Abstract: Semantic retrieval is a very potential approach in improving the accuracy of information retrieval and satisfying the customized requirements. This paper annotates the documents with thesaurus and then builds a bi-level indexing structure which from thesaurus element to thesaurus and from thesaurus to document. It makes a conversion from users’ query to the thesaurus, then calculates the semantic similarity in the thesaurus network. After doing that, the documents will be sorted in order. The system has been deployed and applied in a company, making the top 5 results hit 90% of users’ queries. Experimental results show that the method is effective for semantic retrieval.

Key words: sementic retrieval, knowledge document, thesaurus, similarity

摘要： 语义检索是解决信息检索中准确度、人性化要求的一个非常有潜力的方法。通过对知识文档进行主题词标注，然后建立从词元→主题词→知识文档的二级索引结构；对用户的检索，进行查询词到主题词的转化，计算语义相似度，按照语义相似度算法进行排序文档。目前基于知识文档的语义检索系统已经在某集团公司进行部署和应用，取得了前5项结果命中用户总查询90%的效果，说明这种方法是语义检索的一种有效途径。

关键词: 语义检索, 知识文档, 主题词, 相似度