[paper review] Field-aware Factorization Machines for CTR prediction
1
1. Introduction
CTR prediction task
2. POLY2, FM, FFM
POLY2
FM
POLY2 vs. FM
FFM (Field-aware Factorization Machines)
3. Implementation
Optimization
Parallelization
FFM data format
Early Stopping
4. Experiment
모델 성능 비교
Reference
- CTR 예측 알고리즘
https://brunch.co.kr/@kakao-it/84 - Vowpal Wabbit
https://programmers.co.kr/learn/courses/21/lessons/1851 - PITF(Pairwise Interaction Tensor Factorization for Personalized Tag Recommendation) 논문
https://analyticsconsultores.com.mx/wp-content/uploads/2019/03/Pairwise-Interaction-Tensor-Factorization-for-Personalized-Tag-Recommendation-Steffen-Rendle-Lars-Schmidt-Thieme-WSDM-2010.pdf - negative log-likelihood loss
https://ratsgo.github.io/deep%20learning/2017/09/24/loss/ - LM/FM/FFM 수식 설명
https://teamdable.github.io/techblog/CTR-Prediction-FFM?fbclid=IwAR3ZMCY4qAuRcfr-EnbfBmyGN98u2XSkbaMnnds-8nyJ0Rv7imLsdHP8TVM - HOG-WILD! (병렬처리기법)
https://faculty.ucmerced.edu/frusu/Projects/ScalableGD/scalable-hogwild.html - 논문 리뷰 참고
https://greeksharifa.github.io/machine_learning/2020/04/05/FFM/
https://dos-tacos.github.io/paper%20review/FFM/
