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-46 Fig. 14. The curves cross and the order of superiority depends on the operating range of the system. In short, the association process a) does not detect the same word relationships as a thesaurus or an index, b) does not use them in the same way, and c) does not have the same effect on retrieval. The operation of associative retrieval as a precision device can be illustrated also by noting that the mechanism proposed (promotion of relevant material near the top of the rank list) requires a moderately good performance to begin with. When the performance of associative re- trieval methods are compared between requests with good and bad performance in the stem dictionary, it is seen that hardly any requests with bad per- formance in the stem dictionary are promoted by the association process. This is shown in Table 14. Although there are 20 requests with rank recall above .2, and 22 requests with rank recall below .2, 10 of the 14 requests showing performance improvement with the association process have a stem rank recall of greater than 0.2. Finally, Table 15 shows the 25 requests which have a rank recall increase of at least .1 with some method, with a note of which methods improve the request. only two requests improve by this magnitude on all three methods. More requests are improved by the associative process only, and not by the thesaurus, than are improved by both the thesaurus and the associative process. It seems fair to conclude that associative retrieval and the use of a thesaurus are essentially independent methods of improving request performance.