IRS13 Thesaurus, Phrase and Hierarchy Dictionaries chapter E. M. Keen Harvard University Gerard Salton Use, reproduction, or publication, in whole or in part, is permitted for any purpose of the United States Government. VII-27 only from a performance analysis viewpoint1 since the combinations of document lengths (e.g. titles), overlap correlation and logical vectors are known to be inferior to the regular abstracts cosine nuheric results. Fig. 18 presents data already given in Figs. 8, 9, and 11, but here the performance of different versions of essentially the same dictionary may be compared, the latter version always producing some improvement. B) Phrase and Hierarchy Dictionaries Since both phrase and hierarchy dictionaries are based on the grouping made within a given thesaurus, performance comparisons will be made between the thesaurus alone on the one hand1 and the thesaurus used with either phrases or hierarchy on the other. Using the normalized evaluation measures, four comparisons involving phrases are given in Fig. 19, and five comparisons with hierarchy appear in Fig. 20. For the phrase results in Fig. 19, phrase concept numbers are added to the requests and documents and given a weight of 1.0, equal to the weight of the original concepts in requests and documents. Phrases perform better than thesaurus on the IRE-3 collection, and on ADI, a small improvement for phrases is evident. With the Cran-l collection, phrases perform. a little worse than the thesaurus.. The hierarchy results in Fig. 20 are based only on one particular series of relations searched, in which both requests and documents are expanded by means of the hierarchy, and new concepts added are given a weight of 1.0, equal to the weight of original concepts in the requests and documents. Fig shows that use of the "Sons", "Brothers" and "Cross References" relations in the hierarchy results in a near equivalent, or worse performance than the 20