Complement of incomplete task results for real-time crowdsourcing interpretation

摘要

On the crowdsourcing, there is a problem of unstable task results because of an unspecified number of people entrying the task. It is important for the quality improvement of task result to integrate incomplete task results. However, it is difficult to perform majority vote method for free-text-input style task. In addition, it couldn't be the quality improvement in the cases of many incomplete task results. In this report, we propose an integration method for real-time crowdsourcing interpretation. The present method complements incomplete sentences using co-occurrence words in the other task results. The evaluation results show that the proposed method improves subjective appraisals about sentence complement results.