Dynamic Bayesian Networks with Ranking-based Feature Selection for Dialogue Act Recognition
Anwar Ali Yahya, Abd Rahman Ramli, Thinagaran Perumal · 2009
The automatic recognition of dialogue act is a task of crucial importance for the processing of natural language dialogue. It is also a challenging task as most often the dialogue act is not expressed directly in a speaker's utterance. This paper presents a dynamic bayesian network model for dialogue act recognition. The model is induced from annotated dialogue corpus via machine learning algorithms. Furthermore, the model is based on a proposed systematic approach to specify its random variables. In this approach, each variable is a binary classifier for exactly one dialogue act and constituted from a set of lexical cues selected using a specific feature selection approaches, called ranking approaches. To evaluate the model, three stages of experiments have been conducted. In the initial stage, the model is constructed using sets of lexical cues selected manually from the dialogue corpus. The model is evaluated against two baseline models. In the second stage, several ranking approaches are experimented for the selection of sets of lexical cues which constitute the random variables. In the third stage, the model is reconstructed using the random variables generated from the second stage. The results confirm the effectiveness of the proposed approaches for designing dialogue act recognition model.