Leaning to Train: Linking Financial News Articles to Company Short Names

摘要

As a special type of named entity, company name is frequently mentioned in financial news articles, leading to significant necessity on company-oriented information retrieval and management. However, company names are usually mentioned with short names, which are sometimes ambiguous. For example, apple refers in some cases to Apple Incorporation while in other cases to a kind of sweet fruit. This motivates our research on linking financial news articles to company short name, which aims to determine whether a mention in an article is short name of a company. The supervised approach requires labor on annotation of news article that mention the specific company short name. It is rather unpractical as new company short names appear constantly. In this work, we propose a self-contained unsupervised learning framework, which relies on probabilistic topic model to collect training data automatically. Experimental results show that the performance is close to the state-of-the-art supervised approach which relies on human-judged gold standard.