Enhancing the design of voting advice applications with BERT language model

Area of Science:

• Computational Social Science
• Natural Language Processing
• Political Science

Background:

• Voting Advice Applications (VAAs) are popular tools for voters, with 30% considering their recommendations.
• VAA policy statements are manually created by analyzing large volumes of political data, a time-consuming process.
• Manual analysis of party manifestos and political data for VAA statement formulation is limited by time constraints.

Purpose of the Study:

• To propose an automated system that assists Voting Advice Application designers in creating and revising policy statements.
• To leverage natural language processing and machine learning to streamline the VAA policy statement generation process.
• To provide objective, bias-free suggestions for VAA policy statements, enhancing efficiency.

Main Methods:

• Utilized pre-trained language models, specifically BERT, for processing politics-related textual data.
• Applied machine learning techniques to analyze party manifestos and online political discourse (e.g., YouTube comments).
• Developed a system to generate relevant suggestions for VAA policy statements based on data analysis.

Main Results:

• The system successfully processed Japanese party manifestos and YouTube comments, along with VAA policy statements from Japanese and European VAAs.
• The BERT-based system demonstrated capability in capturing contextual information within political documents.
• The system's output provided valuable, objective suggestions for updating VAA policy statements, though not fully replacing manual review.

Conclusions:

• The proposed system offers a significant aid to VAA designers, improving the efficiency and objectivity of policy statement formulation.
• Automated analysis using language models can help overcome the time limitations inherent in manual VAA design.
• The system contributes to creating more accurate and less biased Voting Advice Applications, enhancing their utility for voters.