DeepMerge: Multiple Result List Merging

 

Background

DeepMerge is a neural net implementation of a learning to rank algorithm that merges multiple search result lists. It was produced by C.J. Lee (2015) as part of her Ph.D. degree at the University of Massachusetts, College of Information and Computer Sciences. It is an extension of Sheldon, et. al. (2011) LambdaMerge technique.

DeepMerge uses a deep neural net to process query-document features and shallow neural nets to process query-list and query-vertical features. Learned parameters from each net are combined to produce an optimized merged ranked list corresponding to best P@k and NDCG@k evaluation metrics.

Detailed information about DeepMerge may be found at:

• Chia-Jung Lee, Qingyao Ai, W. Bruce Croft and Daniel Sheldon. An Optimization Framework for Merging Multiple Result Lists. In Proceedings of CIKM, CIKM '15, pages 303-312, 2015.

 

Dataset

The DeepMerge algorithm is implemented using MatLab and may be obtained by emailing Downloads at the Center for Intelligent Information Retrieval at the University of Massachusetts in Amherst.

Data used for DeepMerge development was derived from TREC federated web search (FedWeb13 and FedWeb14) tracks.

 

Acknowledgements

This work was supported in part by the Center for Intelligent Information Retrieval and in part by National Science Foundation grant IIS-1160894. Any opinions expressed in this material are those of the authors and do not necessarily reflect those of the sponsor.