VeriPromise ESG 競賽的目標是什麼？

開發能夠自動識別、分析和驗證企業永續承諾的 AI 系統，解決企業漂綠 (Greenwashing) 問題，提升 ESG 報告可信度。

競賽提供什麼數據集？

提供 VeriPromiseESG4K 數據集，包含 4,000 筆來自臺灣 50 指數成分股的真實 ESG 報告標註數據。

如何報名參加 AI CUP 2026？

請於 2026 年 2 月起前往 AI CUP 報名系統 (https://go.aicup.tw/) 進行報名，隊伍人數限制為 1-5 人。

AI CUP 2026 - VeriPromiseESG | 永續承諾驗證競賽

Competition Overview

In the era of global ESG (Environmental, Social, and Governance) focus, the authenticity and credibility of corporate sustainability reports are increasingly important. This competition aims to establish an automated sustainability commitment verification system using AI technology.

          Full Name:
          VeriPromise ESG 2026 - ESG Promise Verification Competition


          Objective:
          Develop an AI system capable of automatically identifying,
            analyzing, and verifying corporate sustainability commitments
            through four core tasks (promise recognition, evidence support,
            clarity assessment, and timeline prediction) to comprehensively
            evaluate the authenticity and credibility of ESG reports.
        

Competition Tasks

4,000

Annotated Data

Leading Companies

Industry Sectors

Practical Application

Address corporate 'Greenwashing' issues, enhance ESG report credibility, and help investors and stakeholders make more informed decisions.

Technical Challenge

Combine natural language processing, large language models, and multi-task learning to complete four challenge tasks: promise recognition, evidence linkage, clarity assessment, and timeline prediction.

International Collaboration

Co-hosted by top academic institutions from Taiwan and Japan, featured as an NTCIR-19 International Track project, providing high-quality multilingual datasets to promote global ESG AI research.

Competition Tasks

This competition comprises four core subtasks, covering the complete ESG report verification process from promise recognition to evidence assessment, clarity analysis, and timeline prediction

範例程式碼可參考

中文版檔案英文版檔案

Subtask 1: Commitment Classification

Objective: Determine whether a given sentence expresses a clear corporate commitment to future actions

Output Categories

Yes: Statements containing explicit commitments
No: General statements without commitments

Evaluation Metric

F1-Score (harmonic mean of precision and recall)

Examples

              ✅ Promise:
              "We commit to achieving carbon neutrality by 2030"

              ❌ Non-Promise:
              "We value the importance of environmental protection"
            

🔗 Subtask 2: Evidence Identification

Objective: Determine whether identified promise statements are accompanied by specific action plans or supporting evidence

Output Categories

Yes: Promise has concrete evidence support
No: Promise lacks specific evidence
N/A: Statement is not a promise, not applicable

Evaluation Metric

F1-Score (semantic association judgment capability)

Example

              Promise:
              "Promote low-carbon value chain transformation, continuously
                strengthen supplier energy-saving, carbon reduction, water
                conservation, and waste reduction guidance"

              Evidence:
              "Require setting medium and long-term reduction targets and
                proposing specific actions" → Evidence Supported
            

💡 Subtask 3: Clarity Classification

Objective: Assess whether promise statements are semantically clear without ambiguous language, identifying potential 'greenwashing' risks

Output Categories

Clear: Semantically explicit and verifiable
Not Clear: Semantically ambiguous and difficult to quantify
Misleading: Potentially misleading statements
N/A: Statement is not a promise, not applicable

Evaluation Metric

Macro-F1 (average performance across three categories)

Practical Value

Help identify corporate 'greenwashing' behavior and enhance ESG report credibility

⏰ Subtask 4: Timeline Classification

Objective: Infer the expected completion time of commitments based on statements to establish tracking mechanisms

Output Categories

Already: Commitment already fulfilled (verifiable in current period)
Within 2 years: Short-term target
Between 2 and 5 years: Medium-term target
More than 5 years: Long-term target
N/A: Statement is not a promise, not applicable

Evaluation Metric

Macro-F1 (four-category time inference capability)

Task Relevance and Practical Value

ESG Data Analyst

Corresponding Skills: Subtask 1
Key information extraction, text classification, sustainability report writing

Sustainable Investment Analyst

Corresponding Skills: Subtask 2
Semantic association judgment, logical reasoning, due diligence

