Logistic regression modeling for context-based classification

Abstract

We focus on a machine learning approach to the concept/document classification for IR. We apply a logistic regression-based algorithm to three types of classification tasks: binary classification, multiple classification and classification into a hierarchy. At this stage, for our experiments we use a set of 150 topics from the TIPSTER collection. We develop heuristics as to how to build a logistic regression model for high dimensional, sparse data sets. This research describes work in progress.