Category

Research

Organization

DiCEM programme, UPF Barcelona (2025)

Decidim Barcelona Topics 2016-2024

Visualizing the top 20 topics in 31,775 proposals from over 8 years of open data from decidim.barcelona

What were the most recurring topic in the participatory platform of Barcelona since its launch until today? This web app visualizes the top 20 recurring topics in from text descriptions of more than 31,000 citizen proposals, from decidim.barcelona open data, modeled with BERTopic and an open and public language model from Projecte AINA, fine-tuned for identifying sentence similarity in Catalan, trained and published by the Barcelona Supercomputing Center (BSC).

Large Project Gallery Image #1
Large Project Gallery Image #1
Large Project Gallery Image #1
Project Gallery Image for 50% width of the screen #1
Project Gallery Image for 50% width of the screen #1
Project Gallery Image for 50% width of the screen #1
Project Gallery Image for 50% width of the screen #1
Project Gallery Image for 50% width of the screen #2
Project Gallery Image for 50% width of the screen #2

Exploring the possibilites of guided topic modelling, with BERTopic, to assess the most recurrent citizen needs and ideas.

This inmmersive “scrollytelling” interface was built with the Cosmos WebGL library, Scrollama and Chart.js. Topic modelling was previously processed in Python, using custom Google Colab notebook (Decidim Topic Modeller) to parse, clean and process sentence embeddings and topic clusters, using a customizable BERTopic pipeline for dimensionality reduction (UMAP), clusterization (HDBSAN) and topic extraction (cTF-IDF + Bag of Words from clusters)

  • The image featured in the carousel #1
  • The image featured in the carousel #2
  • The image featured in the carousel #3
  • The image featured in the carousel #4
  • The image featured in the carousel #1
  • The image featured in the carousel #2
  • The image featured in the carousel #3
  • The image featured in the carousel #4
  • The image featured in the carousel #1
  • The image featured in the carousel #2
  • The image featured in the carousel #3
  • The image featured in the carousel #4

Tools:

Python: Sentence-Transformers, BERTopic, UMAP, HDBSCAN, Plotly, NLTK, Pandas. JavaScript/TypeScript: Cosmos, Scrollama, Chart.js, Vite, HTML, CSS; (Other open source tools) Cosmograph, Gephi, Open Refine. Other open source tools: Cosmograph, Gephi, Open Refine, Vercel Models: projecte-aina/ST-NLI-ca_paraphrase-multilingual-mpnet-base, published by the BSC-CNS Barcelona Supercomputing Center, for embeddings generation. Perplexity AI and Anthropic Claude Sonnet for coding assistance.

Credits:

Created by: Diego Arredondo Ortiz Prototyped within the course Data Analysis and Information Visualization Concerning Global Issues within the Digital Culture and Emerging Media (DiCEM) programme at Universitat Pompeu Fabra. Barcelona, 2025.

Create a free website with Framer, the website builder loved by startups, designers and agencies.