I am using LAMBDAMART as the learning to rank algorithm to rank relevant documents for a given query.The model performs well for the queries having lower number(300-400) of candidate documents.However for queries having candidate documents in the range of 1000-5000, the model gives a low training with the same features and ground truth relevance metric. The training score in case of lower documents per query is around 75 % but it falls to around 50 % for queries having higher number of candidate documents.

Kindly suggest the possible reason for this and if I need to use some other algorithm for the queries having more candidate documents.

I am using pyltr module in python and the evaluation metric used is NDCG(K = 40).



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.