Laure Berti-Equille - Challenges in Truth Discovery: From Probabilistic Inference to Cross-Modal Data Fusion

User-generated content on the Web is massive, highly dynamic, and characterized by a combination of factual data and opinion data. False information, rumors, and fake contents can be easily spread across multiple sources, making it hard to distinguish between what is true and what is not. Given such a large number of conflicting and multimodal information shared in various formats (text, video, audio, structured data, Web table, micro-text, etc.), estimating the veracity of available data in a scalable and timely manner is extremely challenging and has only been preliminarily and partly addressed by recent work.
This talk will first review the main models, algorithms, and techniques for truth-finding at the Web scale and expose the current challenges and recent advances, in particular, the ones related to cross-modal truth discovery where consistency checking between textual messages and associated pictures or videos is essential for validating the authenticity of events in social media for example.