Detecting news bias in the emerging market using transformed model

Abstract

The news readers from western countries are exposed to news reports from the global news media outlets that focus fundamentally on the conflicts happening in third-world countries such as those in South Africa and Nigeria. Many financial institutions from western countries provide financial opportunities to emerging markets in Africa. For this purpose, an important source of information used by those institutions is the global media outlets reporting the financial news. Therefore, there is a need to assess the credibility of global news outlets in covering financial news for African countries. This study focuses on detecting financial news bias in the emerging markets countries such as South Africa and Nigeria using a transformer model called BERT. The transformer model for sentiment classification was developed using the pre-trained model to achieve this goal. The financial news titles from the local and global news media outlets about South Africa and Nigeria were downloaded from the online news database called Global Database of Events, Language, and Tone project, and they were injected into the BERT model to compute the sentiment scores and labels for each news title. The pretrained BERT model achieved a good accuracy of 89.76% after being fine-tuned with a sample of 5000 movie review dataset. We found that both the local and global news media outlets report more positive financial news than negative news for both countries based on the results from the sentiment scores and labels. To assess news bias between the local and global news media outlets for South Africa and Nigeria, the average monthly sentiment scores were compared with the average monthly exchange rates for each country to determine the correlation between them and test if that correlation is significant or not. It was found that only the Nigerian global news coverage correlates with the exchange rates. Therefore, it was concluded that there is no evidence to suggest that the global news media outlets are biased in reporting the financial news in emerging markets countries since it was seen that the sentiment scores from the local news outlets for both countries do not correlate with their respective exchanges rates. It is evident that the transformer model can be used to accurately compute the sentiment in the financial news articles for the purpose of detecting news bias.