Ranking :: Learning to Rank
|Work hours||9-17 UTC|
|Official mentor||Parth Gupta|
As my GSoC project I would like to improve the LETOR framework of Xapian. I intend to make improvements in both the existing modules: features and ranking algorithm and also introduce a new module for rank aggregation. In the features module, I wish to incorporate more features including statistical language models among others. Apart from this, a module for deep learning using which we can extract features from documents in an unsupervised manner will also be implemented within the LETOR framework. I intend to add support for listwise approach based ranking algorithms, in both supervised and semi-supervised setting. In order to combine the ranking output of the 2(existing)+2(proposed)=4 ranking algorithms I intend to integrate a rank aggregation module which would combine the all the rankings and provide the final ranking of documents. The project details section is sub-divided into 3 parts: Features Ranking Algorithms Rank Aggregation.
Overall Framework of the LETOR system: