Ergun Biçici
Researcher in Computer Science and Engineering
PhD in Computer Engineering from
Department of Computer Engineering, Koç University.
English Word of the Day
January 6, 2024
Potential for Improvement of Sentence Translations
Instance Weighting in Neural Networks for Click-Through Rate Prediction
Neural Network Calibration for CTR Prediction
Efficiently Sampling in Neural Network Training for Click-Through Rate Prediction
Finding efficient downsampling techniques has become more crucial as the training datasets for advertisement click-through rate (CTR) prediction models are growing to billions in size. We present efficient downsampling to sample CTR datasets with goals of faster training and limited decrease in the performance. We present encouraging results demonstrating the effectiveness of our approach on two publicly available CTR prediction datasets and compare efficient downsampling with stratified random downsampling.
Residual Fusion Models with Neural Networks for CTR Prediction
Ergun Biçici. Residual Fusion Models with Neural Networks for CTR Prediction. 2023 8th International Conference on Computer Science and Engineering (UBMK), Burdur, Turkiye, 2023, pp. 01-04, doi: 10.1109/UBMK59864.2023.10286706. URL: https://ieeexplore.ieee.org/document/10286706
No single prediction model achieves the best performance on all datasets and we are better off combining the strengths of different models for each task. Residual fusion learning is a two step combination method that trains a second model on the residual of the target from the first model's prediction. In the final phase, the predictions of both of these models are added. We use gradient boosting decision trees (GBDT) and neural networks for the initial model in residual fusion and compare three GBDT models and four neural network models. We introduce residual fusion with two different neural network models and show that we can achieve AUC gains that reach 0.95%.
September 19, 2023
Power Loss Function in Neural Networks for Predicting Click-Through Rate
Ergun Biçici. Power Loss Function in Neural Networks for Predicting Click-Through Rate. In Proc. of the 17th ACM Conference on Recommender Systems (RecSys), Singapore, September 2023. URL: https://dl.acm.org/doi/10.1145/3604915.3610658, https://www.growkudos.com/publications/10.1145%252F3604915.3610658/reader
Loss functions guide machine learning models towards concentrating on the error most important to improve upon. We introduce power loss functions for neural networks and apply them on imbalanced click-through rate datasets. Power loss functions decrease the loss for confident predictions and increase the loss for error-prone predictions. They improve both AUC and F1 and produce better calibrated results. We obtain improvements in the results on four different classifiers and on two different datasets. We obtain significant improvements in AUC that reach 0.44% for DeepFM on the Avazu dataset.Calibrating Neural Networks for CTR Prediction (Reklam Tıklama Oranını Tahmin Etmek için Sinir Ağlarının Kalibrasyonu)
Ergun Biçici and Hasan Saribaş. Calibrating Neural Networks for CTR Prediction (Reklam Tıklama Oranını Tahmin Etmek için Sinir Ağlarının Kalibrasyonu). In Proc. of the 31st Signal Processing and Communications Applications Conference (SIU), Istanbul, Turkey, July 2023. URL: https://ieeexplore.ieee.org/document/10223867
Calibration methods fix the prediction errors of a machine learning model after it is trained and enable more robust and more confident prediction. We implement isotonic regression, Platt's scaling, neural networks, spline regression, and temperature scaling as calibration techniques on the prediction of click-through rate (CTR), which is an unbalanced task. We compare the improvements on using 3 neural network based CTR prediction models, Masknet, DeepFM, and DCNv2, on the publicly available CTR dataset Avazu. Our results demonstrate that isotonic and spline regression methods improve the most and isotonic regression is the fastest method.
Kalibrasyon yöntemleri, bir makine öğrenimi modelinin eğitildikten sonraki tahmin hatalarını düzelterek daha gürbüz ve daha güvenli tahmin yapılmasını sağlar. Bu çalışmada, dengesiz etiketlere sahip bir görev olan reklam tıklama oranının (RTO) tahmininde, kalibrasyon teknikleri olarak izotonik regresyon, Platt'ın ölçeklendirmesi, sinir ağları, eğri regresyonu, ve sıcaklık ölçeklendirmesi uygulanmıştır. Bu kalibrasyon yöntemlerinin karşılaştırılması için halka açık veri seti olan Avazu üzerinde, MaskNet, DeepFM, ve DCNv2 olmak üzere sinir ağı tabanlı üç farklı RTO tahmin modeli kullanılmıştır. Yapılan deneyler, izotonik ve eğri regresyon yöntemlerinin en iyi şekilde performansı arttırdığını ayrıca izotonik regresyonun en hızlı yöntem olduğunu göstermektedir.