The focus of this shared task will be on "Cross-lingual Classification of Corporate Social Responsibility (CSR) Themes and Topics" and the findings will be presented at EcoNLP workshop at LREC-COLING 2024. 

This shared task will provide the NLP community with data sets in multiple languages (English, French, and simplified Chinese) for CSR news analysis and will shed light on the feasibility of cross-lingual CSR theme detection. The community will also gain insights into fine-grained topic classification for two large CSR themes, viz. Environment (ENV) and Labour and Human Rights (LAB). 

For over 15 years, EcoVadis has been screening a large variety of specialised sources and newspapers to identify CSR-related content and assess it with respect to CSR themes and topics. A key distinguishing element of EcoVadis’ sustainability monitoring platform is the integration of this external input to augment company-provided documentation and data sources. Sustainability analysts assess news items in a variety of languages (e.g., English, Spanish, French) on how they impact the quality and effectiveness of the sustainability management system or reflect positive innovation. The analyzed results are integrated as part of the EcoVadis sustainability rating, and are displayed on the EcoVadis scorecard.

The shared task includes two subtasks:

Task A: Cross-lingual CSR theme recognition (English, French, simplified Chinese)

Task B: Fine-grained classification of CSR topics for Environment (ENV) and Labour and Human Rights (LAB) themes (English)

Task A is framed as multi-class classification, for which participants output for each news article in the different languages a label. Task B is a multi-label multi-class classification problem whereby an article may be assigned multiple topics within the specified theme (e.g., an article with two topics, Air Pollution and Customer Health and Safety, within the ENV theme). While we encourage participants to contribute to both subtasks, they can also decide to participate in Task A or Task B only.

For cross-lingual CSR theme recognition, the annotation will be done at the news item level whereby each url will be classified into one of four CSR themes: ENV, LAB, Fair Business Practices (FBP), or Sustainable Procurement (SUP). The subset of news items labeled with the ENV and LAB themes will be further annotated into one or more CSR topics. The data set shared with participants will include news item urls and the corresponding labels.

To evaluate system performance for Tasks A and B, the prediction of coarse-grained CSR themes and fine-grained CSR topics for environment and Labour and Human rights, we will use standard evaluation measures, including accuracy, precision, recall and F1-score. The latter three will be reported per class label, and micro-averaged within each task and language.