Multilingual information filtering by human plausible reasoning

Abstract

The theory of Human Plausible Reasoning (HPR) is an attempt by Collins and Michalski to explain how people answer questions when they are uncertain. The theory consists of a set of patterns and a set of inferences which could be applied on those patterns. This paper, investigates the application of HPR theory to the domain of cross language filtering. Our approach combines Natural Language Processing with HPR. The documents and topics are partially represented by automatically extracted concepts, logical terms and logical statements in a language neutral knowledge base. Reasoning provides the evidence of relevance. We have conducted hundreds of experiments especially with the depth of the reasoning, evidence combination and topic selection methods. The results show that HPR contributes to the overall performance by introducing new terms for topics. Also the number of inference paths from a document to a topic is an indication of its relevance. © 2010 Springer-Verlag Berlin Heidelberg.