Social Media – The KGP Chronicle

Facebook Awards IIT KGP’s AI Research

Indian Express Financial Express The Week Times Now Hindustan Times NDTV Business Standard

Animesh Mukherjee, Associate Professor at the Computer Science and Engineering, IIT Kharagpur has been selected for the Ethics in AI Research Award in the ‘operationalizing ethics’ category for his project “Targeted Bias in Indian Media Outlets”.

The Award is part of Facebook’s initiative to encourage research on AI ethics and address intricate challenges and complex ethical questions in the AI domain. The award-winning project is part of collaborative work with Prof. Pawan Goyal and Ph.D. student Souvic Chakraborty.

This award-winning project deals with the burning issue of fake news. The scope of the project involves leveraging available information on online published media to predict fake news or identify bias in the news media articles. One of the major challenges in this work is formulating something as abstract as “bias” in a quantifiable manner.

Fake news has become a major point of concern in India owing to the explosive growth in the total number of smartphone users and a massive increase in the number of overall Internet users. Also, the emergence of social media and other forms of digital media make it difficult to identify the actual source of fake news among the huge number of secondary or even primary sources. In India, most of the laws being of the pre-internet era addressing this issue remains a legal complication. Rise of data analytics is adding to the issue with targeted content creation. When conventional media platforms propagate such bias they end up shaping the view of readers. During elections, such bias could lead to violation of EC regulations through its sparse, abstract and non-quantifiable nature. The situation could lead to narrowing down the voices of individuals and groups who do associate with targeted campaigns making use of such bias.

There has been little research to identify bias automatically in news media apart from manual studies done by independent journalists. As India crosses the half a billion smartphone users mark, it is more important than ever to characterize information available online through automatic algorithms and auto-updating crowd-sourced knowledge bases to restrict the spread of falsehood.

Prof. Mukherjee, who has been working in the areas of AI, ML, big data analytics and information retrieval, has identified the solution in leveraging available information on online published media to predict fake news or identify bias in the news media articles.

“One of the major challenges in this work is formulating something as abstract as “bias” in a quantifiable manner,” he remarked.

The team has collected 20-years’ data of three national media outlets and quantified bias on three metrics – coverage bias, word choice bias and topic choice bias. The team further plans to extend the reach of their study to local and digital media outlets.

Explaining the methodology, Mukherjee said, ”for a study with two datasets, the coverage bias was formulated as the ratio of the number of mentions of terms pertaining to two datasets. In case of more number of datasets, the researchers propose inverse of entropy in the distribution of words pertaining to each dataset. Word choice bias was formulated as the ratio of the positive and the negative words for each dataset. Topic choice coverage was formulated as the divergence score from the aggregate topic distribution for each data set.”

Talking about the future roadmap of this project Mukherjee confirmed they have plans to develop browser extensions to show bias score in real-time for identifiable sources.

On June 17, 2019, Facebook launched an India specific request for proposal seeking projects on Ethics in AI Research under three themes — a) operationalizing ethics, explainability and fairness b) governance and c) cultural diversity. The objective of the initiative is to help support thoughtful and groundbreaking academic research in the field of AI Ethics that takes into account different regional perspectives the three selected thematic areas. Synergy with the line of research that the TUM Institute for Ethics in AI was a key parameter.

A statement by the company says, “AI technological developments pose intricate and complex ethical questions that the industry alone cannot answer. Important research questions in the application of AI should be dealt with not only by companies building and deploying the technology, but also by independent academic research institutions. The latter are well equipped to pursue interdisciplinary research that will benefit society.”

The Social Salesperson

Economic Times Economic Times (Bangalore) Economic Times (Delhi) Economic Times (Kolkata)

Researchers at IIT Kharagpur have developed a new marketing model for influencing sales on social media. While customer reviews play a critical factor in online sales through social media, the researchers have developed an advanced model to identify influencers who could have more influence on potential buyers based on opinions and social ties on a popular social networking platform.

Social media is one of the most popular emerging strategies today with a fundamental goal of increasing sales. According to a study, about 91% of retail brands use multiple social media channels with 81% of SMEs using at least one social platform. The global revenue of social media sales is expected to grow to Euro 39 billion in 2019 as per a report by the business analytics firm Statista. However, on social media, people are more likely to adopt a product recommended received from their acquaintances or based on product reviews.

“We already know that comments on social media affect potential buyers. We have considered the personal valuation of the adapters based on their comments. Initially, we have segregated the adopters and the influencers based on their valuation and the threshold value to become an influencer. This helped us to categorize the users and strengthen their influence on adopters. In our second model, the peers’ connections are considered to influence a user. Additionally, we considered all the users who have purchased the product earlier, as the influencers other than just the potential buyers who are considered as users highly connected in the network,” explained Prof. M K Tiwari who led the research.

The study identified the different set of users based on their level of the tendency towards the product. This helps to segregate the adopters by whom they will be getting influenced instead of using all the buyers to influence all the adopters.

Thereafter the research group targeted the influencers by offering the product either for free or at a discounted price depending on their possibility to diffuse the information and influence their neighbours effectively for revenue maximization.

This finding provides the required benefits for marketers regarding the future of advertising and targeting customers in social networks. Marketers know that following traditional methods to motivate consumers in any social network might not always be effective. If marketers motivate any informal member of social networks without their knowledge by offering free or discounted products to initiate and launch any product related information, this can then be an effective strategy for social network advertisements. Also, the study showed that iterations in the product reviews by the influencers show a sudden increase in the number of people getting influenced. Such changes are to be estimated beforehand by the company and required steps to be taken in order to stay away from the lost sales.

