Towards Secure and Privacy-assured Truth Discovery from the Crowd

Description

Many tasks of data analytics today rely on data collection from massive diverse sources or inputs from enrolled clients, as in crowdsensing, crowdsourcing, and IoT applications. But in practice, the data quality among different sources might vary greatly, leading to information conflicts even on the same object of interest. To extract trustworthy information out of multi-source noisy data, truth discovery, which aims to automatically infer the source reliability, often unknown a priori, and draw truths through reliability-aware aggregation, has received significant attentions recently. While truth discovery is prominent in finding trustworthy information without supervision, the latest advancements have largely overlooked the important data privacy issue. In many scenarios, data from individual clients naturally contain private information that must be well protected, such as personal trajectory data, patient reactions to drug usages, personal preferences to sensitive questions, etc. Without privacy-preserving guarantee, individual clients might be reluctant to submit their private data, which would degrade the performance of data analytics. Considering the growing public awareness of privacy and increasingly strict legal regulations, there is a pressing demand to develop full-fledged privacy-preserving truth discovery (PPTD) frameworks for secure and trustworthy multi-source information integration. In this proposal, we will push forward the frontier of theoretical and applied PPTD research, with two broadly defined objectives. Firstly, we plan to bridge the gap between the nascent PPTD and its thriving plaintext counterpart. Specifically we will investigate: 1) new PPTD frameworks without unrealistic clients' online engagement that is indispensable in prior limited studies; 2) privacy-preserving mechanisms to support diverse data types, including non-numeric, non-uniform, and streaming data; 3) systematic optimizations by exploiting unique task characteristics. Secondly, we consider PPTD in largely under-explored decentralised settings, like open blockchain and smart contract platforms. Unlike server-centric architecture, such platforms are known for low entry barriers. Deploying truth discovery tasks therein enables the crowds to contribute unused computing resources as service providers for rewards, besides contributing data. However, the open settings and the lack of trustworthiness guarantee bring in additional security challenges. Accordingly, we will develop: 1) new PPTD service models in decentralised settings with possible participant rewards; 2) integrity assurances to handle untrustworthy service providers in open settings; 3) privacy-preserving validation techniques to detect potentially faulty clients with invalid private inputs. The results will enable wider truth discovery adoptions that would otherwise be unlikely due to privacy concerns, and contribute to broader roadmap of private and trustworthy blockchain-based knowledge monetisation.