面向电商平台客户持续购买问题的情境化推荐模型研究
发布时间：2018-07-29 05:51
【摘要】：随着移动商务、情境感知、物联网的发展,电子商务的疆界被大大拓展,我们已经步入一个商务信息“大数据”时代。然而,电子商务中海量、无序的业务信息与客户需求之间的矛盾也日益凸显。在这种环境下,一方面电商平台获取新客户的成本急剧增加,另一方面电子商务企业想要维持和提高收益率,保留现有客户和提升客户持续性购买意愿变得非常迫切。以B2C为应用核心的电商平台积累了海量的数据,但是客户面临“信息丰富、但有用信息获取困难”的难题。如何根据客户的喜好、历史网络行为以及其他客户群体兴趣等信息,主动为客户提供符合其偏好的商品,提供个性化的信息推荐服务,从而激发客户持续性地网上浏览、购买等行为是电子商务平台面临的巨大挑战。个性化推荐方法作为客户在海量商务信息中获取偏好商品信息的重要手段,近年来受到了广泛的关注。但是,电商平台的客户兴趣具有复杂性,且购买行为受到情境影响后变得更加地不确定性与跳跃性。现有的个性化推荐服务未能很好的应对上述问题,导致客户不断流失。电商平台迫切需要准确高效地提供既符合客户内外情境,又满足客户偏好的信息服务,即提供情境化推荐来支持客户的持续性购买行为。为此,本论文“面向电商平台客户持续购买问题的情境化推荐模型”,研究对象为B2C平台的客户,研究范畴是客户的持续购买问题,以多维情境影响下的个性化推荐方法为手段,在分析电商客户情境的多样化、个性化以及动态变化等特征的基础上,以网络消费者决策行为理论、分布式认知理论与马斯洛需求层次理论等作为个性化推荐方法的理论基础,综合聚类、决策树、关联规则、马尔科夫、协同过滤、本体建模等方法研究电商平台客户持续购买问题的解决路径,并应用于电商平台不同阶段对客户持续购买的推荐服务中。主要研究工作如下：1.面向电商平台客户持续购买问题的情境化推荐模型研究电商平台客户持续购买问题针对的是已经在平台上购买过商品的老客户,他们的兴趣变化及购物行为可以归为两种情况：第一种客户的兴趣在一定时期内是稳定的,针对这一类型客户,本文提出构建分布式与差异化情境影响下的客户兴趣模型,然后利用情境化推荐方法完成商品的推送；第二种客户的兴趣由于电商多维度情境的变化产生了漂移(分为渐进式与突变式),针对这类型客户,本文建立动态兴趣模型并持续监测来适应客户兴趣的变化,利用自适应的情境化推荐方法完成商品的推送。本研究创新性的提出涵盖上述电子商务平台客户不同兴趣特征的情境化推荐模型,多维度分析电子商务中的情境与客户兴趣特征,建立一个融合情境、客户兴趣的个性化推荐知识模型,作为情境化推荐应用的知识支撑。2.基于客户敏感情境的个性化推荐方法研究针对电商平台兴趣未发生漂移的客户持续购买问题。传统推荐模型未能很好地考虑不同情境类型对客户需求的分布式和差异化影响,以及情境化推荐服务自适应性差等不足,提出了基于客户敏感情境的个性化推荐方法。该方法分析各种敏感情境类型及其具体实例对客户兴趣的差异化影响,设计一种基于分布式认知理论的客户兴趣提取算法;然后,结合分布式影响因子,提取出基于敏感情境认知的多维度情境客户兴趣。在上述客户兴趣提取的基础上,将提取出的敏感情境引入到协同过滤推荐过程,计算情境化客户兴趣之间的相似度,并设计一种融入情境相似度的改进协同过滤推荐算法。3.考虑客户兴趣渐进式漂移特征的情境化推荐方法研究针对电商平台兴趣渐进式漂移的客户持续购买问题,首先提出了基于改进型FP-Tree的关联规则算法,有效的提升了电子商务环境下客户兴趣规则模式挖掘的效率；其次,定义了客户的情境强度和情境关联度,并对其进行了量化处理。在此基础上,提出了融入情境贡献度的客户兴趣挖掘及漂移侦测算法,完成对情境贡献度影响下客户兴趣的建模与表达,并利用关联规则置信度与支持度的变化来对情境化客户偏好模式进行漂移侦测；最终,改进基于项目的协同过滤推荐算法,采用关联规则中项目的关系寻找候选项目集,且提出了将影响客户兴趣的情境贡献度代替评分以应对数据的稀疏性,提高了计算项目间相似度的准确性。4.考虑客户兴趣突变式漂移特征的情境化推荐方法研究针对电商平台兴趣突变式漂移的客户持续购买问题。考虑个性化推荐服务中存在难以有效适应外部情境与用户认知等心理因素变化带来的兴趣显著进化问题,提出了一个新的情境化推荐方法。首先,行为动机经典理论——“马斯洛需求层次理论”表明人的需求是会发生变化的,从而解释了人的兴趣会出现漂移原因。根据该原理设计了客户商品或者类别偏好、购买行为与客户需求层次的对应机制；然后,利用上述机制提出了基于本体与隐马尔科夫的客户兴趣层次判定算法,对客户兴趣进行表达与建模；其次,引入客户活跃度概念并提出融入情境的客户活跃度计算方法来解决推荐服务中的冷启动与稀疏性问题；最终,提出融入客户活跃度的动态协同过滤推荐算法,持续监测、学习客户兴趣变化规律,通过选择性扩充候选推荐内容,以及判定跳跃式兴趣趋势来主动适应突变式漂移问题。5.面向电商平台客户持续购买问题的情境化推荐应用研究将本研究提出的模型与方法应用于电子商务情境下客户持续购买问题中,设计推荐系统的体系框架并展开具体应用,通过实际数据来验证与分析本研究方法在某B2C电商平台客户持续购买问题中的推荐效果。最后,给出了若干提高电子商务个性化推荐质量及客户持续购买意愿的措施和建议,为电子商务企业的个性化推荐应用研究,以及客户保持提供了参考。
[Abstract]:With the development of mobile commerce, situational awareness, and the development of the Internet of things, the boundaries of e-commerce have been greatly expanded. We have entered a "big data" era of business information. However, the contradiction between the mass of the electronic commerce, the disordered business information and the customer needs has become increasingly prominent. In this environment, on the one hand, the e-commerce platform can obtain new customers. On the other hand, e-commerce enterprises want to maintain and improve the rate of return. It is very urgent to retain the existing customers and improve the customer's willingness to purchase continuously. With the B2C as the core of the e-commerce platform, it has accumulated a lot of data, but the customer is faced with the problem of "rich information, but difficult to obtain information." According to customer preferences, historical network behavior and other customer group interest information, it is a great challenge to stimulate customer's continuous online browsing and purchase by providing customers with their preferred goods and providing personalized information recommendation services. In recent years, the important means of obtaining the preference for commodity information in mass business information has attracted wide attention. However, the customer interest of the e-commerce platform is complex, and the purchase behavior has become more uncertain and hopping after the situation is affected. The existing personalized recommendation service has not been able to deal with the above problems well, leading to the customers. The e-commerce platform urgently needs to provide accurate and efficient information services that meet both the customer's internal and external situation and satisfy the customer's preference, that is, providing situational recommendation to support the customer's continuous purchase behavior. Therefore, this paper "the situational recommendation model for the continuous purchase of questions for e-commerce platform customers", the research object is B2C On the basis of analyzing the diversification, individuation and dynamic changes of e-commerce customers, the research category is the customer's continuous purchase problem and the personalized recommendation method under the influence of multi-dimensional situation. On the basis of the characteristics of the diversification, individuation and dynamic change of e-commerce customer situation, the theory of network consumer decision behavior, distributed cognition theory and Maslow's requirement hierarchy theory, etc. As the theoretical basis of personalized recommendation method, comprehensive clustering, decision tree, association rules, Markoff, collaborative filtering, ontology modeling, and other methods to study the solution path of customer continuous purchase problem of e-commerce platform customers, and applied to the recommendation service of customer continuous