结合用户消费水平的商品推荐算法研究

发布时间：2018-06-02 08:14

本文选题：协同过滤 + 用户消费水平　；参考：《东北师范大学》2016年硕士论文

【摘要】：近几年,电子商务这一新兴的购物模式伴随着互联网技术的飞速进步逐步兴起,成为在人群中十分风行的一种新型购物渠道。相较于传统的线下购物方式,电子商务具有非常多的优势,一方面由于具有庞大的网络用户,它可以给企业带来更多的营业利润,另一方面由于用户足不出户就可以随意浏览国内国外,各种各样琳琅满目的商品信息,电子商务可以带给使用者更加便捷更加舒适的消费体验。与此同时,我们依然不能忽视现有的电子商务推荐系统中显现出的一些弊端。用户往往在很多时候不能从极其庞大的商品海洋中准确地找到本人中意的商品。在这种情况下,电子商务针对用户的个性化推荐成为处理这一问题的一个非常有效之办法。目前各种各样有关于推荐方法的研究层出不穷。即便这样仍会存在诸如数据稀疏,冷启动,算法可扩展性差之类的难题。如何突破这些技术瓶颈成为现在研究中的重点和难点。协同过滤算法是在个性化推荐上使用的最为广泛的一项技术,目前基于协同过滤算法的研究主要是基于用户-项目评分这一角度来进行各种各样的改进。本文认为除了从评分这一方面来衡量用户间的相似性,还可以利用用户自身的一些因素来分析用户的购物习惯。因此,本文的设想是把现有协同过滤推荐的推荐过程与用户消费水平因素进行融合,认为具有不同消费水平的用户群具备不同的商品倾向性。本文利用用户背景信息以及购物记录建立用户的二级消费水平模型,对评分矩阵进行降维处理并对空缺项目评分预测评分值。然后结合用户消费水平和评分数据得到综合的用户相似性,从根据消费水平筛选后的用户集中确定目标用户的最近邻居集,最后在目标用户最近邻居集的基础上产生推荐项目集。本文在最后通过观察实验验证得到的数据结果,将本文的改进效果和传统的协同过滤进行对比,实验证明结合用户消费水平的改进推荐算法可以在传统推荐算法的基础上为用户更加准确的推荐倾向商品,并且在一定程度上缓解数据稀疏性问题以及新用户问题,对于电子商务推荐系统的改进有一定促进作用。
[Abstract]:In recent years, with the rapid progress of Internet technology, E-commerce, a new shopping mode, has become a popular new shopping channel in the crowd. Compared with the traditional offline shopping mode, e-commerce has many advantages. On the one hand, it can bring more business profits to enterprises because of its huge network users. On the other hand, because users can browse at home and abroad without leaving home, various kinds of commodity information, electronic commerce can bring users more convenient and more comfortable consumption experience. At the same time, we still can not ignore the existing e-commerce recommendation system in the emergence of some drawbacks. Users are often unable to find their favorite goods accurately from the vast ocean of goods. In this case, e-commerce personalized recommendation for users becomes a very effective way to deal with this problem. At present, there are a variety of recommendations for the study of the endlessly. Even so, there will still be problems such as sparse data, cold start, poor scalability of algorithms and so on. How to break through these technical bottlenecks has become the focus and difficulty of current research. Collaborative filtering algorithm is the most widely used technology in personalized recommendation. At present, the research based on collaborative filtering algorithm is mainly based on the user-item score to carry out a variety of improvements. This paper holds that apart from measuring the similarity of users from the aspect of score, we can also use some factors of users to analyze their shopping habits. Therefore, the assumption of this paper is to merge the current recommendation process of collaborative filtering recommendation with the factors of user consumption level, and consider that users with different consumption levels have different propensity of goods. Based on the user background information and shopping records, this paper establishes a two-level consumption level model for users, then reduces the dimension of the score matrix and predicts the score value of the vacancy items. Then combined with the user consumption level and the score data to get the comprehensive user similarity, from the user set selected according to the consumption level to determine the target user's nearest neighbor set. Finally, the recommended item set is generated on the basis of the nearest neighbor set of the target user. At the end of this paper, the improved results are compared with the traditional collaborative filtering. The experimental results show that the improved recommendation algorithm combined with the user consumption level can more accurately recommend products for the user on the basis of the traditional recommendation algorithm, and to some extent alleviate the problem of data sparsity and the problem of new users. It can promote the improvement of e-commerce recommendation system.
【学位授予单位】：东北师范大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.3

【相似文献】