Introduction : During recent years, we as a society have evolved and transitioned towards online mode of news gathering and sharing. Due to this, news media agencies have witnessed a severe transformation in news consumption behavior of the users. Therefore, there is a focus on development of automated systems that can attract audience attention and retain attention through news recommender systems. While existing news recommender systems can ensure relevant news articles specifically for users, they often lead to echo chambers formation, spread of misinformation, hate speech and polarization as we have observed across different events in recent times. This highlights the requirement of a fair and trustworthy system for Computational Journalism.
To the best of our knowledge, there is no existing tutorial which explores Fairness in Computational Journalism from all aspects. Existing research works mainly focus on disjoint challenges, such as, development of news recommender systems, identification of echo chambers in social media platforms, bias detection in news articles, etc. While these studies are highly relevant, our tutorial would initially involve understanding of Computational Journalism and its objectives through the lens of social media platforms and news media agencies. Most importantly, this tutorial highlights the interrelationship among the different objectives and how it affects the society adversely. For example, we cover the implication of understanding users' preferences and their vulnerability to be in echo chambers on the basis of their news consumption profile and finally, discuss different ways to ensure fairness both at news media agency level and user level. Therefore, we believe that this tutorial will provide a detailed understanding of Fairness in Computational Journalism both theoretically and through applications and further, provide several new research directions.
Outline : The immense rise in on-line news consumption and increase in the number of news sources has increased the competition among news media outlets to select news articles that can draw the most attention. The continuous influx of newsworthy events further aggravates the decision to select news articles. To overcome this information overload, news media sources look up for automated systems that can filter out news items that are worthy of being reported next day considering the present day news events. Therefore, news media agencies are required to understand the significance of the news item on its current day of publication and its ability to retain user popularity the next day. This requires large scale mining of daily news items and modeling their importance with respect to specific user perspectives. In order to do these, news media platforms access the implicit or explicit user feedback on their platform to recommend information specific to user choices. Based on the feedback, news media platforms often align the news topics specifically to user interests and preferences. In order to handle these challenges, a new research area, Computational Journalism, incorporates different techniques which can handle the huge amount of user generated contents along with high data variance and sparsity to aid the journalists and news media agencies. While these automated systems, such as, identifying relevant opinions, understanding of user’s implicit and explicit feedback, news recommender systems, intend to maximize the user interactions with their specific platform, they often have several disadvantages with respect to users.
For example, news media platforms often align the news topics specifically to user interests and preferences. This selective exposure to news representation, i.e., exposure to only one aspect of a news event or even complete absence of a particular news event, can lead to formation of echo chambers among users. An echo chamber could be visualized as an isolated virtual environment where the user's beliefs and interests are reinforced by repeated exposure to information that aligns with the user's specific interests. Therefore, this strategic manipulation and selective exposure in news recommendation coupled with user's confirmation bias can amplify the inherent biases, leading to ideological segregation and polarization in society.
Through this tutorial, we designed it as a two-way hierarchical system, where we initially provide an overview of Computational Journalism followed by the necessity of Fairness in Computational Journalism through different aspects and finally, the challenges and existing research. During the overview of Computational Journalism, we will start with an introduction to Computational Journalism and its different applications both from the news media agency and users perspective, such as, identifying relevant user implicit and explicit interactions both at social media websites and news media aggregators, text mining and graph analysis based techniques to analyze and understand, i.e., summarization and stance detection and its application through news recommender systems. In order to visualize the requirement of Fairness in Computational Journalism, we will discuss the implications and impact of news recommender systems and alignment with respect to user’s interests through echo chambers identification and characterization. Therefore, based on this, we will finally provide a brief overview of the different aspects of fairness in AI, such as, assurance of trustworthiness, explainability, in proposed approaches along with specific fairness related aspects of Computational Journalism.
Further, we would provide a hands-on exercise where we initially show techniques for automatic dataset collection both from social media platforms and news media aggregators given a news article followed by visualization of the atleast one of the graph based and neural network based user opinion summarization techniques discussed in theory of the tutorial. Finally, we will show existing proposed metrics for echo chamber characterization and detection followed by fairness based existing metrics. Therefore, through this tutorial, we will cover both the theoretical and applications concepts, which range across text mining, social network analysis and mining and deep learning based specific models.
Duration : Four Hours (Including Hands-on Session)
Level of the Tutorial : Introductory as the tutorial won't require any pre-requisite information. However, knowledge of basic data science and text mining can help the audience but definitely not required before hand. Tutorial will start with the basics of social media platforms, news media agencies and concepts of online social networks coupled with text mining and finally, cover echo chambers detection mechanisms in detail. Echo chamber detection and understanding might be thought of as an advanced topic.
Instructor :
Dr. Roshni Chakraborty is an Assistant Professor at ABV IIITM Gwalior since August 2024. Prior to this, she was an Assistant Professor at Institute of Computer Science, University of Tartu, Tartu from December, 2022 to August 2024 and a Postdoctoral Research Fellow at the Center for Data-Intensive Systems (Daisy), Aalborg University, Denmark from November 2020 to November 2022. She received her PhD degree from IIT Patna, India in 2020 and M.E. degree from IIEST Shibpur, India in 2014. Her research interests include Computational Social Science in which she specifically works on applications related to Computational Journalism, Crisis Informatics, Time-Series Analytics, Bias and Fairness, and Signed Networks.
Email : roshni@iiitm.ac.in
Google Scholar : https://scholar.google.co.in/citations?user=9yWBPqoAAAAJ&hl=en