基于条件相对平均熵的个性化推荐算法研究与应用

发布时间：2018-08-26 16:51

【摘要】：随着互联网的普及,以及电子商务的迅猛发展,网络交易日渐流行,越来越多的商品从现实交易转向虚拟交易,导致信息资源的数据种类和数据量迅速增长,推动了电子商务个性化推荐的研究与发展。目前,电子商务推荐的核心思想是基于各种相关关系,如用户关系、商品关系、用户与商品的关系。然而,当用户消费行为数据没有或很少,或用户共同的选择较少,或某商品不在历史行为数据中,相关关系将缺乏或不足,导致无法通过相似性进行预测,出现数据稀疏性问题或者冷启动问题,降低推荐准确性,难以恰当地向用户提供推荐服务。另外,用户的消费偏好和消费性格特点,对其消费行为有重要的影响,当商品的效用与用户的消费性格相符,用户才可能有消费行为,这为电子商务的个性化推荐提供了一个新的研究视角。因此,如何在减少或消除上述问题的同时,从海量的消费行为数据中挖掘出用户感兴趣或需要的商品,并精准的推荐给目标用户,已成为个性化推荐相关研究的热点。本文主要工作包括:(1)详细分析了个性化推荐算法和复杂网络社团结构发现算法及其特点。(2)考虑到当前个性化推荐系统对准确性的需求,选择具代表性的CNM个性化推荐算法,通过引入点权、边权、JSD距离计算项目相似度公式进行CNM算法优化和验证。(3)基于对用户消费性格的分析,引入条件互信息和条件相对平均熵来获取K2算法中初始节点输入次序,然后利用CH评分函数和后验概率函数对贝叶斯网络进行学习,分析出用户消费性格。(4)使用学习好的贝叶斯网络进行推理,判断用户待推荐域的商品与用户消费性格的联系,得到最终的商品推荐域。(5)给出了电信闲置资产市场化交易系统的需求分析与系统设计,并将本文研究成果应用于该系统的资产推荐模块。
[Abstract]:With the popularity of the Internet and the rapid development of electronic commerce, network transactions are becoming more and more popular, and more commodities are changing from real transactions to virtual transactions, which leads to the rapid growth of data types and data volume of information resources. It promotes the research and development of e-commerce personalized recommendation. At present, the core idea of E-commerce recommendation is based on a variety of related relationships, such as user relations, commodity relations, the relationship between users and commodities. However, when there is no or little consumer behavior data, or when users have fewer common choices, or when a commodity is not in the historical behavior data, the correlation will be lacking or insufficient, leading to the inability to predict by similarity. The problem of data sparsity or cold start reduces the accuracy of recommendation, and it is difficult to provide recommendation services to users properly. In addition, the consumer's consumption preference and characteristics have an important influence on the consumer behavior. When the utility of the commodity is in line with the consumer's consumption character, the user may have the consumer behavior. This provides a new perspective for the personalized recommendation of e-commerce. Therefore, how to reduce or eliminate the above problems, from the mass of consumer behavior data mining users interested or need goods, and accurately recommend to the target users, has become the focus of research on personalized recommendation. The main work of this paper is as follows: (1) the personalized recommendation algorithm and the discovery algorithm of complex network community structure and their characteristics are analyzed in detail. (2) considering the demand for accuracy of the current personalized recommendation system, the representative CNM personalized recommendation algorithm is selected. The CNM algorithm is optimized and verified by introducing point weight and edge weight distance similarity formula. (3) based on the analysis of consumer character, conditional mutual information and conditional relative average entropy are introduced to obtain the initial node input order in K2 algorithm. Then we use CH score function and posteriori probability function to study Bayesian network and analyze the consumer character of user. (4) using the well-learned Bayesian network to reason to judge the relationship between the products of the user to recommend domain and the consumer character. Finally, the final commodity recommendation domain is obtained. (5) the requirement analysis and system design of telecom idle assets market-oriented trading system are given, and the research results of this paper are applied to the asset recommendation module of the system.
【学位授予单位】：南昌大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.3

【相似文献】