An Empirical Investigation into Complex Fake News Detection Using the Multinomial Naïve Bayes Framework

In recent academic discourse, the escalation of counterfeit news dissemination via social media platforms has been identified as a burgeoning concern. Conventional, labor-intensive screening practices are proving inadequate in curtailing the rapid proliferation of such false information, underscoring the necessity for the deployment of automated, machine learning-based classification systems. This study focuses on the application of the Multinomial Naïve Bayes (NB) algorithm, grounded in Bayes Theorem, notable for its minimal computational demand, reduced energy requirements, and user-friendliness. Utilizing data primarily from the years 2016 and 2017, the research employs this algorithm to develop a model, subsequently achieving a classification precision of 95%. Prior to model training, data preprocessing was conducted to distill the essential elements of each dataset. The model's efficacy was assessed using conventional metrics, including precision, recall, and the F1-score. In addition, the Support Vector Classifier (SVC) was employed as a comparative benchmark. Despite its relative simplicity compared to the SVC, the NB algorithm demonstrated commensurate performance. Further, the research delved into the distinct attributes of counterfeit news, observing a higher prevalence of imagery usage in false reports, while genuine news more frequently cited sources for information. In summary, this paper elucidates the capabilities and constraints of the Multinomial Naïve Bayes algorithm in the context of fake news classification, underscoring its potential utility in this domain.