Output details

Introspective Knowledge Revision in Textual Case-Based Reasoning

Accepted for oral presentation, this paper is an outcome of collaboration with IIT Madras as part of a UKIERI funded exchange. The originality is in challenging the assumption that generalisation always enhances document representations, and in developing a novel approach using introspective learning to select only useful generalisations. The significance of the work is in demonstrating that instance based learners, applying selective introspective learning, can compete with other leading classifiers. Rigor is shown in the extensive evaluation on both simulated and real-life datasets to demonstrate the effectiveness of the approach on both classification and unsupervised tasks.