Financial Regulatory Examiner

Corresponding Skills: Subtask 3
Greenwashing risk identification, semantic quality assessment, compliance audit

Corporate Sustainability Specialist

Corresponding Skills: Subtask 4
Time information extraction, goal management and tracking, project planning

Dataset Introduction

VeriPromiseESG4K - The world's first Traditional Chinese-designed sustainability commitment verification annotated dataset, sourced from Taiwan 50 Index constituents, spanning 15 industries' authentic ESG reports

📊 Dataset Features

Taiwan's Leading Companies

Real sustainability reports from Taiwan 50 Index (0050) constituent stocks, covering Taiwan's top 50 listed companies.

Cross-Industry Diversity

Spanning 15 different industry sectors including technology, finance, manufacturing, energy, etc., providing rich industry perspectives.

High-Quality Annotation

Executed in collaboration between National Taipei University and University of Taipei teams, with multi-stage quality control and Krippendorff's Alpha ensuring annotation consistency.

📈 Dataset Scale

          Dataset Name:
          VeriPromiseESG4K (World's First Traditional Chinese Sustainability
            Commitment Verification Dataset)

          Total Data Volume:
          4,000 high-quality annotated data points

          Data Source:
          Taiwan 50 Index (0050) constituent stocks, covering top 50 listed
            companies

          Industry Coverage:
          Spanning 15 industry sectors (technology, finance, manufacturing,
            energy, etc.)

          Annotation Dimensions:
          Four subtasks (promise recognition, evidence support, clarity
            assessment, timeline verification)

          Data Split:
          Training set + Test set (Public & Private)
        

🔍 Annotation Process

Phase 1: Initial Annotation

Professional annotation team performs initial marking
Establish annotation standards and guidelines
Conduct annotator training

Phase 2: Cross-Validation

Multiple annotators independently annotate
Calculate inter-annotator consistency
Resolve annotation discrepancies

Phase 3: Expert Review

Domain experts conduct final review
Quality control and corrections
Dataset release

📥 下載資料集

🗂️

VeriPromiseESG4K (Google Drive)

包含訓練集、驗證集及測試集 JSON 檔案

Evaluation Methods

Adopt a multi-task comprehensive scoring mechanism to fully assess model performance across four subtasks

📊 Evaluation Metrics for Each Subtask

Subtask 1: Commitment Classification

Evaluation Metrics

F1-Score: Harmonic mean of precision and recall
Measures the model's ability to identify ESG promise statements

Subtask 2: Evidence Identification

Evaluation Metrics

F1-Score: Determines whether promises have sufficient supporting evidence
Core practical capability assessment

Subtask 3: Clarity Classification

Evaluation Metrics

Macro-F1: Three-category (clear/unclear/misleading) average performance
Most challenging task, identifying greenwashing risk capability

Subtask 4: Timeline Classification

Evaluation Metrics

Macro-F1: Four-category time inference capability
Assesses model understanding of commitment timelines

🏅 Award Structure (Student Category)

🥇 First Place

Slots: 1 team
Prize: NTD 80,000
Certificate: Paper & Digital Certificate by Ministry of Education

🥈 Second Place

Slots: 1 team
Prize: NTD 50,000
Certificate: Paper & Digital Certificate by Ministry of Education

🥉 Third Place

Slots: 1 team
Prize: NTD 30,000
Certificate: Paper & Digital Certificate by Ministry of Education

🎖️ Award of Excellence

Slots: 2 teams
Prize: NTD 10,000
Certificate: Paper & Digital Certificate by Ministry of Education

🏅 Honorable Mention

Slots: 10 teams
Prize: NTD 7,000
Certificate: Paper & Digital Certificate by Ministry of Education

          ※ Note:
          Industry professionals category receives digital certificates only
            (no prize money)
        

          📌 Ranking Rules:

          • Final ranking based on Private Dataset test results

          • Public Dataset for reference during competition only

          • Top 25% teams exceeding Baseline receive Program Office digital
            certificates

Competition Schedule

2026.03.04 ~ 2026.04.28

Registration Opens

Training set available for download.

March 2026

Hands-on Workshops

2026.06.03 ~ 2026.06.10

Validation Set Release

Download available.

2026.06.10 ~ 2026.06.17

Test Set Release & Prediction Submission

2026.06.23

Preliminary Results Announcement

Private Leaderboard results announced.

