Dual Degree Ph.D. Student Seeks Sentiments Behind Posts in Social Networks



Vinay Uday Prabhu 2   Over the past years we have witnessed the emergence of a new paradigm of engagement between people in online social networks (OSNs). Facebook or Twitter, for instance, have become an intricate part of many lives. The staggering membership numbers and robust participation of the masses have turned these OSNs into a veritable data gold mine for organizations and individuals who have a strong social, political or economic interest in maintaining and enhancing their clout and reputation. Therefore, extracting and analyzing the embedded sentiment in the microblogs (or Tweets) posted by the users about these organizations or individuals, or specific issues, products and events related to them or their competitors, is of great interest to them.  

Beyond the content of these tweets lies another facet: The connectivity graph itself! Understanding and characterizing the role played by the structure of this connectivity graph in determining the ease of the embedded sentiment prediction is precisely what Vinay Uday Prabhu is doing as part of his doctoral work in the dual degree Ph.D. program in Electrical and Computer Engineering, at Faculdade de Ciências of the Universidade do Porto (FCUP) and Carnegie Mellon University (CMU), as part of the CMU Portugal Program. 

Currently in the second part of his dual degree at CMU, after having spent the first half in Portugal at FCUP, Vinay Uday Prabhu is staying at CMU’s department of Electrical and Computer Engineering. Using mathematical models, algorithms, estimation theory and statistical signal processing, the Ph.D. student is trying to quantify, the statistical worth of these social networks in deciphering the latent sentiments expressed. 

Network Aided Sentiment Classification Expressed in Tweets and Posts 

Studying networks in general, and online social networks in particular, Vinay Uday Prabhu wants to understand and characterize the worth of utilizing the underlying network in classifying and clustering information generated by the users or in general, the nodes of the network.  

“The oft-minable underlying network is an awfully precious side-information source that can potentially turn a really hard problem into a manageable one,” he indicates.  “Let us say you gather a group of tweets and you would like to automatically classify the sentiment behind the tweet. Is it positive or negative? Is it a happy tweet or a sad tweet?” Simply by looking at the tweet, “it is very difficult to do that classification because there’s just a little text involved along with other subtle difficulties such as usage of sarcasm and so on,” Vinay Uday Prabhu explains. “However, if the tweeter belongs to a network of people and you know that many of his friends had exuded a positive sentiment, then on account of the social effect, or what the social scientists term ‘homophily,’ it becomes more likely that the tweet under analysis exudes positive sentiment too.” “Classifying the tweet is a difficult task, but when you have the network, the context, it becomes much easier,” he clarifies. 

Monitoring and Enhancing Social Capital of Organizations with Large Followings 

 “Our goal is to also characterize the relationship between the topology of the networks, or at least macroscopic aspects of it, and the rate of error incurred in estimating the latent common sentiment or the majority sentiment, which is also of paramount importance to organizations. For instance, let us say I am an executive in a large supermarket chain in Portugal and I want to know if my online followers are exuding a net positive or negative sentiment with regard to an online campaign I am carrying out,” the Ph.D. student explains.”The error rate incurred in such tasks is ostensibly related to the nature of connectivity between the followers.” 

“The basic idea here is that if the networks are weakly connected, you’d assume that people will be tweeting and posting on a more independent basis bereft of the peer pressure that might flow in through his or her connections” he explains. “In these cases, the network is not a strong driver of mass opinion. On the other hand, densely connected networks signify a flock mentality ecosystem where opinions polarize pretty fast,” he adds.   

Harnessing the graphical models framework, including specialized focus on a particular kind called “the Ising model,” that are used as a simplistic yet rich mathematical model of ferromagnetism in statistical mechanics, “we can organically model the network as a statistical prior” to understand social network behaviors. “Specifically with regard to my Ph.D. thesis, I’m basically looking at these novel applications where the underlying networks and network information can be used to perform classification tasks more efficiently,” the student explains. 

According to the student, this is the first work that connects probability of error and topology when estimating the latent sentiment in social networks. “We brought in a whole new approach to this open problem that had not been tackled until now whilst amalgamating ideas from classical communication theory, statistical physics and Machine learning,” the student reveals, adding that “we have just presented the results of this work at the IEEE International Conference on Communications (ICC) in Sydney, Australia, a top-tier conference in communication sciences.” 

Vinay Prabhu and João Claro 

The Dual Degree is “Wonderful and Extremely Challenging” 

When asked about his experience as a dual degree Ph.D. student, Vinay Uday Prabhu – who is advised in Portugal by Miguel Rodrigues, professor at FCUP and at CMU by Rohit Negi – stated that the experience has been “wonderful and extremely challenging.” According to the student, “I’ve been receiving a lot of support from both advisors. They have very different personalities and take different approaches towards the same problem, albeit with the same level of insistence on correctness and rigor. This goads me to prepare for my meetings with them slightly differently and I think that that’s a very unique experience in itself that you might not get in other programs,” he explains, adding that “I get to learn so much from both of them and I have no one else but the CMU Portugal Program to thank.” 

The student, who spoke fondly of his admiration for all the Portuguese characteristics, especially the cuisine and Port Wine, believes that Europe – and Portugal in particular – needs to advertise itself more aggressively. For instance, “you have lots of summer and winter schools happening all over Europe. I think it’s a truly unique aspect of the European ecosystem where we can learn a new mathematical technique or the latest state-of-the-art and the research being done by professors,” he exemplifies.  

August 2014