https://github.com/vanessaaleung/ad-ctr-prediction
Ads Click-Through-Rate Prediction
https://github.com/vanessaaleung/ad-ctr-prediction
ctr deep-learning prediction python scikit-learn spark
Last synced: 3 months ago
JSON representation
Ads Click-Through-Rate Prediction
- Host: GitHub
- URL: https://github.com/vanessaaleung/ad-ctr-prediction
- Owner: vanessaaleung
- Created: 2020-06-05T19:01:03.000Z (about 6 years ago)
- Default Branch: master
- Last Pushed: 2020-06-24T01:19:43.000Z (about 6 years ago)
- Last Synced: 2025-11-18T14:37:05.207Z (8 months ago)
- Topics: ctr, deep-learning, prediction, python, scikit-learn, spark
- Language: Jupyter Notebook
- Homepage:
- Size: 494 KB
- Stars: 4
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# Ads Click-Through-Rate Prediction
Predict ads click-through-rate on a user-ads category level. Check the presentation deck [here](https://docs.google.com/presentation/d/1bPIcwX8lUZLoGcVJVeiT4bYTSnznvrhjqSJ-daNgjMA/edit?usp=sharing).
## **Data**
[Ad Display/Click Data on Taobao.com](https://tianchi.aliyun.com/dataset/dataDetail?dataId=56)
This dataset is provided by Alimama and contains 1.14 million users behavior on Taobao.com platform.
| Table| Description| Feature
|---|---|---|
|raw_sample| raw training samples |User ID, Ad ID, nonclk, clk, timestamp|
|ad_feature| Ad’s basic information |Ad ID, campaign ID, Cate ID, Brand|
|user_profile |user profile |User ID, age, gender, etc|
|raw_behavior_log |User behavior log |User ID, btag, cate, brand, timestamp|
## **Data Preprocessing**
We used BigQuery to sample 5 million users from the dataset and merge all the tables.
## **Exploratory Data Analysis**
## **Models**
- Logistic Regression, Random Forest with Sklearn
- Logistic Regression, Random Forest with Spark MLlib
- [Deep Interest Network Model](https://arxiv.org/pdf/1706.06978.pdf)
**Deep Interest Network**
It introduces a local activation unit, with which the representation of user interests varies adaptively
given different candidate ads.