索引策略是中文信息检索系统中重要的研究内容之一。对基于字、词和二元文法的索引策略进行研究,分析各自的特点,进行性能对比,在此基础上提出将词的索引策略与二元文法索引策略进行线性融合。应用2-泊松模型的BM25公式在TREC公开数据集上进行测试。实验表明,融合后的索引策略在主要性能评测参数平均精确率、R-精确率参数上相对较优。
The indexing strategy is one of the important parts of the Chinese information retrieval(IR). This paper focuses on the indexing strategy based on character, word and bigram, analyses the characters of each, compares their performance, and then proposes to linear combination the indexing strategies based on word and bigram. In this paper, the BM25 formula of 2-Possion Model is used as retrieval model and the TREC Mandarin corpus is used as data. Experimental results show that the inosculated indexing strategy is better on MAP and R-precision.