基于多属性评分的电子商务个性化推荐算法研究

发布时间：2018-04-02 15:08

本文选题：多属性　切入点：推荐系统　出处：《江西财经大学》2016年硕士论文

【摘要】：在21世纪的今天,信息爆炸的时代里,每个人面对的信息已经数以亿计了,尤其是在电子商务网站,用户如何找到自己感兴趣的信息,已经不单单局限于自己去寻找,更需要电子商务推荐系统帮助用户发现他感兴趣的信息。所以推荐系统的研究领域变得越来越重要,它可以为用户推荐最感兴趣的信息。目前,大部分的推荐系统是通过用户对产品的评价信息进行个性化推荐的,这些显式或隐式的评价信息被表示成用户对被评分项目在单一维度上的偏好等级,这种单一维度上的评分信息不能有效表达用户对某个产品的各个方面的偏好程度的差异性,进而影响推荐算法的推荐性能。针对传统基于单一维度评分的推荐算法的不足,基于多属性评分的推荐系统考虑用户对产品各个方面评价信息的差异进行个性化推荐决策,本文的主要工作包括如下几个方面。首先,在考虑酒店的多个属性评分的基础上,对改进的基于多属性评分的协同过滤推荐算法的三种方法进行实证分析比较它们的准确性和多样性,这三种方法分别是:基于多属性相似性平均协同过滤、基于多维距离的协同过滤算法和基于层次分析法(AHP)协同过滤算法。其次,再分析酒店多维度评分信息后,把线性规划模型引入到求各个属性的权重中来,提出了基于多属性线性规划的协同过滤算法。再次,如何衡量一个推荐系统好坏的指标有很多,本文主要探讨比较准确性和多样性这两个常用的评价指标,虽然推荐准确度无疑是重要的,但学者日益认识到更高的准确性并不总是意味着对用户有用,也许用户希望所推荐的商品具有多样性。因此,除了分析准确度之外,本论文还考虑了另一个重要的指标多样性来衡量推荐的质量,并探讨了准确性和多样性之间的关系。最后,为了实证和验证所提出的方法,我们通过在酒店网站收集真实用户数据。收集到了 165829位用户对酒店的多属性评分(分别是性价比评分、舒适度评分、位置评分、卫生评分、睡眠评分、服务评分)。平均绝对误差(MAE)以及多样性被用来衡量算法表现。实验结果显示我们提出的改进方法在多属性环境下可以显著的提高推荐准确性和用户多样性。
[Abstract]:Today in the 21st century, in the era of information explosion, everyone has faced hundreds of millions of information, especially in e-commerce sites, how users find their own interested in information, is not limited to their own search,E-commerce recommendation system is needed to help the user to find the information he is interested in.Therefore, the research field of recommendation system is becoming more and more important, it can recommend the most interesting information for users.At present, most recommendation systems make personalized recommendation through users' evaluation information of products. These explicit or implicit evaluation information are expressed as the user's preference level on a single dimension for the rated items.This kind of rating information on a single dimension can not effectively express the difference of the user's preference degree to each aspect of a product, and then affect the recommendation performance of the recommendation algorithm.In view of the shortcomings of the traditional recommendation algorithm based on single dimension rating, the recommendation system based on multi-attribute scoring takes into account the users' differences in evaluation information in all aspects of the product. The main work of this paper includes the following aspects.First of all, on the basis of considering the multiple attributes of the hotel, three improved collaborative filtering recommendation algorithms based on multi-attribute scoring are analyzed and compared with each other in terms of their accuracy and diversity.The three methods are: average collaborative filtering based on multi-attribute similarity, collaborative filtering algorithm based on multi-dimension distance and collaborative filtering algorithm based on analytic hierarchy process (AHP).Secondly, after analyzing the multi-dimension rating information of hotel, the linear programming model is introduced to calculate the weight of each attribute, and a collaborative filtering algorithm based on multi-attribute linear programming is proposed.Thirdly, there are a lot of indicators to measure the quality of a recommendation system. This paper mainly discusses the two commonly used evaluation indicators, namely, accuracy and diversity, although recommendation accuracy is undoubtedly important.But scholars increasingly realize that higher accuracy does not always mean that it is useful to users, who may want a variety of products to recommend.Therefore, in addition to analytical accuracy, this paper also considers another important indicator diversity to measure the quality of recommendations, and discusses the relationship between accuracy and diversity.Finally, in order to demonstrate and verify the proposed method, we collect real user data on the hotel website.A total of 165829 users rated the hotel for multiple attributes (performance-to-price score, comfort score, location score, health score, sleep score, service score, etc.).The mean absolute error (mae) and diversity are used to measure the performance of the algorithm.The experimental results show that the proposed method can significantly improve the accuracy of recommendation and user diversity in multi-attribute environment.
【学位授予单位】：江西财经大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：F724.6

【参考文献】