algorithm did not model this dynamic information of time into the recommendation model. This does not capture the time-related important laws, this is called static recommendation. In order to solve this problem, this paper firstly analyzes all kinds of algorithms in recommendation field, and then discusses the influence of time, a dynamic context information, on the recommendation results, based on collaborative filtering recommendation algorithm. The time information is modeled in the algorithm, and the specific algorithm flow is given. Finally, combining the characteristics of mobile devices, a recommendation scheme is proposed and applied to the online book sales system. The main research work of this paper is as follows: (1) deeply study the current mainstream recommendation algorithms, analyze the recommendation mechanism, advantages and disadvantages of each algorithm, and describe each algorithm through a simple example. The collaborative filtering recommendation is analyzed in detail. (2) by analyzing the time rule in the real life recommendation mechanism and the time phenomenon in Netflix data set, The importance of time information to recommendation system is demonstrated. (3) the time information is modeled, and the algorithm based on user's influence degree (IOE-User CF) and the algorithm based on item coupling and popularity (PC-Item CF) are proposed, and the concrete flow is given. Then the recommendation quality of the two algorithms is evaluated by Netflix data set from the mae value, recommendation accuracy rate, recall rate and F1 value. The experimental results show that the proposed algorithm has better recommendation accuracy than the previous algorithms. (4) finally, using the algorithm of this paper, combined with the advantages of mobile devices can easily obtain the user address book. In order to prove that the recommendation model in this paper has some practical value, a simple online book sales system is designed and applied in this system. (5) A dynamic recommendation model combining the characteristics of mobile devices is established. (5) in order to prove that the recommendation model of this paper has some practical value, a simple online book sales system is designed.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：F724.6;TP391.3

【参考文献】