Towards versatile conversations with data-driven dialog management and its integration in commercial platforms

Abstract

Conversational interfaces have recently become a ubiquitous element in both the personal sphere by easing access to services, and industrial environments by the automation of services, improved customer support and its corresponding cost savings. However, designing the dialog model used by these interfaces to decide system responses is still a hard-to-accomplish task for complex conversational interactions. This paper describes a data-driven dialog management technique, which provides flexibility to develop, deploy and maintain this module. Various configurations for classification algorithms are assessed with two dialog corpora of different application domains, size, dimensionalities and set of possible system responses. The results of the evaluation show satisfactory accuracy and coherence rates in both tasks. As a proof of concept, our proposal has also been integrated with DialogFlow, a platform provided by Google to design conversational user interfaces. Our proposal has been assessed with a real use case, proving that it can be deployed in conjunction with commercial platforms, obtaining satisfactory results for the objective and subjective assessments completed.