Question answering using evolving networks

• Main Authors: See, Solomon Lim, Sih, Marc S., Tacderas, Beehjae F., Teo, Michael G.
• Format: text
• Language: English
• Published: Animo Repository 2004
• Subjects: Machine learning; Algorithms; Question-answering systems; Computer Sciences
• Online Access: https://animorepository.dlsu.edu.ph/etd_bachelors/14111
• Institution: De La Salle University

This study involves the exploration of evolving networks as a viable machine learning approach for question answering systems. Question answering systems engage in analyzing a free text, accepting questions and giving answers based on the input text. There are various works involving question answering systems using rule-based approaches and machine learning approaches. This research extends the work on question answering using machine-learning approaches by using evolving networks. The basic idea of evolving networks to improve performance. Three evolving network approaches on a back-propagation neural network were explored, namely, weights evolution, learning parameters evolution, and architecture evolution. The accuracy of each evolving network approach was benchmarked vis-a-vis other related works on question answering systems and was found to yield performance that is at par with the best performing approaches and in some instances, incrementally better. Thus, the evolving network approach is found to be a viable and competitive machine learning approach for question answering systems.