CrowdDB: Query Processing with the VLDB Crowd

Databases often give incorrect answers when data are missing or semantic understanding of the data is required. Processing such queries requires human input for providing the missing information, for performing computationally difficult functions, and for matching, ranking, or aggregating results based on fuzzy criteria. In this demo we present CrowdDB, a hybrid database system that automatically uses crowdsourcing to integrate human input for processing queries that a normal database system cannot answer. CrowdDB uses SQL both as a language to ask complex queries and as a way to model data stored electronically and provided by human input. Furthermore, queries are automatically compiled and optimized. Special operators provide user interfaces in order to integrate and cleanse human input. Currently CrowdDB supports two crowdsourcing platforms: Amazon Mechanical Turk and our own mobile phone platform. During the demo, the mobile platform will allow the VLDB crowd to participate as workers and help answer otherwise impossible queries.