基于社会标注复杂网络的电子商务推荐方法研究

E-Commerce recommender helps users to find the commodities easily. And it not only improves quality of E-Commerce service, but also increases purchase of commodities. E-Commerce recommendation is an important research task in E-Commerce and customer relationship management. Recent years, as a kind of new application forms, social tagging network can describe the text based products, and it also describes the image, sound, video, etc, which is difficultly described by traditional content based method. The tagging based recommendation has been successfully applied in some notable E-commerce corporation, and bring promising benefit.. This research is focused on E-Commerce recommendation based on social tagging network. It has three parts work. First, it analyzes community structure of social tagging network in use of complex network theory to find the latent relationship of users, and form their community structure to help users to constructor their favorite resource. Furthermore it provides support for extended collaborative filtering. Secondly, the semantic relations among concepts are calculated by reference the theory of Complex Network, and these semantic relations are fused into the E-Commercial Recommendation, so as to improve the infer ability. Thirdly, the tagging based recommendation is combined with many other traditional recommendation technologies, so as to make the decision by considering many recommendation factors. Furthermore, the increment learning mechanism is deeply researched in this projected so as to perform the online learning, which is important when facing the dynamic changing environment of E-commence. This change probably raised by the reason of new product, or the user interesting changing. The dynamic learning technology makes the recommendation show the intelligent beheavior.

电子商务推荐可帮助用户快速找到满意商品，改善电子商务的服务质量，提高产品的销售量，它是电子商务和客户关系管理中的一个重要研究内容。近年来，社会标注网络可描述文本类内容，还可描述图像、声音、视频等传统方法不易说明其特征的产品，目前已在一些电子商务企业中成功采用，并带来可观的效益。本课题研究基于社会标注复杂网络的电子商务推荐方法，主要包含三项工作：一、借鉴复杂网络理论对标签网络的结构进行分析，发现用户之间隐含的联系，帮助用户组织自己喜好的资源，形成自己的网络社团，并为用户兴趣协同扩展提供依据；二、研究标签网络中的广义语义信息提高推荐系统的概念敏感型，改善推荐系统的深层次推理能力；三、研究引入标签的组合推荐方法并进一步研究其增量式学习机制，赋予推荐系统动态学习能力，以解决现有电子商务中因新产品的增加、用户兴趣的转移等引起的推荐环境变化，体现推荐系统的智能化。

电子商务推荐系统直接与用户交互，模拟商店销售人员向用户提供商品推荐，帮助用户找到所需商品，从而顺利完成购买过程。然而随着电子商务系统规模的进一步扩大，电子商务推荐也面临一系列挑战。对此，本项目进行了以下几方面的工作：.（1）鉴于推荐系统在电子商务发展中的普遍性，提出电子商务推荐系统用户采纳影响因素模型。验证兼容性为影响用户感知有用性和易用性的外部因素，感知易用性显著影响有用性，有用性及易用性是影响用户采纳电子商务推荐系统的重要因素，进而为电子商务企业推荐系统的改进提供了意见与建议。. （2）基于整合型技术接受理论通过结合研究对象本身特点，在原有解释度较高的模型基础上进行修改，从理性和感性的角度去全面整体的分析用户对于奢侈品网络商城的接受行为，增加新的变量，以期更好的诠释网络商城用户接受行为的主要影响因素，更实际的研究网络商城用户购物行为。.（3）借鉴复杂网络理论对用户网络的结构进行分析，选取权威用户，发现与推荐网络社团，以及进行网络社团的演化分析，从而发现用户之间隐含的联系，增加资源的分享机会，进而帮助用户组织自己喜好的资源为用户兴趣协同扩展提供依据。.（4）在前面工作的基础上，引入协同过滤的思想，增加具有相似兴趣喜好用户的信息将用户模型与用户对自身信息需求的组织形式相关联，克服用户个人偏好的异质性问题，以及偏置现象。同时设计了网络-行为-协同扩展的推荐系统NE-BE-CO，赋予推荐系统动态学习能力，以解决现有电子商务中因新产品的增加、用户兴趣的转移等引起的推荐环境变化，体现推荐系统的智能化。