Towards Crowd-Assisted Data Mining

Copyright retained by authors. Abstract Mining massive datasets can benefit from human input, but current approaches require making tradeoffs between overburdening end users or under-informing the system – algorithms become more accurate given more training data, but requiring more exemplars takes significant user effort. In this paper, we suggest an approach that engages nonexpert and semi-expert crowds as a supporting “interface layer” between end users and data mining systems. Leveraging human intelligence will allow systems to answer new types of queries (e.g., vague or subjective ones) and generate richer example sets for user-specified patterns.