MONO91 Automatic Assignment Indexing Techniques chapter Mary Elizabeth Stevens National Bureau of Standards Table 2 (cont.) Materials Investigator Principles and Methods Used Tests Remarks Borko and Factor analysis to determine Same corpus as Detailed comparison with Some items canno Bernick distinctive groupings of clue Maron, 405 Maron's technique. For the processed becausE words. Maron's 90 clue computer source items, 63.4% were contain no clue wc words used for word-word abstracts, of correctly classified. For the correlation and factor which 260 new items, 46.5(70 correctly analysis. 21 factors used to indexed, and 48.9% were developed, and items establish correct for those items in manually re-indexed to factors, 145 which 2 or more clue words these categories. as new items. occurred. Swanson Text word lookup against Brief news Machine assignments compared to clue word lists, construct- dispatches manual subject indexing. For a ed by careful analysis of available on first batch of 500 items, 569 assign- sample items to be ex- teletype tape, ments of correct headings, 119 clusively indicative of a wide diversity assignments of irrelevant headings, 0 particular subject heading of topics. and 32 correct headings missed. Machine assigns a subject From study of The clue word thesaurus was then heading to an item if any several 1, 000 revised. For 275 additional test word on its list occurs in items, 24 sub- items, results showed 282 correct that item. ject headings assignments, 29 irrelevant assign- established and ments, 1 missed. For total, aver- word lists se- ages of 17% irrelevant assignments, lected, averag- 3% missed. For 200 items, mach- mg approximat- me and manual assignments were ely one hundred compared with respect to 5 of the per category. subject categories, with the 775 new items following results: then tested. Man Machine Irrelevant 4 25 missed 46 4 ___________ _____________________________ correct 75 116