Crowdsourced Hypothesis Generation and their Verification: A Case Study on Sleep Quality Improvement

Abstract

A clinical study is often necessary for exploring important research questions; however, this approach is sometimes time and money consuming. Another extreme approach, which is to collect and aggregate opinions from crowds, provides a result drawn from the crowds' past experiences and knowledge. To explore a solution that takes advantage of both the rigid clinical approach and the crowds' opinion-based approach, we design a framework that exploits crowdsourcing as a part of the research process, whereby crowd workers serve as if they were a scientist conducting a "pseudo" prospective study. This study evaluates the feasibility of the proposed framework to generate hypotheses on a specified topic and verify them in the real world by employing many crowd workers. The framework comprises two phases of crowd-based workflow. In Phase 1 - the hypothesis generation and ranking phase - our system asks workers two types of questions to collect a number of hypotheses and rank them. In Phase 2 - the hypothesis verification phase - the system asks workers to verify the top-ranked hypotheses from Phase 1 by implementing one of them in real life. Through experiments, we explore the potential and limitations of the framework to generate and evaluate hypotheses about the factors that result in a good night's sleep. Our results on significant sleep quality improvement show the basic feasibility of our framework, suggesting that crowd-based research is compatible with experts' knowledge in a certain domain.