Analysing fake news titles for 2016 Trump-Hillary campaign using contextual-based approaches in text analytics

Abstract

Text analytics is the process of transforming unstructured text data into meaningful information that can be used for fact-based decision making. It is widely used for sentiment analysis, summarising text or searching for useful information from the web. Existing approaches such as machine learning or natural language processing techniques have been proven to obtain significant information from massive amounts of text data. However, these approaches can have issues of obtaining sufficiently accurate results during training or the limitation of linguistic resources for the understanding of slang or acronyms for example. Thus, we propose a new method called Contextual Analysis (CA) that accentuates the relationship of the words and sources that are used for analysis. This approach will create a self-learned knowledge tree of contextual information, based on where words appear in the underlying sources. CA provides an understanding of the degree of relationship between the context of words which is a new technique to understand textual data sources. To evaluate CA techniques, 2000 news items are used that contain fake and actual news during 2016 Trump-Hillary campaign. The results are compared with other prominent Supervised Machine Learning (SML) techniques. CA matched the best classification performance and achieved the best performance of 0.81 accuracy for fake news prediction. Moreover, CA provides a Hierarchal Knowledge Tree (HKT) that helps to understand the context of words used in both fake and real news and is one of the important findings of this method. The experimental results demonstrate that CA has the potential to undertake classification tasks and at the same time reveal the contextual relationship and hierarchy of words which improves upon existing ML methods that treat each word as independent.