Detecting Misuse of Information Retrieval Systems Using Data Mining Techniques

Initially, for each user, we obtain a profile. A system administrator assigns profiles in cases where allowable task vocabularies are known a priori. Otherwise, profiles are generated via relevance feedback recording schemes during an initial proper use period. Any potential misuse is then detected by comparing the new user queries against the user profile. The existing system requires a manual adjustment of the weights emphasizing various components of the user profile and the user query in this detection process. The manual human adjustment to the parameters is a cumbersome process. Our hypothesis is: Data mining techniques can eliminate the need for the manual adjustment of weights without affecting the ability of the system to detect misuse. The classifier learns the weights to be placed on the various components using the training data. Experimental results demonstrate that using classifiers to detect misuse of an information retrieval system achieves a high recall and acceptable precision without the manual tuning.