Tatsuki Tamano, Ryuya Itano, Honoka Tanitsu, Takahiro Koita

Proceedings of the 16th International Multi-Conference on Complexity, Informatics and Cybernetics: IMCIC 2025, pp. 137-140 (2025); https://doi.org/10.54808/IMCIC2025.01.137

Tatsuki Tamano
Graduate School of Science and Engineering, Doshisha University, Kyotanabe, Kyoto, Japan

Ryuya Itano
Graduate School of Science and Engineering, Doshisha University, Kyotanabe, Kyoto, Japan

Honoka Tanitsu
Graduate School of Science and Engineering, Doshisha University, Kyotanabe, Kyoto, Japan

Takahiro Koita
Graduate School of Science and Engineering, Doshisha University, Kyotanabe, Kyoto, Japan

Cite this paper as:
Tamano, T., Itano, R., Tanitsu, H., Koita, T. (2025). Real-Time Performance and Accuracy in Anomaly Detection by a Hierarchy of Crowdworkers. In N. Callaos, N. Lace, B. Sánchez, M. Savoie (Eds.), Proceedings of the 16th International Multi-Conference on Complexity, Informatics and Cybernetics: IMCIC 2025, pp. 137-140. International Institute of Informatics and Cybernetics. https://doi.org/10.54808/IMCIC2025.01.137

For anomaly detection, many proposed systems have used dedicated models and crowdsourcing. Crowdsourcing systems recruit anonymous workers on the Internet to accomplish specific tasks. Anomaly detection by crowdsourcing is achieved using the responses of multiple crowdworkers, but the accuracy of this approach is low. One possible cause is the influence of spam workers, who accomplish tasks in an inappropriate manner to earn a large amount of compensation. To eliminate spam workers, previous work has proposed a filtering method using qualification tests. In this method, workers are required to perform a qualification test before working on a task, and only those workers who meet passing criteria are allowed to work on the task. Unfortunately, such a filtering method can significantly reduce real-time performance. In this paper, we propose a method to improve real-time performance by introducing a partial filtering method through the hierarchization of crowdworkers. Experimental results show improved real-time performance while maintaining nearly the same level of accuracy.