Connectionist interaction information retrieval

摘要

Connectionist views for adaptive clustering in information retrieval (IR) have proved to be viable approaches, and have yielded a number of models and techniques. However there has never been any exhaustive and methodical––i.e., theoretical, formal, practical, simulation- and user-based––evaluation of such a retrieval method and system. The aim of the paper is therefore just this. It suggests a connectionist clustering technique and activation spreading-based IR model using the interaction information retrieval (I2R) method. Theoretical as well as simulation results as regards computational complexity of this method are presented and discussed. Evaluations of relevance effectiveness are also given using standard test collections. Two applications were designed, developed and implemented based on this method. Their relevance effectiveness was evaluated in vivo in experiments carried out with human subjects. The results obtained show that I2R based on a connectionist approach proves useful when emphasis is on high precision.