On-line index term predictions using bigram-term associations

Abstract

Predicting index terms (or keywords) by examining a word's component letter strings is investigated. The weights or string-term associations for the letter strings are determined by using relative frequencies computed from a representative sample of the total abstract (or document) collection. The experimental results indicate that the terms predicted by using bigrams (letter pairs) are effectively the same as those predicted by using bigrams and longer letter strings.