User's utterance classification using machine learning for Arabic Conversational Agents

Abstract

This paper presents a novel technique for the classification of Arabic sentences as Dialogue Acts, based on structural information contained in Arabic function words. It focuses on classifying questions and non-questions utterances as they are used in Conversational Agents. The proposed technique extracts function words features by replacing them with numeric tokens and replacing each content word with a standard numeric token. The Decision Tree has been chosen for this work to extract the classification rules. Experiments provide evidence for highly effective classification. The extracted classification rules will be embedded into a Conversational Agent called ArabChat in order to classify Arabic utterances before further processing on these utterances. This paper presents a complement work for the ArabChat to improve its performance by differentiating among question-based and non question-based utterances.