Design and Evaluation of a Conversational Agent Prototype based on Recurrent Neural Networks using Tensorflow.

Diego San Cristobal. (2018). Design and Evaluation of a Conversational Agent Prototype based on Recurrent Neural Networks using Tensorflow.. Final Career Project (Master Thesis). Universidad Politécnica de Madrid, ETSI Telecomunicación, Madrid.

Artificial intelligence is a multidisciplinary scientific discipline that aims to understand intelligence as well as the construction of intelligent entities. Due to this multidisciplinary character, artificial intelligence is experiencing a great boom because it allows to address a multitude of different problems obtaining good results. Specifically, one of the fields of artificial intelligence that is suffering more development is machine learning. This sub-field is responsible for the development of programs capable of generalizing behaviors from information provided in the form of examples. Within this branch, especially stand out neural networks that constitute a paradigm inspired by the neurons of the animals nervous system. On the other hand, one of the main objectives of artificial intelligence is to equip ma- chines with the ability to understand and generate natural language, allowing communication between humans and machines in an easier way. The Natural Language Processing (NLP) is the field of science responsible for the study of this process. Therefore, the purpose of this project is to analyze the problem of human-machine communication through natural language, as well as the development of a solution using techniques from the field of artificial intelligence. To achieve this goal, a prototype based on neural networks has been developed. For this purpose, different tools and technologies that provide the capacity to carry out this type of development have been analyzed. In addition, different sources of information have been studied for structuring the data set necessary for the prototype to perform the learning process. Finally, different experiments have been carried out with different data sets and the results obtained have been analyzed.