A Deep Learning Model for Classifying Candidate-Related Tweets in the 2023 Nigerian Presidential Election

The increasing use of social media for political communication has transformed electoral discourse in Nigeria, particularly during the 2023 presidential election. While existing studies have largely focused on sentiment analysis or electoral prediction, limited attention has been given to candidate-level classification of political discourse as a distinct computational task. This study develops a deep learning framework for classifying tweets associated with the three leading presidential candidates Atiku Abubakar (PDP), Bola Ahmed Tinubu (APC), and Peter Obi (LP) based on their online presence. Twitter data were collected from six campaign-related hashtags between August 1, 2022 and February 28, 2023 to capture real-time political engagement. After preprocessing procedures including text cleaning, normalization, and duplicate removal, a total of 121,958 unique tweets were retained for model development. Two models were implemented: a baseline Artificial Neural Network (ANN) using TF-IDF vectorized features, and a Convolutional Neural Network (CNN) employing distributed word embeddings trained from scratch to capture localized campaign expressions and Nigerian political slang. Both models were evaluated using accuracy, precision, recall, F1-score, and AUC. The ANN achieved an accuracy of 0.5142, indicating limited effectiveness in distinguishing candidate-related tweet patterns. In contrast, the CNN achieved an accuracy of 0.9673 with balanced precision and recall, demonstrating superior performance in learning contextual and linguistic structures within short, informal political text. These findings confirm that CNN-based architectures are well suited for structuring large-scale candidate-related political discourse on social media. While this study does not infer voter preference or predict election outcomes, it provides a robust computational approach for analyzing digital political communication in Nigeria. Future research may extend this framework using multilingual datasets, transformer-based language models, and cross-platform political discourse analysis.