/SparseRetrieval

Sparse Retrieval using Contextualized Term Weights

Primary LanguagePython

SparseRetrieval

This repo contains information on different experiments conducted in the area of Sparse Information Retrieval using contextualised term weights. We have used the Anserini IR toolkit to perform the experiments.

Sparse Retrieval Systems

Results

Term Weighting Expansion MRR@10 Notes
BM25 None 0.1874
DeepCT None 0.2362 No hyperparameter tuning
DeepCT None 0.2425 Hyper-params fine-tuned: k1=18, b=0.7
DeepImpact doc2query–T5 0.3253
COIL None 0.3536 token dimension d = 32
uniCOIL doc2query–T5 0.3515