A Question Detection Algorithm for Text Analysis

Abstract

In this paper, an effective question detection algorithm for Vietnamese text analysis is proposed. The proposed algorithm takes an audio file as input, converts its speech to text, and returns question detection result. This is extremely useful for a text analyzer to determine if a given sentence generated from an audio file is a question or not, particularly in chatbot or voicebot systems where very often there are needs for automatic replies to questions queried by users. The algorithm uses two tiers of question words and a customized question phrases to achieve 88.64 % accuracy on a sub-dataset of 176 questions prepared based on FPT Open Speech Dataset.