Development of Political QA Systems aimed at Assembly Minutes based on Abstractive Summarization

The assembly minutes published on the Web by local assemblies in Japan are not fully utilized. They are very long and complex in structure, making it difficult for citizens to read, and they only have a simple search function. In this research, we propose a Question-Answering System that allows citizens to solve their questions about the local government without reading the meeting minutes. The system consists of a content selector and a module based on a summarization model, and was trained on the Tokyo Metropolitan Assembly dataset. We built a prototype of the model that outputs answers to questions by referring to the assembly minutes, and it significantly outperformed the baseline method.