Creating A Large-Scale Financial News Corpus for Relation Extraction

As the data accessible is growing exponentially, the unstructured information has become the most common and main resource for decision makers. News, one of the most basic resource containing rich information covers from the company reports to market dynamics, has been regarded as one of the most important information resources. However, the richness of information in the huge volume of news also leads to the difficulty for the managers to get the key information benefit for their decisions, for which the processing on unstructured information is crucial and fundamental. This paper describes the establishment of a Chinese corpus for researches on financial entity recognition and financial relation extraction based on a large financial news data set. We propose a mixed pattern with POS tagging (Part of Speech Tagging) and BIES (Begin, Inside, End, Single) annotation for Generating quadruples (entity1, entity2, relation, text) from unstructured text. To the best of our knowledge this is the first attempt at extracting financial relation concepts. The average accuracy that was achieved by mixed pattern was (88.88%) by manually verification on random sampling dataset. Evaluation of the corpus gives good results and can be used for finance knowledge extraction.