| Entity Linking aims to link the named entities in the free text to the corresponding entities in the given knowledge base (KB). Entity Linking can enrich the information of the entities in the text, so that has remarkable meanings to the understanding of the text by the users and computers. It is widely used in the area of entity extraction, information retrieval, machine learning and so on, and has become one of the basic technologies of natural language understanding and tasks of semantic computing.In classical entity lining tasks, the typical KB often contains plenty additional information for entities. For example, Wikipedia, the infoboxes, text descriptions and anchor text can all do great help to the link generation and disambiguation steps. In our paper, we resolve a problem that linking organization names to a list-like KB. More specifically, the records in the KB are simply organization name full names without any context. For this kind of task, the massive variations or abbreviations in the text cannot be linked to the list directly, and bring about a lot of ambiguities.The proposed method to deal with the problem contains an offline and an online step. In the offline step, making use of various sources, such as Hudong Baike, we design a pattern based method to annotate the organization names. On that basis, possible abbreviations are generated to extend the KB. In the online step, we propose a two-stage link generation method to avoid the ambiguities, utilizing the co-occurrence of abbreviations and full names in the same document or document cluster, where the linked full names in the first stage constraint the linking of abbreviations in the second stage.We apply our approach to police inquiry records as well as industry economy news on Xinhua Net, with the organization name list provided by police as the KB. The results show our annotation strategy effective, and the two-stage entity linking method performs well. |
PDF Full Text Request