MineAr: using crowd knowledge for mining association rules in the health domain

Crowdsourcing, where the power of the human thinking is harnessed to answer queries that are otherwise difficult for computers to answer, has been successfully used in many applications. A particularly interesting application of crowdsourcing is crowd mining, where given a dataset, patterns are learned by asking questions to the crowd. Crowd mining is extremely useful in situations where either the information is complex or it is not available in a systematic manner. In this paper, we target one such scenario, that of common health practices and cures. A web-based framework, called MineAr, is built to ask simple questions to the crowd. The questions ask whether a common product helps in a disease (such as ginger for cold). The crowd worker can choose an answer from different grades varying from "always" to "never", or can skip if she is not sure. Association rules are then mined from these answers using different aggregation techniques. Since not all the crowd workers can be relied upon, the system takes into account the confidence of the workers and, consequently, the rules are ordered according to importance. We also enhance the framework to enable prediction of answers of a new question for a crowd worker using her history. Finally, we construct a knowledge graph for searching and visualization.