Generation of relation-extraction-rules based on Markov logic network for document classification

Abstract

Classifying documents into predefined classes is a very necessary task, especially in extracting information from huge resources such as web. Although a considerable amount of work has been carried out to classify documents into groups according to the subject domain or according to the other attributes. It still prevails as a big challenge in large scale, high dimensional document space. A number of techniques have been presented and proceeded with suggested improvements in order to achieve a higher degree of success in the document class. In this paper, a novel rule-based method for document classification with a combination of relation extraction techniques have been proposed. It is possible to replace overwhelming text classification techniques which involve thousands of words, document features or numerous patterns of word combinations by a set of rules which involves a much smaller number of entities and relations. We further discuss the effectiveness of relation extraction rules in document classification with the use of Markov logic networks for learning the weights of rules efficiently. Our experimental results show that the use of relation extraction rules on document classification yields a very high precision in the selected domain. We also demonstrate the applicability of our method on a benchmark text corpus with good performance measures.