Implementation and Evaluation of Recurrent Neural Network Models for an Intelligent University Admission Chatbot

Throughout history, universities have been fundamental in the advancement of scientific knowledge and technological innovation, allowing students to specialize and develop professionally in various disciplines. However, the university admission process is rigorous, and in the case of the National University of Moquegua (UNAM), it is regulated by a detailed regulation that generates numerous inquiries from applicants. To deal with these queries, the university hires staff and call center personnel, creating a vast historical data set of questions and answers. This research focuses on implementing and evaluating recurrent neural network models to train on this data, based on admissions regulations and other official documents, to deploy a chatbot prototype on a web interface. Four recurrent neural network architectures were implemented and compared: LSTM, BiLSTM, GRU, and SimpleRNN. The results show that all the architectures achieve good performance with an accuracy per nth of 84% according to the return metrics, highlighting the LSTM architecture with an Accuracy of 0.8608%, 0.8704% Precession, 0.8608% Recall and 0.8498% F1-Score.