HyILR

KO | EN

gitlite — search

#python #react #typescript #golang #machine-learning #deep-learning #linux #macos #java #windows #cli #kubernetes #awesome-list #mcp

(★ 16)

Code for the paper "HyILR: Hyperbolic Instance-Specific Local Relationships for Hierarchical Text Classification" acepted in ACL Student Research Workshop 2025

HyILR Latest Version Download

Download Latest Version (.zip)

// repository documentation

Was this content helpful?

★ 0 (0 ratings)

# HyILR (Hyperbolic Instance-Specific Local Relationships)
## Requirements
- Python>=3.6
- torch>=1.13.1
- transformers>=4.30.2
- numpy>=1.23.0

## Data
- All datasets are publically available and can be accessed at [WOS](https://github.com/kk7nc/HDLTex), [RCV1-V2](https://trec.nist.gov/data/reuters/reuters.html), [BGC](https://www.inf.uni-hamburg.de/en/inst/ab/lt/resources/data/blurb-genre-collection.html) [NYT](https://catalog.ldc.upenn.edu/LDC2008T19).
- After accessing the dataset, run the scripts in the folder `preprocess` for each dataset separately to obtain tokenized version of dataset and the related files. These will be added in the `data/x` folder where x is the name of dataset with possible choices as: wos, bgc, rcv and nyt.
- For reference, we have added tokenized versions of the WOS, BGC, and NYT datasets along with their related files in the `data` folder. Due to size constraints in GitHub, we could not upload the tokenized version of the RCV1-V2 dataset, which exceeds 400 MB in size.
- Detailed steps regarding how to obtain and preprocess each dataset are mentioned in the readme file of `preprocess` folder 
## Train HyILR 
`python train.py --name t1_hyilr --batch 10 --data wos --cl_loss 1 --cl_temp 0.07 --cl_wt 1` 

Some Important arguments: 
- `--name` name of directory in which your model will be saved. For e.g. the above model will be saved in `..HyILR/data/wos/t1_hyilr`
- `--batch` batch_size for training. We set it to 10 for all datasets.
- `--data` name of dataset directory which contains your data and related files. Possible choices are `wos`, `rcv`, `bgc` and `nyt`.
- `--cl_loss` Set to 1 for using contrastive loss in Lorentz hyperbolic space
- `--cl_temp` Temperature for the contarstive loss. We use a value of 0.07 for all datasets
- `--cl_wt` weight for contrastive loss. We use the following weights for the contrastive loss across datasets: `WOS:0.3` , `RCV1-V2:0.4`, `BGC:0.4`, and `NYT:0.6`

### For Exponential map transformation and Geodesic distance calculation in Lorentz hyeprbolic space
- The code for operations in lorentz geometry is in the script `lorentz.py` which is obtained from [meru](https://github.com/facebookresearch/meru/blob/main/meru/lorentz.py) repository.
- Specifically we use the functions  `exp_map0()` and `pairwise_dist()` provided in  the script `lorentz.py`

### For Contrastive loss in Lorentz hyperbolic space
- The code for the contrastive loss and **hierarchy-aware negative sampling** is in the script `criterion.py`, where the loss is defined as the PyTorch class `CLLoss`.

## Test
To run the trained model on test set run the script `test.py` 
`python test.py --name t1_hyilr --data wos --extra _macro` 
Some Important arguments
- `--name` name of the directory which contains the saved checkpoint. The checkpoint is saved in `../HyILR/data/wos/` when working with WOS dataset
- `--data` name of dataset directory which contains your data and related files
- `--extra` two checkpoints are kept based on macro-F1 and micro-F1 respectively. The possible choices are `_macro` and `_micro` to choose from the two checkpoints

## Citation
If you find our work helpful, please cite it using the following BibTeX entry:
```bibtex
@inproceedings{kumar-toshniwal-2025-hyilr,
    title = "{H}y{ILR}: Hyperbolic Instance-Specific Local Relationships for Hierarchical Text Classification",
    author = "Kumar, Ashish  and
      Toshniwal, Durga",
    editor = "Zhao, Jin  and
      Wang, Mingyang  and
      Liu, Zhu",
    booktitle = "Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 4: Student Research Workshop)",
    month = jul,
    year = "2025",
    address = "Vienna, Austria",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2025.acl-srw.63/",
    doi = "10.18653/v1/2025.acl-srw.63",
    pages = "872--883",
    ISBN = "979-8-89176-254-1",
   
}

HyILR

HyILR Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

HyILR

HyILR Latest Version Download

Today's picks

Was this content helpful?

Recent Feedback

Download README