Abstract: To quantify the degree of similarity between synonyms by term similarity accurately.Combining this quantification，the simple Bayesian network for information retrieval is improved，a four-layered Bayesian network retrieval model based on the quantified term relationships is proposed.The topology of the new model，probability estimation of all nodes and the whole inference process are described.Experimental results show that the new model in which term similarity is used to encode term relationships behaves better than others，realizing semantic retrieval to some extent，which is the inevitable tendency of information retrieval.

Key words: Bayesian networks, term similarity, information retrieval, synonyms

摘要： 利用术语相似度将同义词间的相似程度数量化，以此量化关系对用于信息检索的简单贝叶斯网络进行若干改进，构造一个四层贝叶斯网络检索模型。给出新模型的拓扑结构、各层节点详尽的概率估计以及文档检索与推理过程。最后，对新模型进行评估，结果表明该模型可以有效地提高检索性能，在一定程度上实现基于语义的信息检索，这正是目前信息检索发展的必然趋势。

关键词: 贝叶斯网络, 术语相似度, 信息检索, 同义词