Regresar

A Real-Time Political Sentiment Analysis System Using DeepSeek-R1 on Multiplatform Social Media Data

Abstract:

Sentiment analysis on social media platforms has become an essential tool for gauging public opinion, especially in political contexts where discourse evolves rapidly and insights must be generated in real time. Traditional sentiment classification systems often struggle with linguistic variability, platform-specific expressions, and the computational demands of large-scale text analysis. Addressing these challenges, this work presents PoliSentiX, a real-time sentiment analysis system designed to classify user-generated comments about political figures across YouTube, Reddit, and X (formerly Twitter). The system integrates advanced Natural Language Processing (NLP) pipelines and leverages DeepSeek-R1 language models to categorize sentiments into positive, negative, neutral, or unidentified. PoliSentiX offers a web-based graphical interface that allows users to configure key parameters such as the target political figure, discussion topic, platform, and number of comments. The backend implements parallelized sentiment analysis to accelerate response times while maintaining classification accuracy. Experimental results demonstrate that the DeepSeek-R1 7B model outperformed other variants in overall accuracy, achieving 0.68 macro F1-score and 0.70 weighted F1-score while balancing performance with computational efficiency. Inference times ranged from 1.16 s for the 1.5B model to 6.73 s for the 8B model, indicating a trade-off between model size and responsiveness.