Anirban Sen Department of Computer Science Email: anirban@cse.iitd.ac.in Linkedin handle: Anirban@Linkedin Phone: +91 78 3816 0298 (Mobile), +91 11 2659 7300 (Office) Office: Room 114, ICTD Lab, SIT Building

Research

• Analysis of mass media data to detect bias in prominent news-sources
• Analysis of corporate-political data to capture the interlocks between corporations and policymakers
• Analysis of mass media, , and social media data to understand the policy discourse

Current Work

There are several factors that can adversely impact development through incorrect or inappropriate policy formulation. These factors often arise due to the lack of proper functioning of the participants in the democratic process, e.g., the mass media, the citizens, and the Parliament. For instance, a biased media leads to a polarized public opinion, resulting in elections not being contested on well informed policy grounds. Similarly, incorrect information propagating through information channels like mass media and social media can also lead to the citizens being misinformed. This results in undeserving candidates securing key positions, often leading to improper policy formulation and implementation. Corruption between the corporate and government entities can lead to corporates influencing policy, to skew it in their favor. Lack of nuanced understanding of the policy requirements can also lead to superficial parliamentary discourse on the citizens' concerns. My research touches upon each of these factors, by analyzing data collected from publicly available web based sources.

For instance, I study large scale news data collected from prominent online newspapers, to understand if they are biased in terms of topics covered on the policy events, the dominant frames of presenting these topics, and the coverage they provide to various entities (politicians, bureaucrats, judiciary members) and political parties. I also study social media data to find out if the followers of these news-sources amplify or counter these biases on Twitter. To study the nature of policy discourse, I study statements made by influential entities (the power elites) in mass media, and the questions asked by the policymakers in the Parliament, using parliamentary Question-Hour data. With the help of my team, I have developed a corporate-government knowledge base for India, which provides information on corporate-government interlocks, for the purpose of observing the collusion between corporate and government entities. I have reported most of my findings in a website titled the Giant Economy Monitor. The purpose of this site is to bring out the results of my research in the public domain. I plan to update it with new data (e.g., new topics, new content on topics under consideration, new policy events, and further analysis of this data) so that it can present a time-evolving and current picture of the political economy.

Previous Work

1. Predicting Virality of Tweets in Online Social Networks: In Twitter, there are many topics (hashtags) which become viral with time, and are retweeted about thousands of times. There are some other topics, which die a natural death within days, hours, or even minutes. This work concerns identifying some structural features that include tweet features, user network based features, conductance based features, and users' geographical features to build a classifier that can predict if a topic will eventually become viral, by just looking at the initial few tweets on the hashtag. We have used a Random Forest based classifier, which with the help of these features solves the task with significantly high precision.


2. Extracting Situational Updates from Microblogs during Calamities [masters thesis]: Twitter gets flooded with tweets immediately after a disaster or calamity. Some of these tweets are purely emotional, while some provide actionable, factual information that can help in knowing important details on an event. The latter are known as Situational Updates. My work in this project involved analysing tweets corresponding to four disasters namely Uttarakhand Flood (India), Hyderabad Blasts (India), Sandyhook School Shooting (US), and Typhoon Bopha (Phillipines), and extracting a set of tweet and user based features to build a classifier, which would effectively identify situational updates. We have used support vector machine (SVM) for this purpose, and have come up with an interesting set of features, which work with significant accuracy.


3. Automated Diagram Generation from School Level Geometry Problems [B.Tech thesis]: In this project, we used natural language processing techniques to build a system, which parses school level geometry problems in English, and generates an intermediate, platform-independent graphical representation. From this representation, information can be extracted into a form which is compatible with any general graphical software (like OpenGL) that can be used to create a diagram in support of the problem statement. We tested our system on a set of around 500 geometry problems of varied complexity, and found it to perform with significant precision.