Multi-objective Learning to Rank through Multi-objective Optimization with LightGBM

Introduction

MO-LightGBM is a gradient boosting framework based on decision tree algorithms, used for multi-label learning to rank tasks. It is based on LightGBM.

Requirements

• Python 3.8 or higher
• matplotlib
• seaborn

Installation

1. Install blas and lapack,

   1. In mac: $brew install openblas $brew install lapack
   2. In Debian-based systems like Ubuntu: $sudo apt install libblas-dev liblapack-dev
   3. In Red Hat systems like CentOS and Fedora: $sudo yum install lapack-devel blas-devel

2. From the scs root, build the package, install cmake if needed

   cd LightGBM/external_libs/scs/ rm -r build mkdir build cd build cmake .. make -j$(nproc) cd ../../../../

3. From the ecos root, build the package

   cd LightGBM/external_libs/ecos.2.0.8/ rm -r build mkdir build cd build cmake .. make -j$(nproc) cd ../../../../

4. Now build LightGBM

cd LightGBM rm -r build mkdir build cd build cmake .. make -j$(nproc) cd ../../ It is fully tested with Red Hat systems.

Usage

Assumptions about Dataset

1. The dataset should be in csv or tsv format, not in the libsvm format.
2. The csv/tsv file should have header that names each column

Process for bi-objective experiments

1. Prepare the <dataset_name>_config.yml
2. Generate configure files for baseline experiment

python baselines.py <dataset_name>_config.yml 3. Go to the baseline_results folder and run the experiments

cd results_${dataset}/lambdarank/baseline_results sh run_experiment.sh 4. Generate configure files for biobjective experiments for preference based methods

python biobjectives.py <dataset_name>_config.yml 5. Go to the biobjective_results folder, and, either run all the experiments or run for individual combinator

cd results_${dataset}/lambdarank/biobjective_results sh run_experiment.sh # will run for the all combinators sh epo_search_run_experiment.sh # will run only for epo_search 6. Generate configure files for biobjective experiments for constraints based methods

python biobjectives_ec.py <dataset_name>_config.yml 7. Go to the biobjective_results folder, and, either run all the experiments or run for individual combinator

cd results_${dataset}/lambdarank/biobjective_results sh ec_run_experiment.sh # will run for the all combinators 8. [optional] Run plotting script 1. Plot bound and ray in one figure python plotbiobjectives.py <datasetname>_config.yml ```

Example for bi-objective experiment

```

dataset=istella

root=/home/user/MO-LightGBM/

folder=${root}/example/

cd ${folder}

python baselines.py ${dataset}_config.yml

cd results${dataset}/lambdarank/baselineresults

sh run_experiment.sh

cd ${folder}

python biobjectives.py ${dataset}_config.yml

cd results${dataset}/lambdarank/biobjectiveresults

sh run_experiment.sh

cd ${folder}

python biobjectivesec.py ${dataset}config.yml

cd results${dataset}/lambdarank/biobjectiveresults

sh ecrunexperiment.sh

cd ${folder}

python plotbiobjectives.py ${dataset}config.yml

```

//: # ()

//: # ()

//: # ()

//: # ()

//: # () //: # ()

//: # ()

//: # ()

//: # ()

//: # () //: # ()

//: # ()

//: # ()

//: # ()

//: # ()

Citation

If you use this work, or otherwise found our work valuable, please consider citing the papers:

MO-LightGBM: A Library for Multi-objective Learning to Rank with LightGBM

Multi-Label Learning to Rank through Multi-Objective Optimization

``` @inproceedings{chaosheng2025mo-lightgbm, title={MO-LightGBM: A Library for Multi-objective Learning to Rank with LightGBM}, author={Dong, Chaosheng and Momma, Michinari}, booktitle={Proceedings of the 48th International ACM SIGIR Conference on Research and Development in Information Retrieval}, year={2025} }

@inproceedings{mahapatra2023multi, title={Multi-label learning to rank through multi-objective optimization}, author={Mahapatra, Debabrata and Dong, Chaosheng and Chen, Yetian and Momma, Michinari}, booktitle={Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining}, year={2023} }

```

Security

See CONTRIBUTING for more information.

License

This project is licensed under the Apache-2.0 License.