Socio Political News Detection Using Enhanced Graph Neural Network

Abstract

Purpose: In the age of data driven by social media, the proliferation of socio-political news has amplified the challenges of detecting misinformation, biases, and polarized news. Socio-political news detection process is essential for promoting knowledgeable decision-making among citizens, assessing public sentiment, influencing policy formulation, tracking crises, strengthening international relations, and maintaining public security.

Design/Methodology/Approach: This paper introduces a hybrid model for detection of socio-political news by merging BERT model (Bidirectional Encoder Representations from Transformers) with Graph Neural Networks (GNN). The proposed model leverages the ability of BERT’s contextual embedding to capture semantic information form text and incorporates it with the structural insights of Graph Neural Networks to model the relational dependencies among news elements, such as headlines, article and entities. RSS feeds are collected to estimate the proposed model’s performance.

Findings: It demonstrated superior performance in terms of accuracy, precision and recall compared to traditional machine learning methods and independent deep learning models.

Originality/Value: The proposed socio-political news detection approach offers a robust solution by incorporating various models. It provides researchers to manage and perform analytical task of socio-political news.

Paper Type: View Point