Revisiting Again Document Length Hypotheses

[摘要] The TREC2004 Genomics track evaluationexperiments at Patolis Corporation are described with afocuson the document length issues in differentretrieval models such as TF*IDF or probabilisticlanguage modeling approaches.In the genomics ad hoc retrieval task, combination ofpseudorelevance feedback and reference databasefeedback is applied. For the triage subtask, we trained a SVM classifierusing leaveoneoutcrossvalidation, and calibratedparameters to be optimal against the training set.

[学科分类] 社会科学、人文和艺术（综合）