Machine Learning in News Media: Case study 24sata.hr
summary
Styria.ai builds tag cleaning, recommenders and personalization for 24sata.hr, claiming concrete lifts: merged ~50,000 legacy tags down to ~8,000 to simplify classification, tag prediction using FastText + time embeddings reaches about 71% recall@10, and a contextual recommender runs a production autoencoder / Doc2Vec pipeline plus triplet CNN embeddings for compact article vectors. A single neural model scores new articles instantly for all users, with GPU batch inference enabling near real‑time recommendations.
AI-written summary · about
chapters (10)
- 0:00Intro
- 6:35How did we do it
- 10:02How it works
- 13:42Data
- 16:14Implementation
- 18:39Problem
- 20:45Results
- 22:56Text Generation
- 25:35Personalization
- 32:50Questions
related talks
35mIndustrial Applications of Machine Learning in Search
30mPerforming Content: Can NLP and Deep Learning algorithms predict reader prefere...
34mApplying Machine Learning Techniques to Detect and Understand Fake News
24mApplications of AI in news and media
38mAutomatically categorizing news articles on nrk.no- 34mApplying Machine Learning Techniques to Detect and Understand Fake News
24mMachine Intelligence gets real
29mAI in Media: Balancing Innovation, Quality, and Responsibility