search

microsoft / TREC-2019-Deep-Learning

Website for the TREC Deep Learning Track 2019
https://microsoft.github.io/TREC-2019-Deep-Learning/
Creative Commons Attribution 4.0 International
87 stars 28 forks source link

Still can not understand why set two task: document ranking and passage ranking. #8

Closed guotong1988 closed 5 years ago

guotong1988 commented 5 years ago

I am new to text retrieval. The two tasks seems are similar. Thank you very much.

spacemanidol commented 5 years ago

The tasks are very similair by design. We have document and passage retrieval to explore how systems scale across document size and with sparcity of labels. Since all relevance labels are on the passage data, any learned methods would be using transfer learning in the document task. We also believe that the passage ranking task is likley easier since it is more simple to set up(and there is already code in methods in the msmarco task) and thus provided the contestants with a tasks with different difficulty.