基于组合算法的电子商务推荐系统的研究

发布时间：2018-03-11 17:38

本文选题：协同过滤　切入点：关联规则　出处：《西安工业大学》2017年硕士论文　论文类型：学位论文

【摘要】：互联网的兴起和发展,而后又带动了电子商务的飞速发展。电子商务将大量的数据带到了我们的面前。面对这些数据,用户却无法从中快速的获取有效的信息,这就形成了信息超载的问题。搜索引擎在解决这个问题上取得了一定的成功,例如美国的谷歌和国内的百度都是搜索引擎,当用户在搜索引擎中输入关键词时,搜索引擎就会在数据库中寻找用户可能需要的信息。然而,互联网的数据浩瀚如海,并非每一个用户都能够明确的知道自己的需求。相比之下,推荐系统则能更好的解决信息超载的问题,也显得更加的智能和主动。推荐系统不需要用户输入关键词,就能够在海量的数据信息中进行快速的搜索。另一方面,它会将用户可能感兴趣的信息主动呈现给用户,也会根据不同用户的不同兴趣为用户提供个性化推荐服务。因此,在很多领域中,都能够看到推荐系统熟悉的身影。尤其在电子商务网站中有越来越好的发展前景。推荐算法有多种,但目前主流的并且使用率最高的三个推荐算法分别是基于内容、基于协同过滤和基于关联规则的。其中,使用人数最多、市场占有率最高的推荐算法是第二种算法。然而随着互联网的快速发展,网站中用户数量和产品数量也在迅速的增长,基于协同过滤的推荐算法其也暴露出来越来越多亟待解决的问题。其中,最受关注的问题是数据稀疏性和冷启动。如何解决这些问题一直是研究者们的难题,为了使基于协同过滤的推荐算法有更加良好的推荐效果,本论文提出了利用组合算法来弥补这些缺陷。另外,当电子商务推荐系统中积累了很多数据时,单机处理这些数据受到了一定的影响,从而也会对推荐的准确度和效率产生影响。因此,对这些数据采用分布式处理,使用Hadoop技术实现推荐系统中大量数据的计算,提高计算的效率,将更加准确的商品推荐给用户,使用户越来越依赖电子商务推荐系统,也使得电子商务网站获得经济效益。本论文主要研究了以下几个方面:1)对几种常见的推荐算法做了详细的介绍,并描述了各种算法的思想和实现步骤,同时了解了各种推荐算法的缺点,尤其是协同过滤推荐系统的不足。2)使用组合算法来解决数据稀疏性和冷启动的问题。根据不同的条件,选用不同的推荐算法。3)使用Hadoop中的MapReduce框架实现推荐系统海量数据的分布式计算,使推荐系统有更好的性能。
[Abstract]:The rise and development of the Internet, and then the rapid development of e-commerce. Electronic commerce will bring a lot of data to us. In the face of these data, users can not quickly obtain effective information. This has led to the problem of information overload. Search engines have achieved some success in solving this problem. For example, Google in the United States and Baidu in China are both search engines. When users enter keywords in search engines, The search engine will search the database for the information users may need. However, the data on the Internet is so vast that not every user can clearly know their needs. Recommendation system can better solve the problem of information overload, also appear more intelligent and active. Recommendation system does not need user input keywords, can quickly search in the mass of data information, on the other hand, It will actively present to the user information that may be of interest to the user, and will also provide personalized referral services to the user according to the different interests of different users. Therefore, in many fields, Especially in e-commerce websites, there are more and more good prospects. There are many kinds of recommendation algorithms, but the three most popular and most popular recommendation algorithms are based on content. Among them, the recommendation algorithm, which has the most users and the highest market share, is the second one. However, with the rapid development of the Internet, the number of users and the number of products in the website are also growing rapidly. The recommendation algorithm based on collaborative filtering also exposes more and more problems that need to be solved urgently. Among them, the most concerned problems are data sparsity and cold startup. How to solve these problems has always been a difficult problem for researchers. In order to make the recommendation algorithm based on collaborative filtering have better recommendation effect, this paper proposes a combination algorithm to make up for these defects. In addition, when there is a lot of data accumulated in the E-commerce recommendation system, The processing of these data on a single machine is affected to some extent, which will also affect the accuracy and efficiency of recommendation. Therefore, distributed processing is used to process these data, and Hadoop technology is used to realize the calculation of a large number of data in recommendation system. Improve the efficiency of computing, more accurate products recommended to users, so that users rely more and more on e-commerce recommendation system, This paper mainly studies the following aspects: 1) introduce several common recommendation algorithms in detail, and describe the ideas and implementation steps of various algorithms. At the same time, we understand the shortcomings of various recommendation algorithms, especially the shortcomings of collaborative filtering recommendation system. 2) using combinatorial algorithms to solve the problem of data sparsity and cold start, according to different conditions, Different recommendation algorithms. 3) using the MapReduce framework in Hadoop to realize the distributed computing of mass data of recommendation system, so that the recommendation system has better performance.
【学位授予单位】：西安工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F713.36;TP391.3

【参考文献】