Le Sun Junlin Zhang Yufang Sun

[摘要] Institute of Software, Chinese Academy of Sciences (ISCAS) participated in TREC2004,submitting 18 runs. We focus on studying the problem of the combination of the user andqueryinformation from clarification forms and metadata. We provided two kinds of ClarificationForm. Our experiment shows the CF2 is more effective than CF1. We use Google as a resource forquery expansion base on metadata subject and familiarity together, and the Rprec is increasedfrom 0.2308 (baseline) to 0.2646 (+14.6%). Our approach to exploiting the metadata Genre andGeography yield negative result when used alone, however, surprisedly, when combinate metadataGenre and metadata Geography with CF2 respectively, we get an increase (+1.2%) and (+5.4%)than use CF2 alone. Our combination of CF2 and metadata relttext is the best results of all theTREC runs (Rprec), and in this run, the Rprec is increased from 0.3303 (CF2 alone) to 0.3766(+14%), and from 0.2888 (metadata reltext alone) to 0.3766 (+30.4%). From the results we cansee the information from user (CF2) and the information from query (metadata relttext) may

[学科分类] 社会科学、人文和艺术（综合）