基于用户特征数据的内容推荐方法的研究与应用

发布时间：2018-03-14 09:21

本文选题：内容推荐　切入点：用户特征数据　出处：《北方工业大学》2017年硕士论文　论文类型：学位论文

【摘要】：近年来,电子商务的快速发展使得信息过载问题越来越突出,为了解决这一问题各大电子商务网站越来越依赖于内容推荐系统。本文首先介绍了内容推荐方法在国内外的研究与应用现状,对电子商务中用户特征数据以及数据挖掘技术做出了研究。通过分析和比较现有的内容推荐技术,针对传统协同过滤推荐方法存在的"冷启动"、数据稀疏和"新兴趣""发现的问题提出了基于用户特征数据的改进方案。对于协同过滤推荐方法存在的"冷启动"问题,利用用户的属性特征数据,挖掘用户属性特征的相似带来的兴趣相似。首先根据新用户的属性特征,为新用户找到属性相似的用户。然后根据相似用户群的喜好,为用户推荐基于用户属性特征的商品。对于数据稀疏问题,本文通过分析用户的多种特征数据,得到用户的属性特征、喜好特征和评分特征。针对协同过滤推荐方法中相似度的计算进行改进。融合用户属性相似度的同时,并以用户的特征数据所表示的用户喜好权重来填充用户-项目评分矩阵。最后加权综合用户的属性相似度和优化后的评分相似度得到用户间的综合相似度。为用户找到相似邻居,通过协同过滤,实现推荐。针对不能发现惊喜商品的问题,引入关联规则数据挖掘技术,挖掘订单中商品的频繁项集,找到商品间的关联规则。通过结合用户的在线特征数据,将关联规则推荐结果进行个性化过滤,得到推荐。最后根据本文提出的推荐方法设计并实现了一个电子商务内容推荐系统,并应用到电子商务网站中进行实际检验。通过对推荐方法准确率、召回率和平均绝对误差的实验分析,证明了本文所述推荐方法的有效性。
[Abstract]:In recent years, the rapid development of electronic commerce makes the problem of information overload more and more prominent. In order to solve this problem, various e-commerce websites rely more and more on content recommendation system. Firstly, this paper introduces the research and application status of content recommendation methods at home and abroad. This paper makes a research on user characteristic data and data mining technology in electronic commerce. By analyzing and comparing the existing content recommendation technology, Aiming at the problems of "cold start", data sparsity and "new interest" found in traditional collaborative filtering recommendation methods, an improved scheme based on user characteristic data is proposed. Using the attribute feature data of the user, the similarity of interest caused by the similarity of the user attribute feature is mined. Firstly, according to the attribute feature of the new user, the user with the similar attribute is found for the new user, and then according to the preferences of the similar user group, For the problem of data sparsity, this paper analyzes the user's multi-feature data, and obtains the user's attribute feature. Preference feature and score feature. The similarity calculation in collaborative filtering recommendation method is improved. At the same time, the similarity of user attributes is fused. The user-item scoring matrix is filled with the user preference weight expressed by the user's characteristic data. Finally, the user's attribute similarity and the optimized rating similarity are weighted to obtain the user's comprehensive similarity. The family found a similar neighbor, Through collaborative filtering, recommendation is realized. In order to solve the problem of finding surprise items, association rules data mining technology is introduced to mine the frequent item set of goods in order to find the association rules between items. By combining the online characteristic data of users, the association rules between items are found. Finally, according to the recommendation method proposed in this paper, an E-commerce content recommendation system is designed and implemented. Through the experimental analysis of the accuracy, recall rate and average absolute error of the recommendation method, the validity of the proposed recommendation method is proved.
【学位授予单位】：北方工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.3

【参考文献】