Financial Customized Information Extraction Platform
Helping financial industry R&D personnel to customize financial information extraction API online. It can quickly cold start without any NLP expertise and a large amount of labeled data, conveniently and quickly build benchmark effects for business scenario verification, and calmly respond to flexible and changeable business scenario requirements.
Product Overview
Common Extraction Tasks in the Financial Industry
Supporting information extraction in different styles, such as news, announcements, and research reports. Common fields such as name, company name, and time have been specially optimized.
Zero Threshold Customized Information Extraction Model
Customizing your multi-field information extraction model online by simply adjusting prompts without NLP expertise and large amounts of annotated data. Responding quickly to the complex and changing needs of business scenarios.
Publishing API and Connecting to Existing Systems Easily
You can preview the extraction effect online, and publish it as an API with one click for existing business platforms to call. Seamlessly connect low-code, RPA, chatbots and other scenarios.
Application Scenarios and Pain Points
There are many information extraction scenarios in the financial industry, and business requirements change rapidly. Using the pure rule method for information extraction is poor in coping with changes. Most scenarios lack labeled data, making model training difficult.
Pure Rule Extraction is Difficult to Cope with Variable Expressions
The description methods of language and characters are varied, and semantics and context cannot be understood only by rules, and it is difficult to obtain better information extraction results.
Long Implementation Cycle of Algorithm Model Customization
The standardization of the algorithm model is low, and good effects often require in-depth customization. The huge investment and lengthy process from project establishment, bidding to business launch made it difficult to quickly respond to business needs.
Difficult to Cold Start Model Training
Deep customization algorithm model not only requires a lot of data annotation, but also involves a high degree of business cooperation and feedback. It is difficult to determine whether a project is worth investing resources in the early stages of implementation.
Product Advantages
Covering Common Information Extraction Scenarios, out of the Box
In addition to common fields such as name, company name, time, etc., targeted optimization is also carried out for common document types in the financial field, such as announcements and news events such as fines, lawsuits, and acquisitions.
Based on Zero-sample Learning Technology, Labeling Requirements are as Low as Zero
Based on Mengzi's multi-task learning technology, it can achieve a certain extraction accuracy even when there is no labeled data at all. In August 2022, it ranked first on ZeroCLUE, the authoritative list of Chinese zero-sample learning, and FewCLUE, the authoritative list of small-sample learning.
Cold Start Before Upgrading, Low Project Investment Risk
You can first establish a benchmark effect through a quick cold start of the customized information extraction model API, and clearly define the input and output. Then increase investment in the link where the effect is not as expected and carry out targeted optimization to effectively reduce project risks.
ZeroCLUE - Chinese Zero Sample Learning Authoritative List
August 22, 2022 Rankings
| Ranking | Model | Research Institutions | Score | EPRSTMT | CSLDCP | TNEWSE | IFLYTEKE | OCNLIE | BUSTM | CHIDE | CSLE | CLUEWSCE |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 | Mengzi-T5-MT | 澜舟科技 | 68.926 | 86.99 | 55.19 | 74.73 | 22.42 | 74.69 | 77.6 | 85.1 | 84.17 | 65.17 |
| 2 | GPT-MoF | 阿里云机器学习平台PAI | 67.458 | 84.2 | 60.29 | 57.73 | 51.31 | 67.59 | 63.9 | 88.2 | 56.53 | 67.59 |
| 3 | 二郎神-MRC | IDEA研究院 | 63.515 | 86.19 | 48.65 | 69.47 | 36.08 | 45.59 | 74.05 | 84.65 | 53.53 | 79.31 |
FewCLUE - Chinese Small Sample Learning Authoritative List
August 22, 2022 Rankings
| Ranking | Model | Research Institutions | Score | EPRSTMT | CSLDCP | TNEWSE | IFLYTEKE | OCNLIE | BUSTM | CHIDE | CSLE | CLUEWSCE |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 1 | Mengzi-T5-MT | 澜舟科技 | 76.612 | 88.18 | 63.15 | 78.73 | 51.54 | 74.75 | 78.85 | 85.25 | 84.2 | 86.9 |
| 2 | FSL++ | Meituan NLP | 76.458 | 88.45 | 68.42 | 75.53 | 54.27 | 77.76 | 80.3 | 83.35 | 79.73 | 79.31 |
| 3 | 玉言 | 网昜伏羲 | 76.306 | 88.45 | 70.92 | 71.87 | 53.12 | 70.23 | 78.75 | 78.15 | 82.83 | 87.93 |
Related Recommendations
Financial Market Search and Judgment Platform
An integrated platform solution for information search, analysis and judgment, and decision support for market research and investment decisions
Learn More
Financial NLP Platform
Providing scenario NLP capabilities with strong versatility in the financial industry, which can be called through APIs
Learn More
Business Cooperation Email
Address
11F, Block A, Dinghao DH3 Building, No.3 Haidian Street, Haidian District, Beijing, China
Business Cooperation:
bd@langboat.com
Address:
11F, Block A, Dinghao DH3 Building, No.3 Haidian Street, Haidian District, Beijing, China
Official Accounts:
