Retrieval of answer-sentences and answer-figures from papers by text searching

作者:

Highlights:

摘要

Retrieval of passages from documents rather than whole documents as units speeds both user access to wanted information and the screening out of false retrievals. Passage retrieval services are already available to many lawyers. Results of an experiment reported here suggest that high quality passage retrieval services for scientists are now feasible. The experimental results are for biomedical retrieval questions, but reasons are given which support generalizing them. Since only titles, abstracts and words from figures (tables graphs, etc.) need be in computer-readable form, the retrieval procedures used are now economically feasible.Several characteristics of the results are especially noteworthy. (1) Search words for input to the computer search were selected by a person with only limited biomedical knowledge, aided primarily by a medical dictionary and medical textbooks (no thesauri or other cross-reference systems were used). (2) Recall averaged 90% and the lowest recall for any question was 75%. (3) The false retrieval rate averaged three falsely retrieved sentences per answer-paper retrieved, though for one question this value rose to 12. (4) Each answer-paper was retrieved by retrieval of a sentence, figure or (occasionally) title which was either in itself an answer-passage or became so when “automatically augmented”. In the latter case the computer annotated the passage with a qualification such as “multiple-case result” on the basis of words in the title or abstract. (5) Search of the words in figures in addition to those in titles and abstracts raised recall from 35% to 80%.

论文关键词:

论文评审过程:Received 28 June 1975, Available online 15 July 2002.

论文官网地址:https://doi.org/10.1016/0306-4573(75)90004-7