Event Reason Extraction Dataset
-
Finreason Reason Extraction of Structural Events from Financial Documents
- Pledge: Pledging of shares is that the shareholders of listed companies use their stocks (equity) as collateral to apply for loans from banks or provide a guarantee for loans from third parties. This kind of event implies the funding status of the related companies and investors will take it into accounts when making decisions.
- O&U: Overweighting\Underweighting of shares refers to the major shareholder's increasing or reducing of their shares over the company. This kind of event indicates the confidence of the major shareholders towards the company and will have an influence on the stock price.
- Lawsuit: Lawsuit and Arbitration are the legal disputes about the listed company or the big shareholders of the company. This type of event will tremendously impact the stock price of the related companies.
- Pei Chen, Kang Liu, Yubo Chen, Taifeng Wang and Jun Zhao, Probing into the Root: A Dataset for Reason Extraction of Structural Events from Financial Documents, in Proceedings of EACL 2021.
FinReason is a financial-domain Chinese corpus regarding extracting the causes of major events in the announcements of listed companies. Each document in this corpus contains one or more structural events, and each event has none, one or more causes in the document. These events are automatically matching to the documents and the causes are manually annotated. In total, there are 3 types of events in FinReason.
The statistics of the dataset is shown in the following table.
| Event Type | Doc Count | Event Count | Reason Count | Doc Count with Reason |
|---|---|---|---|---|
| Pledge | 4,138 | 5,379 | 4,714 | 2,901 (70.11%) |
| O/U | 2,550 | 4,127 | 3,565 | 2,132 (83.61%) |
| Lawsuit | 2,106 | 3,355 | 2,727 | 1,438 (68.28%) |
| Total | 8,794 | 12,861 | 11,006 | 6,471 (73.58%) |
Paper:
Document Level Event Extraction Dataset
-
CFEED Chinese Financial Event Extraction Dataset
- FREEZE: Equity Freezing refers to a compulsory measure adopted by the court to restrict the owner of equity to withdraw or transfer his own equity. The main purpose of such measures is to prevent the improper loss of equity returns.
- PLEDGE: Equity Pledging is that the shareholders of listed companies use their stocks (equity) as collateral to apply for loans from banks or provide a guarantee for loans from third parties.
- OW&UW: Equity overweighting\underweighting refers to the shareholder's increasing or reducing of their shares over the company.
- Hang Yang, Yubo Chen, Kang Liu, Yang Xiao and Jun Zhao, DCFEE: A Document-level Chinese Financial Event Extraction System based on Automatically Labeled Training Data, in Proceedings of ACL 2018, Melbourne, Australia, July 15-20. PDF
- Pei Chen, Hang Yang, Kang Liu, Ruihong Huang, Yubo Chen, Taifeng Wang and Jun Zhao, Reconstructing Event Regions for Event Extraction via Graph Attention Networks, in Proceedings of AACL 2020, Suzhou, China, December 4-7. PDF
Chinese Financial Event Extraction Dataset (CFEED) is a financial-domain Chinese corpus regarding the major events in the announcements of listed companies. Each document in this corpus contains one or more event templates. This dataset is automatically generated by distant supervision method. We crawled the public announcements from sohu.com and the event templates from eastmoney.com. Since the announcements are not in line with their corresponding event templates, we utilize distance supervision to align them. We assume that if the key role fillers in a template appear in an announcement, the announcement is describing the event in the template. There are 3 types of events in CFEED:
For each event type, there are different event roles in it as shown in the following table.
| Event Type | Freeze | Pledge | OW&UW |
|---|---|---|---|
| Name | Shareholder Name | Shareholder Name | Shareholder Name |
| ORG | Freeze Organization | Pledge Institution | - |
| NUM | Number of Frozen Stock | Number of Pledged Stock | Number of Stock Increase or Decrease |
| BEG | Freezing Start Date | Pledging Start Date | Start Date |
| END | Freezing End Date | Pledging End Date | End Date |
The statistics of the dataset is shown in the following table.
| Freeze | Pledge | OW&UW | Total | |
|---|---|---|---|---|
| Training/Development/Testing | 589/150/300 | 3,602/300/300 | 1,303/300/300 | 5,494/750/900 |
Paper: