Center for Business Analytics

Social Media Analytics

Researchers in the Predictive Analytics Lab develop enhanced predictive capabilities leveraging social media content, including online sentiment analysis, analysis of social media networks, and text analytics applied to user-generated content. Some examples include:
 
Social Media Analytics for Smart Health
In recent years, there has been a growing emphasis on the development of information and communication technologies capable of providing users with mechanisms for generating and accessing medical content via social media. Consequently, people are increasingly turning to social media as a source for health-related information and products, to seek information about health and wellness, and to attain advice, share experiences, and voice concerns. Understanding the importance of various channels in the context of smart health and real-time analytics is an important, yet little-explored endeavor. Many stakeholder groups, including patients, physicians, hospitals, pharmaceutical companies, and regulatory agencies are interested in knowing which channels are most suitable with respect to various dimensions. This study presents a framework for examining different social media channels. PDF
 
Modeling Interactions in Web Forums
The ability to accurately identify “reply-to” relations in online discussions has important implications for various social media analytics applications. However, accurately identifying such interactions remains a challenge, with existing methods providing inadequate performance. In this study, we propose a novel method for modeling social media interactions. The proposed method leverages several empirical insights about online interaction patterns, coupled with a robust machine learning algorithm, for enhanced classification of social media interactions. Furthermore, the proposed method also facilitates the creation of more accurate social media networks. As topological information derived from online communication continues to play an integral role in various social media analytics application areas, the results of our work have important implications. PDF
 
Detecting Adverse Drug Reactions using a Sentiment Classification Framework
Medical blogs and forums are a source of sentiment oriented content that is used in diverse applications including post-marketing drug surveillance, competitive intelligence and the assessment of health-related opinions and sentiments for detection of adverse drug reactions. However the application of baseline methods for sentiment analysis to health-related data sets provides moderate classification accuracy. These methods provide less useful features sets, thereby lacking in discriminatory potential. We propose an approach that uses feature set ensembles with novel feature representations that reduce sparsity by adding representational richness. The added feature representations extract important semantic, sentiment, and affect cues, that are better able to reflect the experiences of people when they discuss adverse drug reactions as well as the severity and the emotional impact of their experiences. We have conducted experiments on a test bed of drug-related posts, which have demonstrated improved classification accuracy as compared to existing conventional sentiment analysis methods. Furthermore, we present a case study to illustrate the suitability of the new feature sets to drug related sentiments that serve as useful indicators in predicting adverse drug reactions. PDF