ranxhub

Run

RanxHub ID: msmarco-passage/trec-dl-2020/ranxhub/slim++
Version: 1.0
Description: SLIM++ run reproduced using Pyserini.
Tags: Retrieval
Date: 25 May 2023
Run Authors: Elias Bassani
From Paper:
Paper Authors:

Model

Name: SLIM++
Description: SLIM maps each contextualized token vector to a sparse, high-dimensional lexical space before performing late interaction between sparse token embeddings. It employs a two-stage retrieval architecture that includes inverted index retrieval followed by a score refinement module to approximate the sparsified late interaction, which is fully compatible with off-the-shelf lexical search libraries such as Lucene. SLIM++ enhances SLIM with cross-encoder distillation and hard negative mining.
Tags: BERT · Knowledge Distillation · Late Interactions · Sparse Representations
Paper: SLIM: Sparsified Late Interaction for Multi-Vector Retrieval with Inverted Indexes
Authors: Minghan Li · Sheng-Chieh Lin · Xueguang Ma · Jimmy Lin

Results

NDCG@10 MAP@1000 Recall@1000
0.7021451024094133 0.49063745716330454 0.8550606933791138