After being asked about how you rank in Google there is a document you can send to those who are versed in mathematics. This documentation went beyond the traditional pointwise scoring functions and introduced a novel setting of group wise scoring functions (GSFs) in the learning-to-rank framework.
The documentation implements GSFs using a deep neural network (DNN) that can efficiently handle large input spaces. If you read the link below you will see that GSFs can include several existing learning-to-rank models as special cases. They compared both GSF models and tree-based models based on a standard learning-to-rank data set. Experimental results show that GSFs significantly benefit several state-of-the-art DNN and tree-based models, due to their ability to combine list wise loss and groupwise scoring functions.
The work compiled now opens up a few interesting future research directions: how to do inference with GSFs in a more principled way using techniques as well as helping readers, to define GSFs using more sophisticated DNN like CNN, rather than simple concatenation, and how to leverage the more advanced DNN matching techniques proposed in.
Read more to understand:
