How LambdaMART works – optimizing product ranking goals(softwaredoug.com)
softwaredoug.com
How LambdaMART works – optimizing product ranking goals
https://softwaredoug.com/blog/2021/11/28/how-lammbamart-works.html
3 comments
I work in the similar space (recsys). Learning-To-Rank sounds amazing on paper but logging "good" (yes, those quotes are warranted) ranked data in production is damn near impossible. Every time you come up with strategy X, someone points out how its statistically biased to hell.
At work I've had to create a model to rank our products and was searching for implementations and working examples of Learn To Tank models (especially with lightGBM examples). There are actually very few tutorials out there. I understand the theory but there are a thousand questions I had about the implementation and the data preparation and could not find the answer on the internet.
It was quite surprised given how much Machine Learning has been blog spammed.
It was quite surprised given how much Machine Learning has been blog spammed.
I recently completed a Learn To Rank (with LightGBM) recommendation system at work and I agree. For how powerful the libraries are you would think there would be many more examples. Have you collected or posted your questions anywhere? I hit an error I didn't understand and posted questions on a related LightGBM issue and LightGBM devs responded with enough detail to at least get me unstuck.