Mutual Learning for News Classification

Mutual learning represents a collaborative machine learning paradigm wherein diverse machine learning algorithms assimilate insights from varied sources and subsequently exchange knowledge among themselves. This fosters concurrent enhancement of classification and prediction accuracies across all participating agents. Particularly advantageous in resource-constrained environments such as mobile platforms, this dynamic learning approach circumvents the need for extensive training with voluminous datasets. In parallel, the proliferation of language-capable machines underscores the growing significance of natural language processing (NLP) in our daily interactions. Leveraging memory-efficient training, NLP-enabled machines are poised to seamlessly integrate into our lives. A notable application lies in news classification, a quintessential challenge in NLP, where machine learning algorithms categorize news articles sourced from diverse publications. This study demonstrates the application of Naïve Bayes and Support Vector Machine (SVM) algorithms for news classification. Furthermore, it introduces a novel approach wherein two interconnected networks dynamically assume the roles of teacher and student, facilitating mutual knowledge exchange to enhance the efficiency of both networks simultaneously. Initially, mutual learning is explored within homogeneous algorithms—two Naïve Bayes and two SVM algorithms. Subsequently, the scope extends to heterogeneous agents, involving a Naïve Bayes and an SVM agent, elucidating the comparative efficiency gains following mutual learning.