CONSTRUCTION OF CHINESE FINANCIAL EVENT RELATION CORPUS

Event relation recognition is not only an important task of information extraction but also the key for constructing event knowledge graphs. However, most event corpora in Chinese financial field are only built for event extraction tasks, therefore, it is significant to build a corpus that focuses on event relation recognition. On the basis of analyzing the characteristics of financial news texts, a set of event and relation labeling specifications was formulated, and we carried out the work of labeling corpus with the annotation tool constructed. A Chinese financial event relation corpus was constructed, which contained a total of 367 annotated documents, 11 208 events, and 196 569 event relation pairs. The analysis of annotation consistency and the experiment results of event relation recognition based on the deep learning models show that the corpus has good usability.