III: EAGER: Knowledge Graph Mining for Financial Risk Analytics

通过对金融文档的文本分析进行图形数据挖掘和学习可以成为增强金融风险衡量和管理以及实现监管监督和合规的强大工具。这项提议寻求为大量的金融新闻和事件流以及披露文件(文件、电话报告等)开发新的知识图谱挖掘方法。来自金融机构对银行和金融系统的系统性风险评估。这项提案中提出的新方法将有助于银行和保险公司等金融机构以及此类金融机构的监管机构。基于文本分析的风险指标可用于前者的一系列资产和风险管理决策，以及用于确定对后者进行监管的具体风险类型。由于文本数据无处不在，从文本中提取有价值的见解仍然是一个挑战，本项目中开发的风险分析技术也将有助于分析文本数据可以提供有价值见解的其他领域。在这个研究项目中，PI将开发创新的数据挖掘和学习方法，根据从提交给SEC的公开可用年度和季度报告中挖掘的文本和语义特征，创建“金融风险”知识图谱。PI还将使用新闻文章和信用评估报告中的文本数据。关键的底层方法依赖于开发创新的最先进的文本和图形挖掘方法，以获得有效的特定领域的方法来处理金融文本。特别是，PI计划定义、提取和跟踪与金融机构不同风险敞口对应的细微差别和情绪主题，如信贷、利率、流动性、汇率风险，以及运营、监管和声誉风险的影响，以改进风险预测和监测。这项任务的结果将是基于文本和图形挖掘的丰富的风险知识图。知识图谱将包括风险细微差别的情绪、用于风险分析的词语和短语，以及一套金融风险概念和不同实体之间的关系。

Graph Data mining and learning via text analytics over financial documents can be a powerful tool to enhance financial risk measurement and management, as well as to enable regulatory oversight and compliance. This proposal seeks to develop novel knowledge graph mining methods for the vast set of financial news and events streams, and disclosure documents (filings, call reports, etc.) from financial institutions for systemic risk assessment of the banking and financial system. Novel methods developed in this proposal will help both the financial institutions, such as banks and insurance firms, as well as regulators of such financial institutions. Risk indicators based on text analytics can be used for a range of asset and risk management decisions for the former, and for identifying specific risk type for regulatory oversight for the latter. Since textual data is ubiquitous, and extracting valuable insights from text remains a challenge, the risk analysis techniques developed in this project, will also help analyze other domains where textual data can offer valuable insights. Examples include international crises, natural disasters, humanitarian efforts and so on, where an assessment of the risk is essential for appropriate responses.In this research project the PI will develop innovative data mining and learning methods to create a "financial risk" knowledge graph from textual and semantic features mined from the publicly available annual and quarterly reports filed with the SEC. The PI will also use textual data from news articles and credit assessment reports. The key underlying methods rely on developing innovative state-of-the-art text and graph mining methods for effective domain-specific approaches to deal with financial text. Especially, the PI plans to define, extract and track nuance and sentiment topics corresponding to different risk exposures of financial institutions, such as credit, interest rate, liquidity, exchange rate risks, as well as the impact of operational, regulatory and reputation risks, to improve risk prediction and monitoring. The outcome of this task will be a rich knowledge graph of risk based on text and graph mining. The knowledge graph will comprise risk-nuanced sentiment words and phrases for the use in risk analytics and a set of financial risk concepts and relationships among different entities.