Improving Passage Retrieval Using Interactive Elicition and Statistical Modeling

[摘要] The University of Maryland and Johns Hopkins University worked together in the 2004 High Accuracy Retrieval from Documents (HARD) track to explore de sign options for interactive passage re trieval systems. HARD assessors re sponded to clarification forms by (1) selecting additional search terms from an automatically constructed list of po tentially discriminating terms, (2) se lected relevant passages from an auto matically constructed list of possibly rel evant passages, and (3) entered addi tional search terms. Query expansion based on these three types of elicited information yielded statistically signifi cant improvements in Rprecision over baselines with and without blind rel evance feedback. For topics that re quested passages as answers, a prelimi nary analysis shows that statistical mod els for passage extent trained on HARD 2003 data yielded a significant improve ment over a replication of the Univer sity of Maryland’s HARD2003 tech nique for passage extent determination, and the results of the new technique ap pear to generally be well above the me dian for HARD 2004 systems.

[学科分类] 社会科学、人文和艺术（综合）