基于DBN分类的协同过滤推荐算法研究

发布时间：2018-01-27 03:50

本文关键词： 推荐系统多属性生命周期 DBN 覆盖率　出处：《新疆大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着数字信息化时代的到来,类似于淘宝、京东、亚马逊等各大网络电商的数量与日俱增,电子商务个性化推荐系统亦成为了研究和应用的热门领域。统计大量研究结果显示,目前现有研究方法都是从用户角度出发进行预测和推荐。与传统的视屏或电影推荐不同,电子商务个性化推荐系统不仅要注重用户体验,同时也要注重商家盈利状态,因此对于新项目的推荐、项目推荐的覆盖率及多样性成为商家关注的焦点。如何在现有方法的基础上从商家角度出发研究出高质量、高性能的推荐技术就显得尤其重要。首先,本文提出了基于用户多属性的协同过滤推荐算法(UMACF),该方法从用户的评分、评论、等级及区域多因素计算预测评分值,将预测结果和基于用户的协同过滤推荐算法结合后进行推荐。实验结果表明:(1)在用户的评分、评论、等级及区域4因素中,评分和评论是最影响预测评分值的因素;(2)与传统的协同过滤推荐算法相比,UMACF推荐算法的预测评分准确度提高近10%;与UARCF推荐算法相比,UMACF推荐算法的预测评分准确度提高近5%。其次,本文提出了基于用户多属性和项目生命周期的推荐算法(UAIL),该方法根据评分、评论、等级、区域、用户评论时间和项目发布时间信息使用销售量增长率分析法和商家盈利方式构建了基于项目生命周期的推荐模型,将该推荐模型和UMACF推荐算法的预测评分值相结合后进行推荐。实验结果表明:与UARCF推荐算法相比,覆盖率提高近28%,推荐新项目的新颖度提高近40%。最后,本文提出了基于DBN分类的协同过滤推荐算法研究(DBNCF),该方法使用DBN网络进行学习分类,将分类的结果和UAIL推荐算法的项目生命周期模型结合形成基于DBN分类的项目生命周期推荐模型,将该模型和UMACF推荐算法的预测评分值相结合后进行推荐。实验结果表明:(1)与UAIL推荐算法相比,DBNCF推荐算法的覆盖率提高5%,推荐新项目的新颖度提高近10%。(2)在时间耗能方面,UserCF、UARCF、UMACF和UAIL推荐算法时间消耗较为相近;与这四种推荐算法相比,DBNCF推荐算法需花费大量时间学习,因此该算法的时间消耗呈指数型增长。
[Abstract]:With the arrival of the digital information age, similar to Taobao, JingDong, Amazon and other major network e-commerce number is increasing day by day. E-commerce personalized recommendation system has also become a hot area of research and application. At present, the existing research methods are from the perspective of users to predict and recommend. Unlike traditional video or film recommendation, e-commerce personalized recommendation system should not only focus on user experience. At the same time, we should also pay attention to the status of business profitability, so for the new project recommendation. The coverage and diversity of project recommendations have become the focus of attention. How to develop high quality and high performance recommendation technology based on existing methods is particularly important. First of all. In this paper, we propose a collaborative filtering recommendation algorithm based on user multi-attribute, which calculates the prediction score from user rating, comment, rank and region multi-factor. The prediction results are combined with the user-based collaborative filtering recommendation algorithm. The experimental results show that: 1) in the user's rating, comment, rating and area of four factors. Scores and comments were the most important factors affecting the predicted scores. Compared with the traditional collaborative filtering recommendation algorithm, the prediction accuracy of UMACF recommendation algorithm is improved by nearly 10%. Compared with the UARCF recommendation algorithm, the prediction accuracy of UMACF recommendation algorithm is improved by nearly 5. Secondly, this paper proposes a recommendation algorithm based on user multi-attribute and project life cycle. According to the rating, comment, rating, region, user comment time and project release time information, the method constructs a recommendation model based on project life cycle using the sales growth rate analysis method and the business profit method. The proposed recommendation model is combined with the prediction score of the UMACF recommendation algorithm. The experimental results show that compared with the UARCF recommendation algorithm, the coverage rate is increased by nearly 28%. Finally, this paper proposes a collaborative filtering recommendation algorithm based on DBN classification, which uses DBN network for learning classification. Combining the result of classification with the project life cycle model of UAIL recommendation algorithm, the project life cycle recommendation model based on DBN classification is formed. The model is combined with the prediction score of the UMACF recommendation algorithm. The experimental results show that the coverage of the UMACF recommendation algorithm is 5% higher than that of the UAIL recommendation algorithm. The time consumption of user CFS UARCF UMACF and UAIL recommendation algorithm is similar to that of UAIL recommendation algorithm. Compared with these four recommendation algorithms, it takes a lot of time to learn, so the time consumption of the proposed algorithm increases exponentially.
【学位授予单位】：新疆大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.3

【参考文献】