2026.06.24 ~ 2026.06.30

Additional Deliverables Submission

Submission of model documentation, self-constructed training data, and source code.

2026.07.23

Final Ranking Announcement

2027.03 (To be determined)

Award Ceremony

私有版排行榜

恭喜 AI CUP 2026 VeriPromiseESG 2026 ESG 永續承諾驗證競賽私有版排行榜（Private Leaderboard）前 30 名入圍隊伍！

Private LB 排名	Public LB 排名	隊伍名稱	成績	上傳時間	次數
1	30	TEAM_10671	0.6538186	2026/6/18 23:41	4
2	3	TEAM_10825	0.6528318	2026/6/18 14:30	11
3	13	TEAM_10376	0.647304	2026/6/18 19:49	23
4	36	TEAM_9924	0.6469202	2026/6/17 13:57	20
5	20	TEAM_10339	0.6465768	2026/6/18 20:42	22
6	59	TEAM_10858	0.6457201	2026/6/18 21:58	18
7	47	TEAM_10287	0.6457018	2026/6/18 17:23	15
8	7	TEAM_9944	0.6449874	2026/6/18 15:41	24
9	11	TEAM_10478	0.6435742	2026/6/18 0:26	23
10	9	TEAM_10832	0.6435522	2026/6/17 21:13	20
11	42	TEAM_10289	0.6434421	2026/6/14 20:05	11
12	15	TEAM_10505	0.6432854	2026/6/17 22:05	21
13	28	TEAM_10637	0.6421031	2026/6/18 10:51	16
14	5	TEAM_10365	0.6416516	2026/6/18 23:59	21
15	1	TEAM_10049	0.6404881	2026/6/18 20:03	24
16	16	TEAM_10070	0.6401839	2026/6/18 0:22	18
17	21	TEAM_10029	0.6401161	2026/6/18 23:59	14
18	46	TEAM_10209	0.6397067	2026/6/18 23:43	14
19	44	TEAM_10219	0.6386597	2026/6/18 8:59	23
20	48	TEAM_10036	0.6385083	2026/6/18 19:20	17
21	12	TEAM_9939	0.6375782	2026/6/17 16:57	21
22	33	TEAM_9908	0.637124	2026/6/18 20:08	23
23	75	TEAM_10537	0.6366663	2026/6/18 22:53	5
24	17	TEAM_9910	0.6359202	2026/6/18 0:10	14
25	27	TEAM_10001	0.6356957	2026/6/18 23:24	12
26	23	TEAM_10057	0.6356308	2026/6/18 0:21	24
27	40	TEAM_10801	0.6350402	2026/6/17 23:44	17
28	24	TEAM_10023	0.6348259	2026/6/18 21:58	19
29	51	TEAM_9907	0.634656	2026/6/18 7:28	19
30	37	TEAM_10440	0.6342595	2026/6/18 21:03	16

Organizing Team

Co-hosted by top academic institutions from Taiwan and Japan with industry experts

🎓 Principal Investigator

Prof. Min-Yuh Day

Principal Investigator

National Taipei University
Graduate Institute of Information Management

Website

Specialized in artificial intelligence, generative AI, and sustainable green fintech. Currently Director of Fintech and Green Finance Research Center.

Co-Principal Investigators

Prof. Chung-Chi Chen

Co-Principal Investigator

National Institute of Informatics (NII), Japan

Human-Agent Ally Lab

Founder of ACL SIG-FinTech, specialized in financial opinion mining and natural language processing.

Prof. Yohei Seki

Co-Principal Investigator

University of Tsukuba
Faculty of Library, Information and Media Science

Website

Specialized in natural language processing and information retrieval, organized NTCIR multilingual opinion analysis tasks.

