Hi team!
The client would like to know what is the exact number of clicks that brings search results from, for example, 5th place to 1st place by using Dynamic reranking. Do you have any exact number/percentage of that metrics?
Thanks!
Hey Kaishi!
Dynamic Reranking is calculated using Machine Learning, so changing the order of search results is dependent on the sample size and user behavior. Here’s some math that might help clarify:
For example, if 75% of people choose entity B over entity A, this is a pretty clear preference, so it would require click data from only 4 users before A and B stopped experimentally swapping, and then B is ranked above A for everyone. However, if more users come along and it drops to 60% of users clicking B over A, then it’s a little less clear, and experimental swapping would resume until there have been 30 users. Then B being better than A would be confirmed again, and the experimental swapping would stop.
Hope this helps!
Best,
Daniel
1 Like