【作者】 李辉；

【导师】 杨德礼；

【作者基本信息】 大连理工大学 ， 管理科学与工程， 2008， 博士

【摘要】 为了引领消费者快速的获取商品信息,不仅商场、超市、书店等一些大型购物中心采用了多种导购方式,就连机场、公园、医院都设置了各种指引方式,我们也可以广义的称之为导购。电子商务的导购更是便捷,消费者可以利用互联网的各种搜索工具通过分类查找或关键词检索进行商品信息的获取。在移动商务日益繁荣的今天,同样也需要适合移动商务的导购方式。本文旨在针对目前移动商务导购研究的现状,结合移动通信技术和信息技术对适合移动终端操作的导购方式进行了研究,提出了移动商务导购系统的模型和系统组成。主要研究内容包括导购系统的模型、导购系统信息传递方式研究、基于知网的词义消歧研究、基于知网的相似度计算研究和威胁导购系统安全的垃圾信息过滤研究五个方面。在导购系统模型的研究方面,通过分析移动商务的终端特征及导购的目的,设计了移动商务导购的模型和系统组成,定义了各部分功能,并阐述了研究意义。在导购系统信息传递方式的研究方面,本文通过研究移动设备的技术特征和输入输出方式、移动应用平台所能提供的三种信息服务交互方式,设计了导购系统的信息传递模型。并对导购系统信息传递方式进行了设计和实现,试验结果符合系统应用要求。在基于知网的词义消歧算法研究方面,通过对知网的介绍分析,选择知网作为词义消歧的资源,并以知网的知识描述性语言为基础改写了知网的词语-义项文件成为词语-义项号文件,创造性的引入了距离系数来量化不同距离的实词对歧义词的影响,以此为基础,实现了基于知网的词义消歧算法,试验证明算法是有效的。在基于知网的相似度计算研究方面,通过归纳各种相似度计算方法的使用范围和优缺点,在总结中文语义分析难点的基础上,提出融合消歧策略的相似度计算方法,给出了相似度算法模型。以知网作为语义辞典,实现了基于知网的语义相似度算法,在此基础上实现导购系统问答模块。试验说明相似度算法和导购问答是有效的。在导购系统安全方面,从垃圾信息引发的拒绝服务角度考虑,对导购系统的垃圾信息过滤进行了研究。结合导购问答信息的文本特点,在比较了各种特征提取算法、特征权重计算方法、分类算法在中文信息分类中的优劣的基础上,提出了基于最小风险贝叶斯信息过滤算法,使用自建的短信语料库测试了该算法的性能,实验结果表明该算法能够有效的阻止垃圾信息,满足导购系统的安全要求。更多还原

【Abstract】 In order to guide consumers in masses of merchandise information effectively, multiple methods of guidance are adopted not only by shopping centers, such as emporiums, super markets and book stores, but also by airports, parks and hospitals. All the above are generally defined as shopping-guides. E-commerce provides shopping-guide more convenienc. As a result, consumers may obtain merchandise information by keyword browse or category-based search via internet. However, with the rapid development of Mobility-commerce (M-commerce), it is necessary to develop new methods of shopping-guide fitting for M-commerce nowadays.According to the existing M-commerce shopping-guide researches, this thesis puts forward a shopping-guide system based on M-commerce and its components, combining both mobile communication techniques and information techniques. This thesis mainly discusses the shopping-guide system module, methods of message transfer in shopping-guide system, word sense disambiguation based on Hownet, the computing of semantic similarity based on Hownet, and filtering of spam text messages, which are introduced specifically as followed.Depending on analyzing mobile devices’ technology and shopping-guide targets, a shopping-guide system based on M-commerce is put forward. The formation of the system, the functions of the different components and the significance are also provided.The module of message transfer in M-commerce shopping-guide is designed. This module involves researches on mobile devices’ technical features, input and output modes and three ways of interactive information services providing by mobile platform. It designs and realizes methods of message transfer and the experiment results satisfy the application requirements.The computing of word sense disambiguation is based on Hownet. The thesis introduces and analyzes Hownet as a resource. The computing changes word-sense files to word-sense number files using knowledge description language of Hownet. It also creatively brings in distance parameter to calculate influences from notional words with different distances on ambiguous words. And the computing is verified by experiments.On summing up the applying scope, merits and defects of various similarity computing methods, a new computing method integrated with strategy of diminishing the ambiguity is proposed, based on the summary of Chinese semantic difficulties. Adopting Hownet as a semantic dictionary, the similarity computing is realized, upon which Q& A. in the shopping-guide system is designed. Both the computing and the Q& A. are verified by experiments. In the aspect of the shopping-guide system safety, filtering of spam text messages is researched because of service refusal initiated by spam massages. Considering the text messages using in Q&A shopping-guide, a method of filtering adopting Bayes Risk Analysis is proposed. This thesis compares Algorithm on Feature Extraction, Algorithm on Feature Weights and Classification Algorithm in the Chinese information classification. And then it testifies the performances of self-building short message corpus. The experiments show that such computing can block spam text messages and satisfies the safety requirements.更多还原