The use of second order descriptors for document retrieval

It is proposed that one way to increase the efficacy of document retrieval is to define to the computer the descriptors used to index the file. A computer program written in COMIT to implement the proposal and to facilitate testing its capabilities is described. Definitions are given to the computer as a string of terms called a “definitor.” These terms, which act as “second order descriptors,” are not normally those used as file descriptors. Their introduction provides a controllably broader base for link‐finding and matchcounting operations by the computer. It also makes possible such things as introducing new terminology and biasing existing descriptor indexes towards special interests or languages without having to re‐index the file. The program computes a “pseudometric distance” between a query and each document and prints an ordered list of those documents closer to the query than some chosen cut‐off value. (Large files would probably require some preselection, such as that which would result from use of a concordance.) It then substitutes for each descriptor its definitor and repeats the above process. The result is that the subjective human judgment required to evaluate the efficacy of introducing the definitors is reduced to a statement as to which list would be considered more useful. Use of the program to date has been only as a demonstration so no conclusions can be stated other than that the demonstration results would seem to indicate that testing on a serious scale should be undertaken. (This paper is a result of work sponsored by The MITRE Corporation's Educational Assistance Program).