Estimation of Precedence Relations to Deal with Regional Complaint Reports

Abstract

A system in which citizens and the government work together to solve regional issues is known as Government 2.0. To promote this system, the collection of regional issues through mobile crowd sensing and collaborative IoT is being promoted. On the other hand, although prioritization is essential to solve the collected issues, conventional methods only classify the issues and do not identify the precedence relations between the issues. In addition, the latest deep learning models have not been applied to this task. In this study, we apply BERT to the task to identify the priorities of the collected issues based on the safety and security of citizens. We conduct experiments on a data set of regional complaint citizen reports. Experimental results illustrate that the BERT (fine-tuned approach) outperformed the other baseline methods even in the case of data sets with small vocabulary and biases among priority labels, such as the one in this task.