Table 2.

Summary of the submitted runs. Refer to Figure 2 for a pictorial description.

Run	Name	Description
1	bm25	Run based on the baseline BM25a model using the metadata index
2	bow + rrf	An RRFb combination of BM25, DFRc, and LMDd models computed against the metadata and full-text indices
3	mlm + rrf	An RRF combination of BERT, RoBERTae, and XLNet models applied to run 2
4	bow + mlm + rrf	An RRF combination of runs 2 and 3
5	bow + mlm + lm	A LambdaMART-based model using features from the individual models used to create runs 2 and 3
6	bow + mlm + lr	A logistic regression model using features from the individual models used to create runs 2 and 3
7	bow + mlm + lr + rrf	An RRF combination of runs 2, 3, and 6

^aBM25: Okapi Best Match 25.

^bRRF: reciprocal rank fusion.

^cDFR: divergence from randomness.

^dLMD: language model Dirichlet.

^eRoBERTa: robustly optimized BERT approach.