Click-through rate

Machine Learning

About

CTR (Click-Through Rate) is a marketing metric that measures the percentage of clicks relative to total impressions, indicating the effectiveness of marketing efforts.


Summary

Overview

To build a CTR prediction model and find a strategy that beats standard classification algorithms.


Challenges

  • Efficient processing of large-scale web log data
  • Handling class imbalance in CTR prediction
  • Encoding high-cardinality categorical features effectively




Dataset

Avazu dataset

The dataset consists of 11 days of data, divided into two parts:

  • Train set: Contains 10 days of data
  • Test set: Contains 1 day of data

The dataset comprises 24 categorical features and includes a total of 40,428,967 training samples.

For more details, please refer to the link above.



Baseline model

LightGBM has been selected as the baseline model, as it ensures fast training speed and stable performance.

In CTR prediction tasks, user characteristics change over time, leading to domain shift and, as a result, performance degradation (Staleness Problem). To address this, a periodic retraining strategy can be applied.

  • Gradient Boosting-based models are frequently used as baselines in various CTR prediction competitions and are known for their stable performance.
  • With its fast training speed, it enables efficient experimentation in model development and allows for quicker retraining during periodic updates, thereby saving overall time and resources.



Solutions

애자일(agile) 의사 결정을 위한 직관적인 MMM(Marketing Mix Modeling) 솔루션
에어브릿지는 수개월에 걸쳐 멜리즈와 긴밀히 협업하며 포스트 프라이버시 시대에 맞는 채널 레벨의 마케팅 성과를 성공적으로 측정하기 위해 실행 가능한 MMM 모델을 개발하였습니다.



Final thoughts

멜리즈는 각 채널의 진정한 기여도를 파악하고 전반적인 마케팅 효율성을 높일 수 있는 예산 분배에 대한 몇 가지 중요한 인사이트를 얻었습니다. 사용자 참여율이 30% 이상 증가하고, 전환율도 눈에 띄게 향상되었습니다. CTR Prediction Click-Through Rate (CTR) prediction is the task of estimating the probability that a displayed item will be clicked. It is widely used in the online advertising industry and various digital marketing fields, contributing to maximizing revenue growth and marketing efficiency by optimizing item display strategies in real-world business scenarios. In addition to these applications, CTR prediction also plays a crucial role in sponsored search and real-time bidding.



References

  1. LightGBM: A Highly Efficient Gradient Boosting Decision Tree