Project overview · Project year 2021
CTR Prediction at Scale
Compared neural and factorization models for click-through-rate prediction on Outbrain data.
My role: BSc thesis author
Outcome
Built and tuned multiple model families with Bayesian optimization on HPC resources.
Model quality under a compute budget
My 2021 bachelor’s thesis, Evaluating Model Tradeoffs for Click Prediction, compared model families on Outbrain click data. I implemented and tuned neural and factorization approaches, using HPC resources and Bayesian optimization to examine the tradeoff between prediction quality and execution time.
The evaluation compared AUC and runtime before and after tuning. Those two axes matter together: a small improvement in ranking quality can come with a substantial increase in computation, and tuning can change the ordering of model families.
Research outcome and scope
The thesis provides the experiment design, model comparisons, and conclusions in Chapters 3–5. It is evidence of experimental modeling and tuning, not an advertising revenue or deployed click-through-rate uplift. The date shown here follows the thesis cover, Ljubljana, 2021; an earlier portfolio label of 2022 was unsupported.
Technology focus
- CTR Modeling
- Deep Learning
- Bayesian Tuning
Evidence and further reading
Evidence reviewed . Project year is separate from this review date.
- Read Evaluating Model Tradeoffs for Click Prediction (PDF) (opens in new tab)
Thesis cover dated 2021; AUC/runtime comparisons in Chapter 4. Reviewed 6 September 2026.