ANtIDOTE: Realtime Analysis of Information Diffusion for Trust and Relevance

• Applicant: Professor Dr. Peter M. Fischer
• Subject Area: Data Management, Data-Intensive Systems, Computer Science Methods in Business Informatics
• Term: since 2017
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 325007547

Modern social media like Twitter or Facebook encompass a significant and growing share of the population, which is actively using it to share messages. Given this broad coverage of the world as well as its fast reaction times, social media acts as a powerful "social sensor", while activities originating on social media can also have significant impact on the physical world. While scalable methods have been developed to detect trends and events in such media, assessing the relevance of social media messages and their trustworthiness is mostly driven by manual, after-the-fact checking. One of the most important factor for such an assessment is an understanding of the information diffusion. A typical use case for such analyses is online journalism, where journalists a) need to understand where information came from and how it reached them b) whom information published by themselves influences and in which way c) how the overall diffusion process is proceeding. Yet, existing work on analyzing information diffusion is centered on complex models with offline computations, making them unsuitable for real-time, large-scale analyses. In this proposal we aim to develop algorithms and systems to trace the spreading of information in social media that produce large scale, rapid data. The results of these analysis are then used to assess trustworthiness; the results are compared against complex, state-of-the-art methods. We identified three crucial building blocks for we need to investigate for such a real-time tracing system: 1) Design, implement and evaluate algorithms and systems that can trace information spreading and assign influence at global scale while producing the results in real-time. With the diffusion provenance generated by such a system, end-users can judge the relevance and trustworthiness of information based on its source and the users who contributed in propagating it. 2) Perform real-time predictions of the remaining lifetime of an information, as the process of information diffusion varies significantly in speed and duration. This provides insights on the diffusion processes to observe further, the completeness of results and - based on the temporal distribution (which is otherwise incomplete) - how trustworthy a message is. 3) Compute the credibility of users in an incremental way, addressing the large amount of data and the high data rates. Scores are computed for both trust and distrust and clustered afterwards to determine a stable assignment. Several technical challenges need to be addressed to provide rapid answers from iterative algorithms while dealing -at the same time- with high-volume/high-speed streams and low-latency access to very large graphs. As solutions to these challenges we investigate a novel processing architecture as well as graph partitioning strategies.

• DFG Programme: Research Grants