Performing Content: Can NLP and Deep Learning algorithms predict reader prefere...
summary
Predicts average article dwell time for a regional newspaper using a two‑step pipeline: a GAM on section and length then a fine‑tuned German BERT on residuals to isolate content effects. Trained on ~36k articles from Nov 2019–Sep 2020, explained variance ranges from ~5% in culture to 45–55% in regional sections. Uses SHAP explanations to highlight which phrases increase or decrease dwell time. Prototype stage, not yet in production.
AI-written summary · about
chapters (8)
- 0:00Start
- 1:53General Newsroom Challenges
- 7:23Transfer Learning
- 9:03NLP methods
- 9:58Modelling Expectations
- 10:58Modelling Results
- 12:37Isolating content effect on Dwell Time
- 15:32Black Box enlightenment with SHAP
related talks
36mMachine Learning in News Media: Case study 24sata.hr
25mBeyond scraping: how to use machine learning when you're not sure where to start
49mDeep Learning Applied to Natural Language Processing (NLP)
13mBeyond Surveys: Real-Time User Satisfaction Prediction
38mAutomatically categorizing news articles on nrk.no
22mEmpowering journalism-approaches to automatic text summarization
29mA content-based recommender system for financial news
35mIndustrial Applications of Machine Learning in Search