Presentation

The Portuguese Data Augmentation Workshop (PaDAWan) aims to gather the community working on Data Augmentation, particularly employing Large Language Models (LLMs), in Portuguese.

With the advancement of LLMs, many traditional Natural Language Processing (NLP) tasks are being revisited. One traditional key challenge is gathering high-quality data for training and evaluating specific tasks. This has often been the main bottleneck in developing machine learning models. Data augmentation has become a crucial technique for enhancing the performance of these models across various tasks, especially when reliable data are limited. Nowadays, particularly with the use of LLMs, it has become feasible to apply sophisticated text data augmentation techniques effectively.

The use of LLMs is still very restricted due to several factors, such as costs, privacy concerns, latency issues, and other challenges. Given the current scenario, using LLMs to generate synthetic data to train classical models for specific tasks is a viable approach. Moreover, while many works in the industry consider synthetic data, scientific discussions on methods and evaluations are not always aligned with market necessities. 

This workshop aims to delve into the use of LLMs for data augmentation, exploring possible methods, evaluation techniques, and associated ethical considerations. The goal is to bring together both industry professionals and academics to deeply discuss the topic.

We invite researchers to submit papers that discuss challenges and advances in Portuguese data generation, including but not limited to the following topics:

1. Data creation and data labeling

2. Data reformation and anonymization

3. Data contamination and noise

4. Co-annotation

5. Augmented data evaluation and controlled data augmentation

6. Ethics in generated data and unbiased data generation

7. Practical applications or case studies of data augmentation techniques

8. Challenges in Portuguese Synthetic/Augmented Data

 Accepted Papers

• Augmenting Data to Improve the Performance of Recommender Systems - Leticia Freire de Figueiredo, Joel Pinho Lucas and Aline Paes

• Automated Topic Annotation in Brazilian Product Reviews: A Case Study of Adversarial Examples with Sabia-3 -  Lucas Nildaimon dos Santos Silva and Livy Real

• Brazilian Consumer Protection Code: a methodology for a dataset to Question-Answer (QA) Models - Aline Athaydes, Lucas Bulcao, Caio Sacramento, Babacar Mane, Daniela Claro, Marlo Souza and Robespierre Pita. 

• Getting Logic From LLMs: Annotating Natural Language Inference with Sabiá -  Fabiana Avais, Marcos Carreira and Livy Real

• LLM-SEMREL: Towards a Better Coreference Resolution for Portuguese - Evandro Fonseca and Joaquim Neto

• Text extraction from Knowledge Graphs in the Oil and Gas Industry -   Laura Milena Parra Navarro, Elvis A. de Souza and Marco Aurelio Pacheco