-1

For e-commerce company, how to pick up features when doing Click Through Rate prediction using logistic regression, SVM or other machine learning models. I tried gender, statistic features from goods tags, and used SVM, NN. but the result was very bad. Is there any suggestions or best practices about the important factors for CTR prediction in e-commerce? THANKS!

Yi De Zhao
  • 1
  • 1

1 Answers1

0

When you use a library like scikit-learn, you can use GridSearchCV the best parameter for the model you're building! You can specify the evaluation metric that you want to optimize! In your case, you need to understand what the evaluation metric is!

Read about it here:

http://scikit-learn.org/stable/modules/generated/sklearn.model_selection.GridSearchCV.html

ababuji
  • 1,683
  • 2
  • 14
  • 39
  • Thanks for your timely reply, it is a great way to determine the important features. I know it depends on the specific problems and the obtained raw data. But I'm still wondering is there any existing known important factors for predicting the CTR. – Yi De Zhao Jun 26 '18 at 05:29
  • @YiDeZhao It is hard to say without performing an Exploratory Data Analysis! These are things which COMPLETELY depend on the data you have! I could say something that is very important at predicting CTR, but something which you do not have a recording for and therefore won't help you predict it! – ababuji Jun 26 '18 at 05:32
  • Got it, thanks again. I'll do more exploratory data analysis first. – Yi De Zhao Jun 26 '18 at 06:04