Interdisciplinary Design for a Semi-Automatically Updating Voting Advice Application System

Besides the purely technical part for providing a Voting Advice Application (VAA) as a web service, VAA designers are confronted with two main challenges largely defining the workload they are facing for the preparatory phase before being able to provide a VAA: a) annotation of VAA statements for all political parties or candidates represented in the VAA, b) coming up and testing up to date and relevant VAA statements. The latter task is particularly time critical in electoral systems in which parliaments can be dissolved and snap elections can be called in a relatively short time frame. In Japan, for example, political parties only submit their respective party manifestos shortly before the official election campaign period starts. According to the Japanese election law, the election campaign for the Lower House only lasts for twelve days. In our experience, VAA designers in Japan therefore roughly only have two weeks to complete both mentioned tasks, not even speaking about proper pre-testing. However, in this conference paper we would like to focus on the latter task. With the advancement of Natural Language Processing (NLP) and Machine Learning (ML), efficient techniques from computer and data science can brought into the designing phase of a VAA. Whereas some of those techniques have been applied and tested before, a more general design for a semi-automatically updating VAA system is still lacking and therefore presented here. The design phase for such a VAA system would draw on 1) qualitative social science approaches such as focus groups as well as expert interviews, input from 2 ) political science mainly based on survey research, and on computer science for 3), the development of a pre-trained general language model for Japanese politics as the basis for VAA statement generation. However, the proposed system can be generalized and implemented for other electoral systems as well, not only for the Japanese one. The paper will in detail showcase those three main parts and based on previously conducted research (see references) show how a semi-automatically updating VAA can be achieved.