set-encoder-base Open Source Model - Free Deployment for Efficient Paragraph Rearrangement Function

Set Encoder Base

Developed by webis

Set-Encoder is a cross-encoder architecture specifically designed for efficient and permutation-invariant paragraph reordering.

Text Embedding

Safetensors

Open Source License:Apache-2.0 #Paragraph Reordering #Permutation Invariance #Cross-Encoder

Downloads 295

Release Time : 7/5/2024

Model Overview

This model is based on a cross-encoder architecture, focusing on listwise paragraph reordering tasks, featuring a permutation-invariant inter-paragraph attention mechanism, suitable for paragraph reordering scenarios in information retrieval.

Model Features

Permutation Invariance

The model features a permutation-invariant inter-paragraph attention mechanism, effectively handling changes in paragraph order.

Efficient Reordering

Optimized for paragraph reordering tasks, capable of efficiently addressing listwise paragraph ranking problems.

Cross-Encoder Architecture

Adopts a cross-encoder architecture, enabling simultaneous encoding of queries and paragraphs for better relevance judgment.

Model Capabilities

Paragraph Relevance Assessment

Information Retrieval Result Reordering

Text Ranking

Use Cases

Information Retrieval

Search Engine Result Reordering

Reorders initial search engine results to improve relevance

Performs excellently on the TREC DL 19/20 dataset

Document Retrieval System

Ranks candidate paragraphs by relevance in document retrieval systems

🚀 Set-Encoder: Permutation-Invariant Inter-Passage Attention for Listwise Passage Re-Ranking with Cross-Encoders

This project presents a cross - encoder architecture for efficient and permutation - invariant passage re - ranking. It's introduced in the paper Set - Encoder: Permutation - Invariant Inter - Passage Attention for Listwise Passage Re - Ranking with Cross - Encoders, aiming to solve the problem of passage re - ranking in an effective way.

🚀 Quick Start

To reproduce the results, you can run the following command using the Lightning IR library and the configuration files from the linked repository:

lightning-ir re_rank --config ./configs/re-rank.yaml --model.model_name_or_path <MODEL_NAME>

✨ Features

Cross - encoder Architecture: Designed for efficient and permutation - invariant passage re - ranking.
Pre - trained Models: We offer pre - trained models for general - purpose re - ranking.

📦 Installation

The README doesn't provide specific installation steps, so this section is skipped.

💻 Usage Examples

The provided command is a usage example for reproducing the results:

lightning-ir re_rank --config ./configs/re-rank.yaml --model.model_name_or_path <MODEL_NAME>

📚 Documentation

Pre - trained Models

We provide the following pre - trained models for general - purpose re - ranking. The performance on TREC DL 19 and TREC DL 20 (nDCG@10) is shown in the table below:

Model Name	TREC DL 19 (BM25)	TREC DL 20 (BM25)	TREC DL 19 (ColBERTv2)	TREC DL 20 (ColBERTv2)
[webis/set - encoder - base](https://huggingface.co/webis/set - encoder - base)	0.746	0.704	0.781	0.768
[webis/set - encoder - large](https://huggingface.co/webis/set - encoder - large)	0.750	0.722	0.789	0.791

Model Information

Property	Details
Model Type	Cross - encoder
Base Model	google/electra - base - discriminator
Library Name	lightning - ir
Pipeline Tag	text - ranking

Code Repository

The code for this project is available at: https://github.com/webis - de/set - encoder

🔧 Technical Details

The README doesn't provide specific technical details, so this section is skipped.

📄 License

This project is licensed under the Apache 2.0 license.

📚 Citation

If you use this code or the models in your research, please cite our paper:

@InProceedings{schlatt:2025,
  address =                  {Berlin Heidelberg New York},
  author =                   {Ferdinand Schlatt and Maik Fr{\"o}be and Harrisen Scells and Shengyao Zhuang and Bevan Koopman and Guido Zuccon and Benno Stein and Martin Potthast and Matthias Hagen},
  booktitle =                {Advances in Information Retrieval. 47th European Conference on IR Research (ECIR 2025)},
  doi =                      {10.1007/978-3-031-88711-6_1},
  month =                    apr,
  publisher =                {Springer},
  series =                   {Lecture Notes in Computer Science},
  site =                     {Lucca, Italy},
  title =                    {{Set-Encoder: Permutation-Invariant Inter-Passage Attention for Listwise Passage Re-Ranking with Cross-Encoders}},
  year =                     2025
}

Featured Recommended AI Models

Empowering the Future, Your AI Solution Knowledge Base

English 简体中文繁體中文にほんご