用于科技文献检的扩展信念网络模型.pdf

摘 要 摘 要 参考文献信息是科技文献中的一部分主要内容，用来反映作者所做工作的主要理论 基础，同时也提供了一些与该文献相关的其他科技文献的信息。传统的信息检索系统只 是把文档的内容作为判断文档是否与查询相关的依据，没有考虑到参考文献信息对查询 结果的影响。 基于以上原因，本文利用科技文献间的参考情况信息来分析文献的重要程度，并将 这种方法组合到信念网络模型中，形成一种组合内容和参考文献证据的扩展的模型。详 细描述了根据文献间参考情况计算文献权重的方法，定义了衡量文献权重的两个指标： 价值度和关注度，并通过改进的艾宾浩斯遗忘曲线合理化了两个指标的计算结果。给出 了组合参考文献证据的扩展的信念网络模型的拓扑结构，并介绍了网络中文档的表示方 法、模型中各层节点的计算方法和模型推理过程。最后通过实验对扩展模型进行了评价， 结果显示扩展模型能够有效的提高检索性能。 关键词 信息检索 科技文献 信念网络 I Abstract Abstract Reference is a part of important information for scientific document. It presents the basic knowledge of document and shows the information of other scientific document which is related to this one. The traditional information retrieval systems merely use content information and not relate to reference information while computing. Base on this phenomenon, this paper compute the important degree of document based on the reference information. Reference evidence is combined with the original belief network model. Then we get the extended model which combines content information evidence and reference information evidence. This paper introduces an algorithm to compute the two metrics: worth and concern, which are used to measure the important degree of document based on the reference evidence. The metrics are rationalized by Hermann Ebbinghaus curve of forgetting. The model of extended belief network is presented. The way to express a document, probability computation and the whole inference process are described in detail. Finally, new model is used in ex