基于总剩余最大化和物品上下文约束的协同推荐算法研究
发布时间：2018-11-22 18:38
【摘要】：电子商务中产生越来越多的产品和交易信息,使得用户快速找到自己想要的产品变得越来越困难。同时,电子商务企业也面临着如何推荐让用户满意的产品从而提高销售量的问题。电子商务推荐系统就是为了解决这些问题而产生的。协同过滤这类推荐技术更多的是基于用户产生的数据直接进行分析,很少基于经济行为来进行电子商务的推荐。本文结合经济学中的总剩余最大化(Total Surplus Maximization)和物品上下文约束,提出了基于总剩余最大化和物品上下文约束的协同推荐模型。首先,根据用户的购买记录情况,综合用户的关联购买和时间局部性,构建物品相似度矩阵;其次,通过矩阵分解和物品上下文约束构建用户的个性化效用模型,并根据最后一单元零剩余法则,构造用户效用的目标函数,使用消费数据训练得到用户的个性化效用预测模型;最后根据总剩余最大化模型(TSM),得到用户对物品的购买预测模型,使消费者利益和生产者利益总和达到最大。本文通过引入物品的上下文约束,缓解消费记录矩阵的稀疏问题,更好的预测用户的个性化效用,最终得到更好的购买行为预测。论文在TaFeng超市销售数据集上进行实验对比和结果分析。从而得出结论,我们的模型与过去相关的协同过滤和基本的TSM算法相比,在稀疏数据集上取得了更好的推荐效果。
[Abstract]:More and more products and transaction information are generated in electronic commerce, which makes it more and more difficult for users to find the products they want quickly. At the same time, e-commerce enterprises are faced with the problem of how to recommend products that satisfy users to increase sales volume. E-commerce recommendation system is to solve these problems. Collaborative filtering is more based on the analysis of user-generated data, and rarely on the basis of economic behavior to recommend e-commerce. In this paper, a collaborative recommendation model based on total residual maximization and article context constraint is proposed, which combines the total surplus maximization (Total Surplus Maximization) and the article context constraint in economics. Firstly, according to the user's purchase record, the article similarity matrix is constructed by synthesizing the associated purchase and time locality of the user. Secondly, the user's personalized utility model is constructed by matrix decomposition and article context constraint, and the objective function of user's utility is constructed according to the zero residue rule of the last unit. The user's personalized utility prediction model is obtained by using consumer data training. Finally, according to the total surplus maximization model (TSM), the prediction model of the consumer's purchase of the goods is obtained, which makes the sum of the consumer's interest and the producer's interest maximum. By introducing the contextual constraints of items, this paper alleviates the problem of sparse consumption record matrix, better predicts the personalized utility of users, and finally obtains a better prediction of purchasing behavior. This paper carries on the experiment contrast and the result analysis on the TaFeng supermarket sales data set. It is concluded that compared with the previous collaborative filtering and basic TSM algorithms, our model achieves better recommendation results on sparse datasets.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.3
本文编号：2350210
[Abstract]:More and more products and transaction information are generated in electronic commerce, which makes it more and more difficult for users to find the products they want quickly. At the same time, e-commerce enterprises are faced with the problem of how to recommend products that satisfy users to increase sales volume. E-commerce recommendation system is to solve these problems. Collaborative filtering is more based on the analysis of user-generated data, and rarely on the basis of economic behavior to recommend e-commerce. In this paper, a collaborative recommendation model based on total residual maximization and article context constraint is proposed, which combines the total surplus maximization (Total Surplus Maximization) and the article context constraint in economics. Firstly, according to the user's purchase record, the article similarity matrix is constructed by synthesizing the associated purchase and time locality of the user. Secondly, the user's personalized utility model is constructed by matrix decomposition and article context constraint, and the objective function of user's utility is constructed according to the zero residue rule of the last unit. The user's personalized utility prediction model is obtained by using consumer data training. Finally, according to the total surplus maximization model (TSM), the prediction model of the consumer's purchase of the goods is obtained, which makes the sum of the consumer's interest and the producer's interest maximum. By introducing the contextual constraints of items, this paper alleviates the problem of sparse consumption record matrix, better predicts the personalized utility of users, and finally obtains a better prediction of purchasing behavior. This paper carries on the experiment contrast and the result analysis on the TaFeng supermarket sales data set. It is concluded that compared with the previous collaborative filtering and basic TSM algorithms, our model achieves better recommendation results on sparse datasets.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.3
【参考文献】
相关期刊论文 前2条
1 高全力;高岭;杨建锋;王海;;上下文感知推荐系统中基于用户认知行为的偏好获取方法[J];计算机学报;2015年09期
2 薛福亮;马莉;;利用动态产品分类树改进的关联规则推荐方法[J];计算机工程与应用;2016年04期
相关硕士学位论文 前1条
1 蔡瑞瑜;基于社会上下文约束和物品上下文约束的协同推荐[D];浙江大学;2012年
，本文编号：2350210
本文链接：https://www.wllwen.com/jingjilunwen/dianzishangwulunwen/2350210.html
