MONO91 Indexes Generated by Machine-Automatic Derivative Indexing chapter Mary Elizabeth Stevens National Bureau of Standards This approach to the selection problem can be combined with other devices, as in the "Selective Dissemination" system described by Kraft in which keyword extraction indexing is applied to abstract, title, author's name and manually assigned index terms, after processing of all input material against both "in" and "out" dictionary lists. 1/ The use of abstracts rather than full text as source material makes the selection criteria problems somewhat less severe. In addition, there is evidence to suggest that the abstract does contain much of the significant information that would normally be indexed and the text of the abstract is therefore a fertile field for title augmentation. In experiments conducted by Slamecka and Zunde on the comparison of indexing terms manually assigned with the occurrences of the names of these terms in abstracts used in NASA's STAR system, it was found that 80.4 percent of the assigned terms were contained in the abstracts. 2/ Swanson, on the other hand, suggests that, at least for short articles having homogeneous subject matter, title and first paragraph "are nearly as good as full text , 3/ A combination inclusion-exclusion list system may involve prior "weighting for relevance "of words that are judged by human analysts to be significant for purposes of search and retrieval, as suggested by Swanson, for example: "The computer first separates those words which are important for purposes of information retrieval from those which are unimportant. This is accomplished by means of looking up each word in an alphabetized word list with which the computer is furnished. Each word in this word list carries a `weight' which reflects an estimate of its importance for retrieval purposes. Words of zero weight are completely unimportant and discarded by the computer for indexing entries." 4/ Continuing work at Thompson Ramo-Wooldridge on automatic indexing methods includes further investigation of assignments of relevance weight estimates to words and phrases, (1959 [490] and[491], 1963 [602]). 3.2.2 Book Indexing By Computer For internal indexing, that is, the subject indexing of the contents of a single book or report, automatic indexing experiments are usually directed toward the processing of full text, with use of stop lists of various lengths. The work of Artandi for her doctorate 1/ 2/ 3/ 4/ Kraft, 1963E334], pp.69-70. Slamecka and Zunde, 1963 [561]. In addition they report (p. 139) that a large number of the terms not found were "either broad, general terms (i.e., `device') or generic level concepts of terms contained in the abstracts." Swanson, 1963[580J, p. 1. Ibid, p. 1. 71