Leveraging on Semantic Textual Similarity for Developing a Portuguese Dialogue System

Authors

Abstract

We describe an IR-based dialogue system that, in order to match user interactions with FAQs on a list, leverages on a model for computing the semantic similarity between two fragments of Portuguese text. It was mainly used for answering questions about the economic activity in Portugal and, when no FAQ has a higher score than a threshold, it may search for similar interactions in a corpus of movie subtitles and still tries to give a suitable response. Besides describing the underlying model and its integration, we assess it when answering variations of FAQs and report on an experiment to set the aforementioned threshold.

Keywords

Semantic Textual Similarity, Question answering, Dialogue systems, Conversational agents, Machine learning

Related Project

AIA: Agente Inteligente para Atendimento no Balcão do Empreendedor

Conference

13th International Conference on the Computational Processing of the Portuguese Language (PROPOR 2020), February 2020

DOI