【作者】 谭婷婷；

【导师】 蔡淑琴；

【作者基本信息】 华中科技大学 ， 管理科学与工程， 2011， 博士

【摘要】 互联网的快速发展为信息的获取带来极大方便,但由于互联网信息的海量、无序、去中心化等特点,用户很容易产生“信息迷失”。如何采用一定的技术即时找到用户需要的高质量信息,并通过一定的方式呈现给用户,是推荐技术要解决的核心问题。Web2.0技术的应用让用户参与创造的内容加入了网络信息资源的阵营,进一步加大了信息推荐的难度。传统的推荐技术一般都通过描述用户与资源的简单对应关系来表达个性化需求,而微内容推荐的用户行为分析既要考虑作为消费者角色的用户偏好,同时也要考虑作为生产者的用户偏好。本文提出将用户行为以及用户之间的交互关联纳入考虑范畴来构建整个推荐体系,针对推荐中的关键问题(社会网络影响、冷启动、可扩展性、人机认知等)对微内容推荐方法进行了以下方面的研究：(1)通过实证提出基于用户关注度的微内容过滤评价指标,并根据用户社会关系的分析,识别出影响关注度的指标,利用这些指标有针对性的对有价值信息进行预测和过滤导向。(2)构建了基于超网络的推荐路径。从微内容的社会性入手,提出微内容的互动与传播网络分三个层次：用户对象关联网,信息资源对象关联网,用户一信息对象二分网络等。通过信息与资源的映射关系,基于超图实现用户对信息的选择(评价)过程以及由传送路径构成的推荐网络的形式化描述。(3)结合微内容信息节点推荐的特征,利用加速遗传算法对微内容推荐路径进行优化。将信息节点标签相似度、基于关注度的信息价值以及信息节点距离度等作为推荐路径计算的多维约束指标,构建出优化的适应度函数,实现了推荐算法的全面性考虑并借用加速遗传使算法得到有效精简。(4)引入多智能体技术,以平台视角构建了上述的各项功能模块,并利用智能体技术优势扩展了人机交互和知识学习等功能,使微内容推荐实现了虚拟空间的人机自动协作,构建了微内容推荐集成平台的原型系统。本文以互联网企业的内容生产加工应用领域为研究背景,以超图、回归分析、遗传算法、智能体技术和平台方法等多种理论和方法为基础,采用定性与定量建模、实证相结合的方法,研究微内容推荐的方法和支持平台等相关问题。更多还原

【Abstract】 The rapid growth of internet has brought great convenience for achieving information. Meanwhile, since the internet information tends to be larger, more disordered and more decentralized, it is already beyond human’s information processing capabilities. People usually feel it not easy to find the required information, namely the phenomenon of information overload. The key issue is that how to provide high quality of information with some kind of techniques in a short time. The Web2.0 technologies expand the boundary of internet information resources with user-generated-content (UGC), which makes the recommendation more difficult. So it’s no longer that recommendation techniques just rely on the simple relations between users and various resource objects. What’s more, the requirement for Internet information could no longer be explained only by personalized recommendation, because the influencing factors include the needs of both producers and consumers. We bring users’ behavior and their cross-correlations into the recommendation system. Aiming at the crucial issues (social network influences, cold boot, expandability, human-technology interaction, etc.), our research on recommending micro-content are:(1) We propose the parameter index for micro-content filtering based on users’ attention rate. Then we identify the indices influencing attention rate based on social relations analysis, and according to these, valuable information could be forecasted and filtered.(2) We establish the recommendation paths based on super-network theories. Start with the sociality of micro-content, we propose that the network for micro-content interaction and communication is a super-network, which could be divided into three levels.Through mapping from users’relations to information relations, we can achieve the hypergraph-based description of the user selection (evaluation) process and the the recommendation network by the transmission path.(3) Combining with the characteristics of micro-content nodes, we bring in accelerating genetic algorithm to optimize the recommendation paths. We regard dimensional binding targets as tag similarities of information nodes, information values based on attention rate, and the recommendation distance, obtain fitness function, consider the matter of information recommendation in its entirety and the accelerating genetic algorithm reduces decision-support complexity.(4) We bring in multi-agent techniques and establish the functional modules based on preamble analysis from the perspective of decision support systems. Multi-agent techniques expand on the capabilities of human-computer interaction and knowledge learning. So the micro-content could achieve man-machine collaboration automatically and the integration platform of micro-content recommendation is established.The background of this research is the micro-content producing and processing applications of Internet enterprises. Many theories and methods including hypergraph theories, genetic algorithm, agent techniques, decision support systems, are used in the study. The methodologies include both qualitative modeling and empirical studies, to look into the issues of techniques and support systems of micro-content recommendation.更多还原