Revert "brief description."

This reverts commit 0b3817a.

Author: trivialfis

doc/tutorials/learning_to_rank.rst

@@ -93,8 +93,6 @@ Position Bias
 *************
 Real relevance degree for query result is difficult to obtain as it often requires human judegs to examine the content of query results. When such labeled data is absent, we might want to train the model on ground truth data like user clicks. Another upside of using click data directly is that it can relect the up-to-date relevance status `[1] <#references>`__. However, user clicks are often nosiy and biased as users tend to choose results displayed in higher position. To ameliorate this issue, XGBoost implements the ``Unbiased LambdaMART`` `[4] <#references>`__ algorithm to debias the position-dependent click data. The feature can be enabled by the ``lambdarank_unbiased`` parameter, see :ref:`ltr-param` for related options and :ref:`sphx_glr_python_examples_learning_to_rank.py` for a worked example with simulated user clicks.
 
-For details about the bias estimation, users are encouraged to visit the original paper. We will briefly go through the intuition behind the algorithm in this section. When conducting a study, we might accidentally introduce selection bias to the experiment by selecting subjects with certain conditions not being distributed evenly. For instance, in a medical study, the group of patients selected to be part of the treatment group might consist of only a particular range of age. If this happens, then we have introduced age as a confounding variable in the study. The effect (or the lack thereof) might be caused by age instead of by the effect of the treatment. However, if we know the confounder (the age), we can use the (extended) inverse propensity score to re-weight the data and remove the bias introduced in the study. The propensity score is defined as the probability of a patient receiving the treatment given all the relevant conditions. In the context of learning to rank, the position bias is cast as a form of selection bias, a user might not observe a document if it's placed at the end of the list. As a result, that document is not selected to be judged by the user and cannot be clicked.
-
 ****
 Loss
 ****