Popping the Filter Bubble
summary
A tool that measures diversity in news recommendations is built and released as open source to evaluate actual recommender outputs rather than debate the idea of diversity alone it starts from the claim that recommender design can create filter bubbles and demonstrates measurable signals using 1,100 articles a day as motivation and Elasticsearch SpaCy Wikidata for entity linking and metadata extraction The prototype computes personalization, popularity, complexity and entity/geo counts visualized in Cabana and yields useful comparative insights while admitting metrics are rudimentary not definitive
AI-written summary · about
chapters (10)
- 0:00News recommendation challenges
- 1:44Filter bubbles in news
- 3:07Project goals and data
- 4:34Normative frameworks
- 7:32Expectations and limitations
- 10:27Metrics and visualization
- 13:18Entity-based recommendations
- 15:51Emotive content metrics
- 16:53Future directions
- 18:16Conclusion
related talks
09mOrganizing Chaos: The Growth of Collaborative Filters
40mThe Echo-Chamber of Your Social Media Feed
47mAlgorithmic bias is everywhere (especially at Breeze) - what can we do about it?
43mBuilding your first recommender
46mDiversification in Recommender Systems
18mLearning How To Cook...
16mMaking recommendations explainable
28mBreaking the Chains of Oppressive Software