purchase at different stages of the e-commerce platform. The main research work is as follows: 1. power facing electricity. A situational recommendation model for the continuous purchase of business platform customers research on the continuous purchase of e-commerce platform customers is aimed at the old customers who have purchased goods on the platform. Their interest changes and shopping behavior can be classified into two situations: the first customer's interest is stable for a certain period of time, for this type of customer. In this paper, we propose a customer interest model under the influence of distributed and differentiated situations, and then use a situational recommendation method to complete the push of goods. The interest of the second customers is drifting due to the changes in the multi-dimensional context of e-commerce (divided into gradual and mutational type). In this paper, the dynamic interest model is established and held for this type of customer. Continuous monitoring to adapt to the change of customer interest, use the adaptive situational recommendation method to complete the push of goods. This research innovatively proposes a situational recommendation model which covers the different interest characteristics of the customers in the e-business platform, analyzes the situation and customer interest characteristics in the e-commerce, and establishes a fusion situation and the customer. The personalized recommendation knowledge model of interest, as the knowledge support of the situational recommendation application,.2. based on the personalized recommendation method based on the customer sensitive situation. The traditional recommendation model fails to consider the distributed and differentiation of different situation types to customer needs. In this way, a personalized recommendation method based on the sensitive situation of customers is proposed. This method analyzes the difference between the sensitive situation types and their specific instances, and designs a customer interest extraction algorithm based on the distributed cognitive theory. On the basis of the above customer interest extraction, the extracted sensitive situation is introduced into the collaborative filtering recommendation process, and the similarity between the situational customer interests is calculated, and an improved collaborative filtering recommendation algorithm.3. for situational similarity is designed, and an improved collaborative filtering algorithm is designed. The situational recommendation method, which concerns the progressive drift characteristics of customer interest, aims at the customer's continuous purchase problem for the gradual drift of the interest of e-commerce platform. First, a association rule algorithm based on improved FP-Tree is proposed, which effectively improves the efficiency of customer interest rule mining under the e-commerce environment. Secondly, the customer is defined. The situation intensity and context correlation degree are quantified. On this basis, the customer interest mining and drift detection algorithm which integrates the situation contribution degree is proposed, and the model and expression of customer interest are modeled and expressed under the influence of the situation contribution, and the situational customer preference pattern is used by the change of the confidence and support degree of the association rules. In the end, the collaborative filtering recommendation algorithm based on the project is improved, the relationship between the items in the association rules is used to find the candidate item set, and the situation contribution degree which affects the customer's interest is replaced by the score to deal with the sparsity of the data, and the accuracy of the similarity between the calculated items is improved by.4., which considers the mutation of the customer's interest. The situational recommendation method of the drift feature studies the problem of customer continuous purchase for the abrupt drift of the interest of the e-commerce platform. Considering the significant evolution of the interests that are difficult to adapt to the changes of the psychological factors such as the external situation and the user cognition in the personalized recommendation service, a new situational recommendation method is proposed. The "Maslow demand hierarchy theory" shows that human needs will change, which explains the cause of people's interest drift. According to this principle, we designed the customer's commodity or category preference, the corresponding mechanism of the purchase behavior and the customer demand level; then, the mechanism based on the above mechanism is put forward based on this mechanism. The customer interest level determination algorithm of body and hidden Markov is expressed and modeled for customer interest. Secondly, the concept of customer activity is introduced and the customer activity calculation method is put forward to solve the cold start and sparsity problems in the recommendation service. Finally, a dynamic collaborative filtering recommendation which is integrated into the customer activity is proposed. Algorithm, continuous monitoring, learning of changing