Desarrollo de un asistente virtual de información utilizando pnl y machine learning para la Universidad Técnica de Machala.

Abstract

Currently, the technological field has presented great advances, especially in the area of Artificial Intelligence (AI), which has led to its implementation in different types of technologies, areas and modes of use, such as: voice recognition, facial recognition, image classification, content marketing, content generation, chatbots, among others; however, the present research focused on chatbots or also known as virtual or conversational assistants, and that, thanks to their AI mimic human behavior, thus managing to give assertive responses corresponding to the user's intention, as configured, being able to respond not only through textual messages but also by images, videos or audios. A virtual assistant can be available 24 hours a day, 7 days a week, being able to respond to an endless number of questions without the need for external help, that is, with human help, depending on the knowledge base that the AI handles. For this reason, this work was oriented to the development of a virtual assistant named "UTI-bot", which would be in charge of the tedious and repetitive task of providing information, due to the need of the users of the Technical University of Machala when requiring information or help regarding their concerns, with less time and greater availability. What was intended to be achieved was the creation of an online user support service capable of solving the doubts of approximately 12578 young graduates aspiring students of UTMACH (external users) and 12425 young current students of the University (internal users), and that, as mentioned above, by means of this virtual assistant, more than 75% of internal and external users can be satisfactorily attended. For the development of the virtual assistant, two methodologies were used: XP and CRISP-DM, where, in Chapter II, the procedure carried out according to the phases of these methodologies is shown. Among the technologies used for the creation of the "UTI-bot" AI for the Web portal and the Facebook social network were: Natural Language Processing and Machine Learning. The research for this work was correlational and transversal, with an exploratory scope, applied to a sample of 379 users. The survey was used as the main technique and a Google form as an instrument, in which the level of satisfaction of the respondents with respect to the prototypes used was analyzed and appreciated. To measure the degree of satisfaction, the Likert scale was used, immersed in the questions posed in the Google