“Here, we aim to increase the influence on people by offering the product for free to potential buyers who are capable of influencing more people and then the product is offered at an increasing price, i.e., decreasing discount rates and increasing the revenue as well as the growth of the influence among customers’ acquaintances,” he confirmed.

Computational experiments were conducted on real-world networks representing different scenarios with varying complexities and tested the effectiveness of these algorithms.

The research work was conducted under the research project EBusiness Center of Excellence (ECO) funded by the Ministry of Human Resource and Development (MHRD), Government of India under the scheme of Center for Training and Research in Frontier Areas of Science and Technology (FAST).

This work can be extended by implementing this algorithm on dynamic networks and budget and time constraints can be imposed for influencing. The results of this study shows that mixed influence model can be used to identify the potential users whom a company can target and also can decide the budget that can be spent on each category of such users based on their level of influencing others.

A future extension would be interesting by adding more social media futures such as likes and shares from sources like Twitter, Instagram, Pinterest etc.

Graphic Credit: Suman Sutradhar

Social Media to Help Disaster Relief

Economic Times Gadget Now India Today Business Standard Business Standard (PTI) Devdiscourse VC Circle

Your posts on social media and internal messaging applications can now help victims receive relief during natural disasters. While you tweet about the situation of the victims in specific locations in English or Hindi, a computer program can read through your empathetic post and send the relevant information to nearby relief operators. A research team led by Dr. Saptarshi Ghosh from the Department of Computer Science and Engineering at IIT Kharagpur is developing the algorithms for this task.

At the wake of the recent Kerala floods and other disasters in the past few years, the country has witnessed a significant rise in the real-time news regarding the disaster locations, victims, relief operations and call for help. Following these disasters, various social media platforms become important sources of real-time information regarding the disasters, coming from victims, on-site volunteers and empathizers all of whom act as “social sensors”.

In the aftermath of the April 2015 earthquake in Nepal, an NGO of medical practitioners “Doctors For You” posted information on their WhatsApp group, on the medical requirements after the earthquake. They shared access to the WhatsApp group with the research team of Dr. Ghosh who studied the exchanges between medical practitioners during a three-month period following the earthquake. Several insights were extracted on what are typical medical requirements after a major earthquake, for better preparedness at the wake of such events in fthe uture. “For instance, in the week just after the earthquake, the main concern is to deal with bone injuries and psychological trauma. In the later weeks, the focus has to shift to taking care of pregnant women and children, and to water-borne diseases,” confirmed Dr. Ghosh. The findings from the study were published in the journals International Journal of Disaster Risk Reduction and Disaster Medicine and Public Health Preparedness.

The team did further research with tweets posted during Chennai floods in 2016. Using the algorithm they found tweets with contact information, asking for drinking water, or calls for help required by adults and infants. There were also tweets that informed about the availability of resources such as drinking water in particular regions. Such posts on social media and internal messaging services could be extremely helpful for carrying our relief operations or plan in advance for disaster management. However, due to the huge volume and the rapid rates of posts related to the disasters, it is cumbersome to locate relevant messages which can contribute to enhancing situational awareness in the disaster-affected region.

“Only about 2% of the information tweeted in a disaster scenario turns out to be useful for relief operations, which is humanly not possible to identify. The critical information is immersed in large amounts of conversational content where people mostly express sympathy for the victims of the disaster. Our intelligent automated methods can identify and extract such critical information in real-time and in summarized form, which could be extremely helpful towards disaster management,” explained Dr. Saptarshi Ghosh.

The algorithms developed by the IIT KGP research team can search English and Hindi posts on various social media handles, especially micro-blogging sites, and IM services, and can extract and summarize critical situational information during disasters. The team has used advanced Neural Network and Deep Learning models to identify critical information from the informal language of social media posts, which lack grammar, contain arbitrary shortenings of words (e.g., ‘medicines’ shortened as ‘meds’), etc. Attempts are going on towards filtering and extracting information from disaster-related images posted on social media, which is also a challenging problem.

At present, the team is developing Web-based systems and mobile applications for aiding post-disaster relief operations. The systems will utilize the algorithms and perform tasks like identifying and extracting actionable information, summarizing the actionable information, etc. “The technology has reached a level where it can be deployed for use by relief operators. For instance, a person sitting in the control room can get live updates about what resources are needed where what resources are available and coordinate the relief operations accordingly. Or, a relief worker can get updates on his/her smartphone about people being trapped in the vicinity, so that they can be rescued.” said Dr. Ghosh.

The overall research work is being carried out in collaboration with Prof. Niloy Ganguly and Dr. Pawan Goyal from IIT Kharagpur (part of the Complex Network Research Group with which Dr. Ghosh is also associated), Prof. Somprakash Bandyopadhyay from IIM Kolkata, Dr. Arnab Jana from IIT Bombay, and Dr. Muhammad Imran, Qatar Computing Research Institute, and several students, notably Koustav Rudra and Moumita Basu.

The project has been jointly funded by IIT Kharagpur ISIRD grant, Microsoft Research India and ITRA, Media Labs Asia and DeITY, Govt. of India. Under the National Disaster Management Plan launched in 2016 development of such technologies are being supported. Recently, the Geological Survey of India has come up with a community-driven early warning system for landslide-prone areas in Darjeeling and training people to report rainfall threshold measurements which could lead to landslides.

“Availability of such trained volunteers operating our technology will be all the more effective in identifying and responding to social media queries and even promote the relevant information to people who could put them to use at the ground level” concluded Dr. Ghosh.

The innovation has been reported in several papers, the latest being in the popular journal ACM Transactions on the Web, and further publications are scheduled in the Springer journal Information Systems Frontiers.

Watch Videos

Utilizing Social Media for Disaster Relief OSM Disaster System Narmada – Disaster Relief Assistant