4
$\begingroup$

In search, position of the search result affects the click-through rate a great deal. How do people usually deal with this ? In practice how to remove such bias to create unbiased training data for training learning to rank model ?

$\endgroup$
1
$\begingroup$

What are your output (prediction) and inputs?

The bias in itself is very important in predicting the click-through rate, but definitely less useful if you want to assess your ad copy/ page title in relation to the click-through rate.

One potential way to do that is to model the impact of the position of the ad / (or page, if it's for SEO), and try to normalize the data either by including the impact factor in your model or adjusting the result by the impact factor.

I would suspect it would be a log function, as the number of clicks tend to drop significantly after the top 1/2 of the first page.

| improve this answer | |
$\endgroup$
0
$\begingroup$

Discounted cumulative gain (DCG) is a common way to measure the quality of search results by measuring usefulness, or gain, of a document based on its position in the result list.

| improve this answer | |
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.