IRS13 Word-Word Associations in Document Retrieval Systems chapter M. E. Lesk Harvard University Gerard Salton Use, reproduction, or publication, in whole or in part, is permitted for any purpose of the United States Government. IX-18 associations appears to offer no advantages over first-order associations, and loses a great amount of material. We can therefore conclude that the use of associative procedures for the determination of word meaning in a general sense is not advisable with moderate sized collections of text, since the vast majority of the associations produced reflect specific local meanings of words. No choice of word frequencies or associative procedures appears to offer a way around this difficulty. 4. Retrieval Experiments Although the association process is not suited for investigations of absolute word meanings, it is nevertheless useful in retrieval systems. Fig. 4 shows a comparison between word-word association retrieval runs and straight word stem matching for three collections. It is seen that for two of the collections (the ADI and IRE collections) the improvement offered by associative strategies is only over small ranges and of doubtful significance. For the Cranfield collection, the associative strategy shows a definite superiority. The purpose of the associative method, originally, is to produce word relations missed by the stem matching procedure, and thus to take the place of a synonym list or thesaurus. It would be expected that such a procedure would be a recall-oriented device. However, this is not quite what happens. Associative procedures improve performance in two distinctly different ways First, they do occasionally retrieve a document that is missed in stem matching, by introducing new word relations which provide some request-document overlap. More often, however, precision is improved