The Development of an Architecture Using Traditional and Modern Recursive Neural Networks and Natural Language Processing to Increase the Accuracy of Response on Client Doubts Related to Public Services

In modern days customer service has become fundamental part in companies. The offer of customer services has been applied to many international companies, unfortunately domestic companies in Latin America lack the implementation of these services especially those companies that offer basic services. Most of these companies offer hotlines, and websites but these options may seem like time consuming and frustrating for customers to interact to. For this reason, we propose an innovative architecture to implement Spanish chatbots in companies that offer basic services. Modern international companies already offer constructed chatbots, but unfortunately these chatbots tend to be constructed with English language. This architecture aims to increase the accuracy in which it responds to a client’s doubts, saving them time and avoiding causing them frustration. In the case of this research, we have used a Latin American electric company that offers basic services, in this case there is no such dataset constructed in Spanish for these companies and for this reason we constructed our own dataset using the company’s website information. We compared the results of the model based on a rnn transformer architecture and seq2seq. As a result of this research, the proposed method achieves an accuracy of 91.57% and a loss of 1.76% with the transformer model. As a future research we suggest the implementation of special characters such as emojis and the increase of dataset to analyze the behavior and accuracy of these models.