基于强化学习模型的科技用户学习机制研究

【作者】 贺娟；

【导师】 甘利人；

【作者基本信息】 南京理工大学 ， 情报学， 2008， 硕士

【摘要】 本论文为国家自然科学基金项目“数据库网站用户信息搜索中的学习行为研究”(编号:70773054)的一个子课题,核心任务在于通过强化学习模型的拟合实验研究来挖掘科技文献数据库用户学习行为的内在机制。本论文首先对科技文献数据库检索方法功能进行剖析,并提出检索方法的分类体系框架;在此基础上,基于符号理论对科技文献数据库检索方法符号表征体系进行分析。然后,运用认知心理学理论,对检索方法符号体系的认知机制以及认知改变的学习机制进行分析,并提出检索方法符号选择行为改变的影响因素体系框架;同时,基于博弈学习理论,对本文要研究的科技用户对于检索方法的选择学习行为的机制与博弈学习行为的机制进行了比较。接下来,对两个重要的强化学习模型——BM模型、RE基本模型及其修正模型进入深入的理论分析,对模型的产生背景、理论假设、基本原理、验证方法、实验拟合情况、修正点、适用性等进行全面考察。最后,在模型分析的基础上,选择了其中的六个模型,对应科技用户检索方法选择调整行为中的学习规则进行解释与量化,并通过实验室控制实验,对模型的拟合预测效果进行了实证性分析。通过研究发现,利用强化学习模型对科技用户的学习机制进行探索是可行的。本论文的理论研究与实验研究在情报学领域尚属第一次,这对下一步研究的深入发展具有重要意义。更多还原

【Abstract】 As a fundamental part of the NSFC project titled "Research of the Learning Behavior of Scientific and Technological Database Users in the Process of Information Searching" (No.70773054), this paper is focused on studying learning mechanism of scientific and technological database users by analyzing learning models and organizing fitting experiment.After analyzing the function of search methods of scientific and technological database, a classification system framework of search methods is proposed, and the representation system of search methods is analyzed based on symbol theory.Secondly, the cognitive mechanism for symbol system of search methods and the learning mechanism for cognitive change are analysed based on cognitive psychology, and the influence factors system framework on choosing search methods is proposed, and the learning mechanism is compared between choosing search methods and games based on game theory.Thirdly, two important reinforcement-based learning models which are BM model and RE basic model and their modified models are analyzed, including background, theory hypothesis, basic principle, verification method, experimental fitting situation, modification, applicability, and so on.The explanation and quantization of six learning models according to search methods choosing behavior is conducted after models analysis, and fitting experimental is organized. The result is that it is reasonable that exploring learning behavior of scientific and technological database users using reinfoecement-based learning models.As the first time of exploration in the related field, the research conducted in this paper also formed a solid foundation for later researchs.更多还原