关键词: 位置语言模型,近邻性,段落检索,传播数

Abstract: In most existing retrieval models,the facts are often overlooked that the proximity of matched query terms in a document and passage retrieval used to score can also be exploited to promote scoring for documents.Inspired by this,a Chinese information retrieval system based on the positional language model was proposed.Firstly,we defined the concept of propagated count to establish a positional language model for each position.Then through combing KL-divergence retrieval model and positional language model,we scored for each individual position.Finally,we scored the document by the multi-parameter strategy.The experiment also focuses on comparing the retrieval effect of the two Chinese indexing approaches named multi character-based and dictionary-based on positional language models.Experiments on standard NTCIR5,NTCIR6 test sets show that the performance of the two indexing approaches of IR system improves greatly and it performs better than the vector space model,okapi bm25 model and classical language model.

Key words: Positional language model,Proximity,Passage retrieval,Propagated count