According to anserini's experiment, using BM25 can get NDCG@10 at 78.8%, which is much better than the reported number Precision@10 at 45% in https://qdrant.tech/articles/bm42/. Why the BM25 performance in Qdrant in much worse than anserini using Elasticsearch?
What Python version are you on? e.g. python --version
What happened?
https://github.com/castorini/anserini/blob/5eb46b9f9bd563c34deca85a5c7417c068348972/docs/regressions/regressions-beir-v1.0.0-quora.flat.md
According to anserini's experiment, using BM25 can get NDCG@10 at 78.8%, which is much better than the reported number Precision@10 at 45% in https://qdrant.tech/articles/bm42/. Why the BM25 performance in Qdrant in much worse than anserini using Elasticsearch?
What Python version are you on? e.g. python --version
NA
Version
0.2.7 (Latest)
What os are you seeing the problem on?
Linux
Relevant stack traces and/or logs
No response