Overview of the Multi-Task Mutual Learning Technique: A Comparative Analysis of Different Models for Sentiment Analysis and Topic Detection

This research aims to provide a clearer overview of a new technique called Multi-task Mutual Learning in the field of Natural Language Processing, specifically in sentiment analysis and topic detection. The objective is to understand whether employing different models within this technique may impact its performance. With the growing collection of natural language-based data, private companies, public organizations, and various entities are increasingly seeking to extract information from this vast amount of data, which can be in the form of audio, text, or video. This underscores the need to study systems that can analyze this data effectively and do so in the shortest possible time, providing a competitive advantage in the private sector and a social analysis of the current historical moment in the public domain. The method employed is Mutual Learning, and within this technique, we analyzed specific models, including Variational Autoencoder, Dirichlet Variational Autoencoder, Recurrent Neural Network, and Bidirectional Encoder Representation from Transformer. These methods were executed with two datasets: YELP, containing reviews of commercial activities, and IMDB, containing reviews of films. The main findings highlight the complexity of the model, the computational power required, and the customization of the model according to specific needs.