报告题目：Towards Truth: Mechanism Design vs. Data Mining
报告摘要：Eliciting true information from the input is of critical importance for meaningful output. There are two ways to secure true information. One is mechanism design where agents reporting wrong information will not gain them any benefit. The other is data mining where true information is dug out from multi-sourced (and perhaps multi-modal) data upon which cross validation can be carried out.
About mechanism design, we use facility location games as an example to illustrate how the truthfulness can be guaranteed by sacrificing some efficiency. Various models are discussed for this setting of mechanism design without money. Especially for one of the variants, we consider how agents affect each other in their utilities. We assume that the coefficients are already known as input when the mechanism is applied. However, in reality, the way people (agents) affect each other needs to be learned from historical data. Therefore, we also analyze how to learn those coefficients in the underlying social networks. For data mining, we use social event discovery as an example, where multi-sourced social media data is collected, integrated, and fused, upon which clustering and classification algorithms are then applied to discover or detect significant/interested events. In dealing with incomplete or inconsistent data from multiple sources, a multi-dimensional model is devised to facilitate cross checking and validation, thereby achieving the objective of digging out true information.