On using the DIET architecture for sentiment analysis and emotion detection

The Dual Intent and Entity Transformer (DIET) architecture has recently been proposed to perform intent classification and entity recognition in conversational agents. In this paper, we show that this architecture is also effective at other common tasks, such as sentiment analysis and emotion classification. The results have been validated in 4 different datasets and they show that DIET exhibits a comparative performance to other state-of-the-art methods, at the same time it provides a low code and fully configurable alternative that can be easily trained and deployed by using the Rasa conversational toolkit.

I. Introduction

Rasa is an open-source machine learning framework to help the design of conversational systems. The system works by processing the user’s utterances to identify the intention and entities associated with them. Intent refers to the user’s objective behind the utterance, and entities are additional information related to the intention. For example, in the sentence “I would like to order a pizza”, the intention is to make a purchase and “pizza” is an entity that refers to the contents of the order.