Improving user engagement with dialogue systems through meaningful response generation

Abstract

The sequence-to-sequence (Seq2Seq) model can be used to generate responses to various input sentences. However, these responses are often dull, such as in the form of simple consent, which reduces people’s willingness to continue the dialogue with a dialogue system. To overcome this limitation, this work was aimed to develop meaningful response generation methods, specifically, by (1) combining multiple response generation modules to the Seq2Seq model and (2) generating responses by introducing randomness to the Seq2Seq model. The results indicated that the adding randomness method could generate satisfactorily meaningful responses, thereby improving the user engagement with the dialogue systems.