THE USE OF TERM POSITION DEVICES IN RANKED OUTPUT EXPERIMENTS

• Published in: Journal of documentation Vol. 47; no. 1; pp. 1 - 22
• Main Author: MICHAEL KEEN, E.
• Format: Journal Article
• Language: English
• Published: Bradford MCB UP Ltd 1991; Emerald; Aslib, etc
• Subjects: Algorithms; Boolean Search Strategy; Computerised information retrieval; Computerized information storage and retrieval; Exact sciences and technology; Free text searching; Information and communication sciences; Information processing and retrieval; Information retrieval systems. Information and document management system; Information retrieval. Man machine relationship; Information science. Documentation; Information storage and retrieval; Information work; Natural language; Online information retrieval; Online Searching; Output; Rank Order; Ranking; Relevance (Information Retrieval); Research process. Evaluation; Sciences and techniques of general use; Search Strategies; Searching; Strategies; Subject Index Terms; Subject indexing; Tables (Data); Term position devices; Terms; Weighted Term Searching; Evaluation; Query formulation; Search strategy; Word; Position; System response; Search result; Use; Boolean function; Optimization; Research and development; Information science; Accuracy; Experimental result; Document retrieval system; Models; Full text
• ISSN: 0022-0418; 1758-7379
• DOI: 10.1108/eb026869

Term position information, as provided in some Boolean systems in the form of field restriction and term proximity, is reviewed and its value assessed. Non-Boolean retrieval in the form of the ranked output experiment has not so far used term position information but has concentrated on schemes of term weighting. The use of term proximity devices is proposed here by analogy with Boolean techniques and seven algorithms are devised to incorporate the ideas of sentence matching, proximate terms, term order specification and term distance computations. It is hypothesised that term position will act as a precision device. A new search experiment is then described in which a test collection is processed into sentences and then output ranking using term position is obtained. Results are given for five algorithms compared against quorum searching as the benchmark. The best result increased the precision ratio by 18% and used proximate matching term pairs in sentences plus a distance component.