Un-polarizing news in social media platform

Abstract

A person with incorrect information on a given subject/topic mays act against his/her own best interest due to the faulty believes. This is the misinformation problem and the rise of internet and social media has only worsened the problem as false stories are spread six times quicker than the correct one. Moreover, due to the nature of social platform, users unknowingly lock themselves in their own echo-chamber, amplifying news that strengthen their viewpoints while disregarding the opposition information. With the inspiration and knowledge gained from the public project: "Value from Public Health Data with Cognitive Computing project" at the University of Jyvaskylä (2017), I started this thesis with one main goal: to fight these problems concerning our modern society: misinformation, the spread of misinformation and the echo-chamber in social media platforms. By utilizing different sub-fields of Natural Language Processing (NLP) technology such as: Sentiment Analysis, Named Entity Recognition (NER) and Open Information Extraction (OIE), I created two hypotheses with two different approaches to suggest articles with different points of view to any given article. The main emphasis is that, by showing various news documents from diverse perspectives, a person gets a possibility to identify and discard the misinformation as well as crushing his/her own echo-chamber due to the exposure to the "other sides".

With a handcrafted evaluation database and benchmarks, I develop two prototypes to test the correctness and rigidity of our hypotheses. The first approach: the "Sentiment-based" solution achieves a satisfactory benchmark level by finding articles with similar topic/subject to the comparing article as well as suggesting ones with different sentiments/attitudes (negative, positive, neutral) using Sentiment Analysis and NER. The second approach: the "Statement/Triples-based" solution, by suggesting articles with relating or contradicting facts in the form of semantic-triples using OIE and NER, while fails our evaluation tests due to technical issues, has some convincing evident of a promising solution that can reliably detect contradictions spanning throughout multiple news sources. Thus, with a successful solution and many captivating findings, I hope that with the works described below, I could contribute to help battling the echo-chamber and misinformation as well as inspire other scholars and companies to do the same: help creating a better world.