rules of customer interest, selective extension of candidate recommendations, and determination of jumping interest trends to adapt to abrupt drift problem.5., a situational recommendation Application Research on e-commerce platform customers' continuous purchase problem, the model and method proposed in this study are applied to e-commerce situation. In the problem of customer continuing purchase, the system framework of the recommendation system is designed and the specific application is carried out. Through the actual data, the recommendation effect of this research method in the continuous purchase problem of a B2C e-commerce platform customer is verified and analyzed. Finally, some measures to improve the personalized recommendation quality of electronic commerce and the continuous purchase intention of customers are given. And suggestions for e-commerce enterprises personalized recommendation application research and customer retention provides a reference.
【学位授予单位】：浙江工商大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：F724.6
[Abstract]:With the development of mobile commerce, situational awareness, and the development of the Internet of things, the boundaries of e-commerce have been greatly expanded. We have entered a "big data" era of business information. However, the contradiction between the mass of the electronic commerce, the disordered business information and the customer needs has become increasingly prominent. In this environment, on the one hand, the e-commerce platform can obtain new customers. On the other hand, e-commerce enterprises want to maintain and improve the rate of return. It is very urgent to retain the existing customers and improve the customer's willingness to purchase continuously. With the B2C as the core of the e-commerce platform, it has accumulated a lot of data, but the customer is faced with the problem of "rich information, but difficult to obtain information." According to customer preferences, historical network behavior and other customer group interest information, it is a great challenge to stimulate customer's continuous online browsing and purchase by providing customers with their preferred goods and providing personalized information recommendation services. In recent years, the important means of obtaining the preference for commodity information in mass business information has attracted wide attention. However, the customer interest of the e-commerce platform is complex, and the purchase behavior has become more uncertain and hopping after the situation is affected. The existing personalized recommendation service has not been able to deal with the above problems well, leading to the customers. The e-commerce platform urgently needs to provide accurate and efficient information services that meet both the customer's internal and external situation and satisfy the customer's preference, that is, providing situational recommendation to support the customer's continuous purchase behavior. Therefore, this paper "the situational recommendation model for the continuous purchase of questions for e-commerce platform customers", the research object is B2C On the basis of analyzing the diversification, individuation and dynamic changes of e-commerce customers, the research category is the customer's continuous purchase problem and the personalized recommendation method under the influence of multi-dimensional situation. On the basis of the characteristics of the diversification, individuation and dynamic change of e-commerce customer situation, the theory of network consumer decision behavior, distributed cognition theory and Maslow's requirement hierarchy theory, etc. As the theoretical basis of personalized recommendation method, comprehensive clustering, decision tree, association rules, Markoff, collaborative filtering, ontology modeling, and other methods to study the solution path of customer continuous purchase problem of e-commerce platform customers, and applied to the recommendation service of customer continuous purchase at different stages of the e-commerce platform. The main research work is as follows: 1. power facing electricity. A situational recommendation model for the continuous purchase of business platform customers research on the continuous purchase of e-commerce platform customers is aimed at the old customers who have purchased goods on the platform. Their interest changes and shopping behavior can be classified into two situations: the first customer's interest is stable for a certain period of time, for this type of customer. In this paper, we propose a customer interest model under the influence of distributed and differentiated situations, and then use a situational recommendation method to complete the push of goods. The interest of the second customers is drifting due to the changes in the multi-dimensional context of e-commerce (divided into gradual and mutational type). In this paper, the dynamic interest model is established and held for this type of customer. Continuous monitoring to adapt to the change of customer interest, use the adaptive situational recommendation method to complete the push of goods. This research innovatively proposes a situational recommendation model