Application of the Maximum Entropy Principle to Retrieval from Very Large Databases

摘要

A study of the implications of the maximum entropy principle for retrieval from very large data bases, with associated investigations of the problems of combining expert judgments for retrieval and other problems. The project has established algorithms for selecting the combinations of characteristics most likely to yield useful retrieved items. Programs in FORTRAN run on an IBM PC/XT compatible computer with run times of less than 7 seconds for up to 4 terms. Computation time increases by a factor slightly less than three with each added term. The computation can be cast in parallel form to control the growth.