This repository contains codes of Jung et al's EACL 2023 paper titled "Cluster-Guided Label Generation in Extreme Multi-Label Classification"

Requirements

We use python 3.6.5. Please run pip install -r requirement.txt to install python dependencies.

Running XLGen with EUR-Lex dataset

We provide an example of XLGen training/evaluation on EUR-Lex dataset with t5-base model. To test with other benchmark datasets (e.g., Wiki10-31K, AmazonCat-13K, Wiki-500K) or t5 model (e.g., t5-large), simply change the corresponding arguments.

Data Download

bash ./download_data/download_data.sh EUR-Lex

Get Label Clusters

1. Get t5 label representations

   bash ./cluster/run_t5_rep.sh EUR-Lex t5-base

2. Get kmeans label clusters

   bash ./cluster/run_cluster.sh EUR-Lex t5-base

Running XLGen models

1. Train XLGen-base / XLGen-bcl / XGLEN-mcg

   bash ./xlgen/run_train.sh EUR-Lex t5-base base
   bash ./xlgen/run_train.sh EUR-Lex t5-base bcl
   bash ./xlgen/run_train.sh EUR-Lex t5-base mcg

2. Inference for XLGen-base / XLGen-bcl / XGLEN-mcg

   bash ./xlgen/run_test.sh EUR-Lex t5-base base
   bash ./xlgen/run_test.sh EUR-Lex t5-base bcl
   bash ./xlgen/run_test.sh EUR-Lex t5-base mcg

Evaluation

bash ./evaluation/run_fscores.sh EUR-Lex t5-base base
bash ./evaluation/run_fscores.sh EUR-Lex t5-base bcl
bash ./evaluation/run_fscores.sh EUR-Lex t5-base mcg

Citation

@Inproceedings{Jung2023, author = {Taehee Jung and Joo-Kyung Kim and Sungjin Lee and Dongyeop Kang}, title = {Cluster-guided label generation in extreme multi-label classification}, year = {2023}, url = {https://www.amazon.science/publications/cluster-guided-label-generation-in-extreme-multi-label-classification}, booktitle = {EACL 2023}, }