Research Assistants

Hsin-Ting LU

Researcher

Website

Wen-Ze Chen

Researcher

Website

Wei-Chun Huang

Researcher

Website

Yu-Han Huang

Researcher

Website

Jun-Yu Wu

Researcher

Website

Partner Institutions

National Taipei University

Graduate Institute of Information Management
Fintech and Green Finance Research Center

University of Taipei

Department of Computer Science

National Institute of Advanced Industrial Science and Technology (AIST)

Artificial Intelligence Research Center

University of Tsukuba

Faculty of Library, Information and Media Science

標註範例說明

以下提供 E、S、G 三類文本的標註範例，協助參賽者理解標註規則與判斷標準

ESG 類別 E

類型有承諾 - 有證據

段落內容

承諾遵循國際人權公約與基本勞動人權原則，落實性別平等、結社自由，並致力於消弭歧視與強迫勞動。

承諾狀態是 (Yes)

證據狀態是 (Yes)

證據品質清晰 (Clear)

驗證時間軸已執行 (already)

註解有制度性承諾與明確人權框架。

ESG 類別 E

類型有承諾 - 有證據

段落內容

除非資金明確用於綠能轉型計畫，不再新增投資燃煤比重超過 50% 的電廠；同時針對燃料煤相關產業制定嚴格准入與撤資標準，積極引導資金流向低碳與可再生能源領域。

承諾狀態是 (Yes)

證據狀態是 (Yes)

證據品質不清晰 (Not Clear)

驗證時間軸已執行 (already)

註解行動方向明確，但量化成效與第三方驗證描述不足。

ESG 類別 E

類型有承諾 - 無證據

段落內容

落實環境永續、實踐永續承諾是本公司的企業使命，將以 2050 年淨零排放為首要目標，致力實現多項環境永續承諾。

承諾狀態是 (Yes)

證據狀態否 (No)

證據品質不適用 (N/A)

驗證時間軸 5 年以上 (more_than_5_years)

註解屬於願景型承諾，未提供具體行動或成果。

ESG 類別 E

類型無承諾

段落內容

應用人工智慧強化良率監控，輔助辦識模型有效性達 98%。

承諾狀態否 (No)

證據狀態不適用 (N/A)

證據品質不適用 (N/A)

驗證時間軸不適用 (N/A)

註解僅敘述產品效果，未有行動承諾。

ESG 類別 S

類型有承諾 - 有證據

段落內容

我們將持續推動零職災文化，完善承攬商納管與教育訓練機制，並加強現場稽核頻率，以降低高風險作業事故率。為此，我們正調整標準作業流程並提高稽核覆蓋率。

承諾狀態是 (Yes)

證據狀態是 (Yes)

證據品質不清晰 (Not Clear)

驗證時間軸 2-5年 (between_2_and_5_years)

註解明確表達企業將持續推動零職災文化，屬於清楚的行動承諾；相關行動未提供量化指標、明確頻率或制度細節，證據支撐力道有限。

ESG 類別 S

類型有承諾 - 無證據

段落內容

制定企業內部衝突礦產管理聲明，承諾遵守 RBA 不使用衝突礦產政策。

承諾狀態是 (Yes)

證據狀態否 (No)

證據品質不適用 (N/A)

驗證時間軸已執行 (already)

註解有制度性承諾與明確人權框架。

ESG 類別 S

類型無承諾

段落內容

推出金來寶小額終身壽險，提供高齡者基本保險保障與終身壽險服務。

承諾狀態否 (No)

證據狀態不適用 (N/A)

證據品質不適用 (N/A)

驗證時間軸不適用 (N/A)

註解僅是產品描述，未包含未來承諾語句。

ESG 類別 G

類型有承諾 - 有證據

段落內容

我們將設立由三位獨立董事組成的永續委員會，每季檢視重大風險議題與對應計畫，並在年度股東會前完成報告揭露與外部查驗。

承諾狀態是 (Yes)

證據狀態是 (Yes)

證據品質清晰 (Clear)

驗證時間軸 2-5年 (between_2_and_5_years)

註解明確表達企業將設立永續委員會，並規劃其運作方式，屬於具體且可執行的治理承諾；清楚說明組織架構、執行頻率及查驗機制。

ESG 類別 G

類型有承諾 - 有證據

段落內容

未來改選董事時將提高女性董事席次，以增進董事會性別多元化。

承諾狀態是 (Yes)

證據狀態是 (Yes)

證據品質不清晰 (Not Clear)

驗證時間軸 2年內 (within_2_years)

註解有明確方向，但缺乏具體目標與追蹤機制。

ESG 類別 G

類型無承諾

段落內容

2024 年研發人員數達 446 人，完成研發專案 17 件，智慧財產權累計 202 件。

承諾狀態否 (No)

證據狀態不適用 (N/A)

證據品質不適用 (N/A)

驗證時間軸不適用 (N/A)

