A search result can have parts, similar to Google search results.
The Learning to Rank model is currently unaware of parts during training or at query (inference) time.
A nice change to make would be to make the model aware of parts. We could either exclude parts clicks from the training data set, or make the model aware of parts during training and inference time. Either of these options could make it easier for the model to be trained; the latter option may have a good impact on search ranking performance, since many popular documents have parts.
A search result can have parts, similar to Google search results.
The Learning to Rank model is currently unaware of parts during training or at query (inference) time.
A nice change to make would be to make the model aware of parts. We could either exclude parts clicks from the training data set, or make the model aware of parts during training and inference time. Either of these options could make it easier for the model to be trained; the latter option may have a good impact on search ranking performance, since many popular documents have parts.