which covers the different interest characteristics of the customers in the e-business platform, analyzes the situation and customer interest characteristics in the e-commerce, and establishes a fusion situation and the customer. The personalized recommendation knowledge model of interest, as the knowledge support of the situational recommendation application,.2. based on the personalized recommendation method based on the customer sensitive situation. The traditional recommendation model fails to consider the distributed and differentiation of different situation types to customer needs. In this way, a personalized recommendation method based on the sensitive situation of customers is proposed. This method analyzes the difference between the sensitive situation types and their specific instances, and designs a customer interest extraction algorithm based on the distributed cognitive theory. On the basis of the above customer interest extraction, the extracted sensitive situation is introduced into the collaborative filtering recommendation process, and the similarity between the situational customer interests is calculated, and an improved collaborative filtering recommendation algorithm.3. for situational similarity is designed, and an improved collaborative filtering algorithm is designed. The situational recommendation method, which concerns the progressive drift characteristics of customer interest, aims at the customer's continuous purchase problem for the gradual drift of the interest of e-commerce platform. First, a association rule algorithm based on improved FP-Tree is proposed, which effectively improves the efficiency of customer interest rule mining under the e-commerce environment. Secondly, the customer is defined. The situation intensity and context correlation degree are quantified. On this basis, the customer interest mining and drift detection algorithm which integrates the situation contribution degree is proposed, and the model and expression of customer interest are modeled and expressed under the influence of the situation contribution, and the situational customer preference pattern is used by the change of the confidence and support degree of the association rules. In the end, the collaborative filtering recommendation algorithm based on the project is improved, the relationship between the items in the association rules is used to find the candidate item set, and the situation contribution degree which affects the customer's interest is replaced by the score to deal with the sparsity of the data, and the accuracy of the similarity between the calculated items is improved by.4., which considers the mutation of the customer's interest. The situational recommendation method of the drift feature studies the problem of customer continuous purchase for the abrupt drift of the interest of the e-commerce platform. Considering the significant evolution of the interests that are difficult to adapt to the changes of the psychological factors such as the external situation and the user cognition in the personalized recommendation service, a new situational recommendation method is proposed. The "Maslow demand hierarchy theory" shows that human needs will change, which explains the cause of people's interest drift. According to this principle, we designed the customer's commodity or category preference, the corresponding mechanism of the purchase behavior and the customer demand level; then, the mechanism based on the above mechanism is put forward based on this mechanism. The customer interest level determination algorithm of body and hidden Markov is expressed and modeled for customer interest. Secondly, the concept of customer activity is introduced and the customer activity calculation method is put forward to solve the cold start and sparsity problems in the recommendation service. Finally, a dynamic collaborative filtering recommendation which is integrated into the customer activity is proposed. Algorithm, continuous monitoring, learning of changing rules of customer interest, selective extension of candidate recommendations, and determination of jumping interest trends to adapt to abrupt drift problem.5., a situational recommendation Application Research on e-commerce platform customers' continuous purchase problem, the model and method proposed in this study are applied to e-commerce situation. In the problem of customer continuing purchase, the system framework of the recommendation system is designed and the specific application is carried out. Through the actual data, the recommendation effect of this research method in the continuous purchase problem of a B2C e-commerce platform customer is verified and analyzed. Finally, some measures to improve the personalized recommendation quality of electronic commerce and the continuous purchase intention of customers are given. And suggestions for e-commerce enterprises personalized recommendation application research and customer retention provides a reference.
【学位授予单位】：浙江工商大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：F724.6
【参考文献】
相关期刊论文 前4条
1 叶作亮;王雪乔;宝智红;陈滨桐;;C2C环境中顾客重复购买行为的实证与建模[J];管理科学学报;2011年12期
2 张s，
本文编号：2151730
本文链接：https://www.wllwen.com/shoufeilunwen/jjglbs/2151730.html
