From Chaos to Insight: automatic proposal classification and recommendation on Brasil Participativo with BERTopic

Description:
Brasil Participativo, Brazil’s national participation platform based on Decidim, collected over 36,000 civic proposals in a few months. Manually organizing this input is unfeasible due to scale, expert knowledge demands, and alignment with official taxonomies.

To address this, we implemented a BERTopic-based pipeline to classify proposals using Brazil’s controlled vocabulary (VCGE), enhanced by expert seed words and LLM validation. The resulting clusters align with institutional categories, enabling automated, detailed reports for decision-makers. This structure also supports cross-process analysis, reveals shared subtopics, and lays the groundwork for future recommendation systems that assist voters and proposal authors. It helps governments deliver broader, theme-based feedback, enhancing transparency and responsiveness.

This lighting talk presents the technical pipeline, results, and how semantic clustering can enrich Decidim-based platforms.

Language: English