Experiments on company name disambiguation with supervised classification techniques

Entity disambiguation is the task of identifying the real world entity was referred to in a context. Ambiguous references to entities can occur due to variations of how entity was referenced (BT, British Telecom) or inherit ambiguities of the names used for entities (Orange Telecom vs. fruit orange) and misspellings (Best Buy vs. BestBuy). Ambiguities in company names however come with a price, when it comes to finding information about the company on the Web. Recently, tracking social media for brand management has become a very important part of the process in marketing, public relations, and product marketing. Therefore, resolving references to the real world objects has become an important part of the social media analytics systems. In this paper, we study different machine learning techniques for entity disambiguation in micro-blogging posts. Our experiments show that using supervised algorithms with carefully selected features, one can improve the disambiguation quality significantly.