An analysis of ill-formed input in natural language queries to document retrieval systems

摘要

We analyzed natural language document retrieval queries from the Thomas Cooper Library at the University of South Carolina in order to investigate the frequency of various types of ill-formed input, such as spelling errors, co-occurrence violations, conjunctions, ellipsis and missing or incorrect punctuation. The primary reason for analyzing ill-formed inputs was to determine whether there is a significant need to study ill-formed inputs in detail. After analyzing the queries, we found that most of the queries were sentence fragments and that many of them contained some type of ill-formed input. Conjunctions caused the most problems. The next most serious problem was caused by punctuation errors. Spelling errors occurred in a small number of the queries. The remaining types of ill-formed input considered, ellipsis and co-occurrence violations, were not found in the queries.