Sequential adaptation of retrieval systems based on user inputs

Abstract A general model for utilizing user-generated feedback in the modification of a retrieval system is developed. The process is depicted as a sequential decision process and an adaptive estimation scheme for minimizing expected cost is obtained by means of dynamic programming, with the user responses treated as random variables. The paper assumes no knowledge of probability distributions and achieves a significant reduction in the dimension of the problem. Some tactics for the further reduction of computational effort are mentioned.