Homotransphobia Detection in Italian (HODI) @ EVALITA 2023

Official repository of HODI, the shared task on Homotransphobia Detection in Italian at Evalita 2023. See the paper for additional details.

Fill out this form to access the training and testing set, an email notification will be sent with instructions on how to download the data.

---

NOTE! For participants using Windows as OS we suggest unzipping the files with WinRAR zip.

See the task guidelines in this repository and the task web page for more details.

Evaluation Script Installation

Dependencies

To install the required python packages, run:

pip install -r requirements.txt

In the case of problems, try to run pip install --upgrade pip setuptools first.

Evaluation Script Usage

The evaluation script can be used for evaluating the results both of Subtask A and B given a zip file containing submission files as input. For submission files formats check the HODI Guidelines.

Subtask A

For running the evaluation script for Subtask A, you can run something like the following:

python compute_metrics.py \
--submission_path teamName.zip \
--gold_path HODI_2023_train_subtaskA.tsv \
--task a \
--output_path result.tsv 

Subtask B

For running the evaluation script for Subtask B, you can run something like the following:

python compute_metrics.py \
--submission_path teamName.zip \
--gold_path HODI_2023_train_subtaskB.tsv \
--task b \
--output_path result.tsv 

Baseline Script Usage

The baselines are a TF-IDF Logistic Regression for Subtask A and a random baseline for Subtask B.

For running the baseline script, you can run something like the following:

python compute_baseline.py \
--train_path train.tsv \
--test_path test.tsv \
--task a \
--output_path result.tsv 

Reference

If you use the data or code please cite the following paper:

   @inproceedings{hodi2023overview,
    title = {{HODI at EVALITA 2023: Overview of the Homotransphobia
    Detection in Italian Task}},
    author = {Nozza, Debora and Cignarella,
    Alessandra Teresa and Damo, Greta  and Caselli, Tommaso and Patti, Viviana},
    booktitle = {{Proceedings of the Eighth Evaluation Campaign of
    Natural Language Processing and Speech Tools for Italian. Final
    Workshop (EVALITA 2023)}},
    publisher = {CEUR.org},
    year = {2023},
    month = {September},
    address = {Parma, Italy} }

Contacts

If you find issues on the evaluation script, please contact Debora Nozza: Twitter | Github | Webpage