How Can a Novel Hybrid Sentiment Analysis System be Created to Extract Language Bias in News Media?

Abstract:

The modern news media is often biased. By pushing a certain perspective through its news narrative, biased news outlets are able to use their influence to manipulate public perception of information. Amongst other methods of swaying their audience, strategic word choices are frequently incorporated in news reports. This method is usually successful, the primary reason being the audience’s unawareness of most news articles’ biased natures. It is, therefore, the aim of this study to detect and quantify the affective states in the language used in modern news articles using a novel sentiment analysis model.

VADER (Valence Aware Dictionary and sEntiment Reasoner) is a lexicon and rule-based sentiment analysis tool specially adapted to analyzing short social media texts. In this paper, an error analysis of VADER is performed in a transductive transfer learning scenario; the pre-trained tool is then optimized using a Multinomial Naive Bayes classifier machine learning-based model trained on tweets for detecting and quantifying sentiments in news articles. This will be done manually after analyzing the outputs and combining the strengths of each model to create an improved hybrid system. This novel model, which would give insight into the severity of bias in major global news outlets for later analysis, is presented and evaluated.

The application of this model is that it successfully demonstrates bias in news articles concerning climate change, sea rise, air pollution, and environmental topics.