Multi-Lingual ESG Impact Duration Inference (ML-ESG-3)

Introduction

In FinNLP-2022, we proposed a FinSim4-ESG shared task, which is related to the topic of environmental, social, and corporate governance (ESG). To continue exploring ESG topics, FinNLP@IJCAI-2023 shared a new dataset for the FinNLP community to explore the multi-lingual ESG issue identification task. Based on the MSCI ESG rating guidelines, ESG-related news can be classified into 35 ESG key issues. The system needs to be aware of the ESG issues of each article. We used multilingual news articles as the raw material, and conduct annotation on the articles. The target languages include English, Chinese, Japanese, and French. Note that, in the Chinese dataset, we merge issues in SASB Standard into MSCI guidelines. In ML-ESG-2, we introduced a new task to continue the discussion on ESG rating. The task we proposed is ESG impact type identification. That is, the models need to identify the given news is an opportunity or risk from the ESG aspect. 

In ML-ESG-3, We seek to determine the duration (length) of the impact an event in the news article might have on the company. More details will be released soon. The overview of this shared task series (Chinese) can be found in [1].

Dataset

The design of the task may be slightly different among all subsets. Below are the introductions of the task design. 

Chinese: Based on the distinction between short-term and long-term defined, we present three labels: "Less than 2 years", "2 to 5 years", and "More than 5 years". [1]

English & French: There are two annotations in this dataset, "Impact Level" and "Impact Length." Impact Length was selected from "Less than 2 years", "2 to 5 years", and "More than 5 years", which is the same as the Chinese dataset. Impact Level qualifies the opportunity or risk as being of "low", "medium" or "high." Please refer to our guidelines for more details. [2] 

Japanese: Based on the distinction between short-term and long-term defined, we present three labels: "Less than 2 years", "2 to 5 years", and "More than 5 years". [1] The labels proposed in [3] are also shared for further exploration. 

Korean: In this dataset, there are two annotations: 'Impact Type' and 'Impact Length.' 'Impact Type' is categorized as 'opportunity,' 'risk,' or 'cannot distinguish,' while 'Impact Length' is classified into three durations: 'less than 2 years,' '2 to 5 years,' and 'more than 5 years.' For more detailed information, please refer to the following source [1]

[1] Yu-Min Tseng, Chung-Chi Chen, Hen-Hsen Huang, and Hsin-Hsi Chen. 2023. DynamicESG: A Dataset for Dynamically Unearthing ESG Ratings from News Articles. In Proceedings of The 32nd ACM International Conference on Information and Knowledge Management (CIKM'23).

[2] ML-ESG 2024 for Social Good (ESG) - 3rd Edition Guidelines by 3DS Outscale 

[3] Naoki Kannan and Yohei Seki. 2023. Textual Evidence Extraction for ESG Scores. In Proceedings of The 5th Workshop on Financial Technology and Natural Language Processing (FinNLP).

Important Dates: Time zone: Anywhere On Earth (AOE)

• Registration Open: Nov. 25th, 2023

• Training set release: Dec. 13th, 2023

• Test set release: Feb. 8th, 2024

• System's outputs submission deadline (Registration Close): Feb. 20th

• Release of results: Feb. 25th

• Shared task paper submissions due: March 7th

• Notification: March 20th

• Camera-Ready Version of Shared Task Paper Due: March 27th

  • Shared Task Paper Submission System: https://softconf.com/lrec-coling2024/finnlp-kdf2024/

Policies

• The LREC-COLING template MUST be used for your submission(s).

• The reviewing process will be single-blind. Accepted papers proceedings will be published at ACL Anthology.

• Shared task participants will be asked to review other teams' papers during the review period.  

• Submissions must be in electronic form using the paper submission software linked above. 

• At least one author of each accepted paper should register and present their work (either online or in-person) in FinNLP-KDF-2024. Papers with “No Show” may be redacted. Authors will be required to agree to this requirement at the time of submission. 

Official Evaluation

Shared Task Organizers

• Chung-Chi Chen - AIRC, AIST, Japan

  • Yu-Min Tseng - Department of Computer Science and Information Engineering, National Taiwan University, Taiwan

• Juyeon KANG - 3DS Outscale, France

• Anaïs Lhuissier - 3DS Outscale, France

• Hanwool Lee - NCSOFT, South Korea

• Min-Yuh Day - Graduate Institute of Information Management, National Taipei University, Taiwan

• Teng-Tsai Tu - Graduate Institute of Information Management, National Taipei University, Taiwan

  • Yohei Seki - University of Tsukuba, Japan

• Hsin-Hsi Chen - Department of Computer Science and Information Engineering, National Taiwan University, Taiwan