Semantic Similarity Controllers

Currently, automatic solutions for measuring semantic similarity based on deep learning offer good accuracy, but little interpretability. The reason is that these solutions are trained over very huge deep neural networks that are imposible to understand by an human operator. To overcome this limitation, we have been working on the design and implementation of what is known as semantic similarity controllers, which allow for high degrees of interpretability without sacrificing good results. The reason is that fuzzy logics allows representing the underlying aggregation stategy in the form of fuzzy rules, that are more likely to be understood by people.