在线推荐系统的算法研究及其应用

发布时间：2018-08-04 16:24

【摘要】：近年来，随着互联网迅速发展及智能手机、相机等个人电子设备的迅速普及，大量信息被上载到互联网。互联网信息的快速增长，产生了信息疲劳的问题。人们每天面对海量数据，却难以从这些数据中快速获取自身需要的信息，而往往被各类垃圾信息所包围。人们迫切需要一种方式能够主动地、便利地、准确地获取并处理信息。推荐系统正是解决这一问题的最新尝试之一，并成为近几年的研究的热点。这种系统通过主动或是被动地获取用户的各类信息，如对项目的评分，，或是访问的网页、搜索的关键词、点击的链接等数据，主动推断用户的偏好，并向用户提供用户需要的信息。然而，现有推荐系统在精确度、功能、适用范围等方面仍然不能完全满足用户的需求，新的推荐系统及推荐算法仍在不断涌现。基于上述背景，本文对推荐系统的研究现状及发展进行了研究，重点关注推荐系统的评价以及基于社交网络的推荐系统两个方面，并提出了一些改进算法，取得了一定成果。本文主要包括以下内容：一是研究了推荐系统的评价方法，分析了现有方法中存在的误差及其对系统性能的影响。提出了一种基于分离因素法的推荐精度评价算法，并对算法进行了分析。基于KDD Cup2012Track2数据对相关算法进行了仿真实验，得到了搜索引擎中显示在不同位置广告，对用户相对吸引力数值。二是研究了在社交网络中推荐系统的作用及实现方法，并提出一种信任传递及聚合机制，用于度量社交网络中原本不是好友的两人之间的相关程度。同时，给出一种算法快速计算一个用户与其它用户之间的相关程度，进而快速给出推荐结果。基于KDD Cup2012Track1数据对算法进行了仿真实验，并给出了分析结果。
[Abstract]:In recent years, with the rapid development of the Internet and the rapid popularity of personal electronic devices such as smartphones and cameras, a large number of information has been uploaded to the Internet. The rapid growth of Internet information has caused the problem of information fatigue. People face huge amounts of data every day, but it is difficult to get the information they need quickly from these data, and they are often surrounded by all kinds of junk information. There is an urgent need for a way to access and process information actively, conveniently and accurately. Recommendation system is one of the latest attempts to solve this problem, and has become a hot research topic in recent years. This system actively infer user preferences by actively or passively obtaining user information, such as scoring items, visiting web pages, searching keywords, clicking links, and so on. And provide users with the information they need. However, the existing recommendation systems still can not meet the needs of users in the aspects of accuracy, function and application scope, and new recommendation systems and recommendation algorithms are still emerging. Based on the above background, this paper studies the status quo and development of recommendation system, focusing on the evaluation of recommendation system and the recommendation system based on social network, and puts forward some improved algorithms, which have achieved some results. The main contents of this paper are as follows: firstly, the evaluation methods of recommendation system are studied, and the errors in the existing methods and their effects on the system performance are analyzed. A recommendation accuracy evaluation algorithm based on separation factor method is proposed, and the algorithm is analyzed. Based on the KDD Cup2012Track2 data, the relevant algorithms are simulated, and the relative attractiveness value of the ads displayed in different locations in the search engine is obtained. Secondly, the function and implementation of recommendation system in social network are studied, and a mechanism of trust transfer and aggregation is proposed to measure the correlation between two people who are not friends in social network. At the same time, an algorithm is presented to calculate the correlation between one user and other users, and then the recommendation results are given quickly. The algorithm is simulated based on KDD Cup2012Track1 data, and the analysis results are given.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP391.3

【共引文献】