A MODULE FOR EVALUATING THE QUUALITY OF A DIALOGUE WITH CONVERSATIONAL SYSTEMS

Collaboration project with El Corte Inglés S.A. The main objective of the project was the design and implementation of a module to control the state of the conversation (evaluator of the quality of a dialogue). To do this, an initial study of the information sources and dialogue parameters to measure the quality of the dialogue was completed (dialogue acts, previous history of the dialogue, semantic representation, external sources of information, etc.). The project was structured in three phases. The main objective defined for the first phase was the design of rule-based models that considered these evaluation parameters to provide in real time a quality estimator between 0 and 1. In a second phase, complementary models based on machine learning were elaborated to predict the following system response. The previous phases have been complemented with an additional model for the detection of emotions based on deep learning, that was implemented in the third phase of the project.