註解營運成果描述，非治理承諾。

Related Resources

Important links to related competitions and research papers

PromiseEval-2025 @ SemEval

SemEval 2025 shared task on promise verification

Visit Website →

Multi-Lingual ESG Issue Identification

FNP@IJCAI-2023 paper on ESG issue identification

Read Paper →

Multi-Lingual ESG Impact Type Identification

FNP@IJCAI-2023 paper on ESG impact type classification

Read Paper →

Multi-Lingual ESG Impact Duration Inference

FNP@EMNLP-2024 paper on ESG impact duration prediction

Read Paper →

NTCIR-19 RegCom

NTCIR-19 Regulatory Compliance task

Visit Website →

ML-Promise: A Multilingual Dataset for Corporate Promise Verification

EMNLP 2025 paper on multilingual promise verification dataset

Read Paper →

論文發表

本競賽相關研究論文

IEEE EM-RITE 2026 @ IEEE IRI 2026

VeriPromiseESG4K: A Traditional Chinese ESG Promise Verification Dataset for AI CUP 2026

Min-Yuh Day, Wei-Chen Huang, Hsin-Ting Lu, Wen-Ze Chen, Yu-Han Huang, Ching-Tai Chen, and Yohei Seki

The 15th IEEE International Workshop on Empirical Methods for Recognizing Inference in TExt (IEEE EM-RITE 2026), in Proceedings of the 27th IEEE International Conference on Information Reuse and Integration for Data Science (IEEE IRI 2026), Seattle, WA, USA, July 31 – August 2, 2026.

📄 下載 PDF

引用格式 (BibTeX)

@inproceedings{day2026veripromiseesg4k,
  title     = {VeriPromiseESG4K: A Traditional Chinese ESG Promise Verification Dataset for AI CUP 2026},
  author    = {Day, Min-Yuh and Huang, Wei-Chen and Lu, Hsin-Ting and Chen, Wen-Ze and Huang, Yu-Han and Chen, Ching-Tai and Seki, Yohei},
  booktitle = {Proceedings of the 27th IEEE International Conference on Information Reuse and Integration for Data Science (IEEE IRI 2026)},
  year      = {2026},
  address   = {Seattle, WA, USA},
  month     = {July}
}

ESG Promise Verification Competition

Competition Overview

Practical Application

Technical Challenge

International Collaboration

Competition Tasks

Subtask 1: Commitment Classification

Output Categories

Evaluation Metric

Examples

🔗 Subtask 2: Evidence Identification

Output Categories

Evaluation Metric

Example

💡 Subtask 3: Clarity Classification

Output Categories

Evaluation Metric

Practical Value

⏰ Subtask 4: Timeline Classification

Output Categories

Evaluation Metric

Task Relevance and Practical Value

ESG Data Analyst

Sustainable Investment Analyst

Financial Regulatory Examiner

Corporate Sustainability Specialist

Dataset Introduction

📊 Dataset Features

Taiwan's Leading Companies

Cross-Industry Diversity

High-Quality Annotation

📈 Dataset Scale

🔍 Annotation Process

Phase 1: Initial Annotation

Phase 2: Cross-Validation

Phase 3: Expert Review

📥 下載資料集

Evaluation Methods

📊 Evaluation Metrics for Each Subtask

Subtask 1: Commitment Classification

Evaluation Metrics

Subtask 2: Evidence Identification

Evaluation Metrics

Subtask 3: Clarity Classification

Evaluation Metrics

Subtask 4: Timeline Classification

Evaluation Metrics

🏅 Award Structure (Student Category)

🥇 First Place

🥈 Second Place

🥉 Third Place

🎖️ Award of Excellence

🏅 Honorable Mention

Competition Schedule

私有版排行榜

得獎名單

🏆 教育部獎狀

📋 計畫辦公室獎狀

Organizing Team

🎓 Principal Investigator

Prof. Min-Yuh Day

Co-Principal Investigators

Prof. Chung-Chi Chen

Prof. Yohei Seki

Research Assistants

Hsin-Ting LU

Wen-Ze Chen

Wei-Chun Huang

Yu-Han Huang

Jun-Yu Wu

Partner Institutions

National Taipei University

University of Taipei

National Institute of Advanced Industrial Science and Technology (AIST)

University of Tsukuba

協辦單位

標注團隊致謝

校園巡迴課程

北區

中區