Semi-supervised Statistical Inference for Business Entities Extraction and Business Relations Discovery

The sheer volume of user-contributed data on the Internet has motivated organizations to explore the collective business intelligence (BI) for improving business decisions making. One common problem for BI extraction is to accurately identify the entities being referred to in user-contributed comments. Although named entity recognition (NER) tools are available to identify basic entities in texts, there are still challenging research problems such as co-reference resolution and the identification of abbreviations of organization names. The main contribution of this paper is the illustration of a novel semi-supervised method for the identification of business entities (e.g., companies), and hence to automatically construct business networks. Based on the automatically discovered business networks, financial analysts can then predict the business prestige of companies for better financial investment decision making. Initial experiments show that the proposed NER method for business entity identification is more effective than other baseline methods. Moreover, the proposed semi-supervised business relationship extraction method is more effective than the state-ofthe-art supervised machine learning classifiers when there are not many training examples available. Our research work contributes to advance the computational methods for the extraction of entities and their relationships from texts.