A Genetic-CBR Approach for Cross-Document Relationship Identification

• Main Author: Jaya Kumar, Yogan
• Format: Book Section
• Language: English
• Published: SPRINGER VERLAG 2012
• Subjects: T Technology (General)
• Online Access: http://eprints.utem.edu.my/id/eprint/6671/1/AMLTA_2012_revised.pdf; http://eprints.utem.edu.my/id/eprint/6671/; http://link.springer.com/chapter/10.1007/978-3-642-35326-0_19

Various applications concerning multi document has emerged recently. Information across topically related documents can often be linked. Cross-document Structure Theory (CST) analyzes the relationships that exist between sentences across related documents. However, most of the existing works rely on human experts to identify the CST relationships. In this work, we aim to automatically identify some of the CST relations using supervised learning method. We propose Genetic-CBR approach which incorporates genetic algorithm (GA) to improve the case base reasoning (CBR) classification. GA is used to scale the weights of the data features used by the CBR classifier. We perform the experiments using the datasets obtained from CSTBank corpus. Comparison with other learning methods shows that the proposed method yields better results.