Pre-training Expanded-SPLADE models on general corpora with higher learning rates improves fine-tuned retrieval effectiveness under both normal and strict pruning, but increases retrieval cost and variance in postings list lengths.
One possible interpretation of metrics is that the MLM model gives higher logit-score-std, non-neg-terms-avg, and non-neg-terms-std when the input data is in-fine-tuning
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.IR 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
The Pre-Training Study of Expanded-SPLADE Models on Web Document Titles
Pre-training Expanded-SPLADE models on general corpora with higher learning rates improves fine-tuned retrieval effectiveness under both normal and strict pruning, but increases retrieval cost and variance in postings list lengths.