Hype Detection in Social Media

• Applicants: Professor Dr. Oliver Hinz; Professor Dr. Thorsten Strufe
• Subject Area: Security and Dependability, Operating-, Communication- and Distributed Systems ; Accounting and Finance
• Term: from 2015 to 2018
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 263254412

This project studies the emergence and development of hypes. Hypes are defined as very fast diffusion processes that emerge without prior announcement and quickly peak out. We thus examine subjects that abruptly arouse interest like unpredictable events or innovations as well as so called shitstorms that hit individuals, companies or other institutions. Understanding this phenomenon early will help identifying and mediating the process in time. In this course, the project integrates diffusion research and trend detection by means of an online analysis of dynamic graphs that are modelled from structural characteristics of user interactions, which are continuously retrieved from social media. The project consists of six well-connected work packages: Terminology and Requirements will be derived, and a system for the early detection of hypes will be designed first. Based on diffusion theory, work package two compiles a catalogue of indicators that identify characteristics based on observational data from the Web. Using these indicators, we aim to detect individuals that emerge as short-term opinion leaders and who foster the onset and development of hypes. Using this information, we develop data collection and management methods that allow for the identification and retrieval of data that needs to be processed and analyzed in real-time in work package three. Work package four develops graph theoretic methods to analyze the dynamic systems of interacting users. The data available from work package three and four defines the model space, which is evaluated in the fifth work package where we develop new diffusion models. Based on the insights from this research process we provide a proof of concept implementation, which detects hypes accurately in their early phases. The project provides contributions on targeted data retrieval and the online analysis of dynamic graphs to computer science. Collaborating with experts from diffusion research will help developing novel metrics to analyze interaction structures and to detect hypes. This will yield insight into patterns in social media, to detect meaningful data for targeted monitoring. Modelling the dynamically changing data as graphs, a novel and efficient method for the analysis of dynamic graphs is developed, which will enable their timely online retrieval and analysis. Business and management science can benefit from the project in several dimensions: Developing a catalogue of indicators we are able to predict the characteristics of individuals on the Web. This approach closes a large gap between survey oriented research that allows identifying individual traits and characteristics quite accurately for a small number of individuals, and large scale empirical analyses of observational data that can be found on the Web. The project also advances research in the diffusion domain and analyzes hypes, i.e., fast and short living diffusion processes that are not well-understood yet.

• DFG Programme: Research Grants