Dynamic Bayesian Networks in Classification-and-Ranking Architecture of Response Generation | Science Publications

Journal of Computer Science

Md. N. Sulaiman¹  Ramlan Mahmod¹  Aida Mustapha¹  Mohd. H. Selamat¹ 

关键词: Bayesian networks;    Conditional Probability Distributions (CPDs);    dynamic Bayesian networks;    Probabilistic Network Libraries (PNL);    classification-and-ranking;    Directed- Acyclic Graph (DAG);    natural language generation;    Natural Language Processing (NLP);   

DOI  :  10.3844/jcssp.2011.59.64

学科分类：计算机科学（综合）

来源: Science Publications

【 摘 要 】

Problem statement: The first component in classification-and-ranking architecture is aBayesian classifier that classifies user utterances into response classes based on their semantic andpragmatic interpretations. Bayesian networks are sufficient if data is limited to single user inpututterance. However, if the classifier is able to collate features from a sequence of previous n-1 userutterances, the additional information may or may not improve the accuracy rate in responseclassification. Approach: This article investigates the use of dynamic Bayesian networks to includetime-series information in the form of extended features from preceding utterances. The experimentwas conducted on SCHISMA corpus, which is a mixed-initiative, transaction dialogue in theaterreservation. Results: The results show that classification accuracy is improved, but ratherinsignificantly. The accuracy rate tends to deteriorate as time-span of dialogue is increased.Conclusion: Although every response utterance reflects form and behavior that are expected by thepreceding utterance, influence of meaning and intentions diminishes throughout time as theconversation stretches to longer duration.

【 授权许可】

Unknown