Temporal Knowledge Graph Construction from Text

本研究旨在从文本中构建一个时间知识图。在本财政年度，我们利用SEC和市场交易所报告的信息构建了金融领域的时间知识图FinKG。使用我们的财务本体作为模板，提取包括报告细节和股票价格的时态信息。在知识图的边缘处对时间信息进行编码。FinKG包含了超过3000万条事实。此外，我们证明了FinKG的有用性与两个应用程序：知识检索和股票价格预测。知识检索揭示了实体之间的复杂联系，而FinKG的聚合特征有助于神经模型更好地预测股票价格。

This research aims to build a temporal knowledge graph from the text. In this fiscal year, we constructed FinKG, the temporal knowledge graph in the financial domain using information reported from SEC and market exchange. The temporal information including the report detail and stock price, is extracted using our financial ontology as a template. The temporal information is encoded at the edge of the knowledge graph. Overall, FinKG contained more than 30 million facts. Furthermore, we demonstrate the usefulness of FinKG with two applications: knowledge retrieval and stock price prediction. Knowledge retrieval reveals the complex connection among entities, while aggregated features from FinKG help neural models to better forecast stock prices.