Linked Open Data Based Entity Linking

Entitylinkinghasbeenahottopicindataminingfeldnowadays. Thistaskaimstolink the mentioned entities from given piece of natural text, to a given knowledge base,in which entities means concepts or objects, such as persons, locations. It difers fromthe traditional NER (Named entity recogition), in that, the latter only recognize theexistence and the location of specifc types of entities, without linking to a knowledgebase, hence it cannot provide detailed information about the entities. An excellent en-tity linking system or algorithm will annotate the text automatically, and provide usefulhelp for tasks such as text structuring. In this paper, we propose a domain topic modelbased entity linking algorithm, targeting the Chinese knowledge base Zhishi.me. Weextract certain domains and use the domain specifc topic model to capture the seman-tic fetures of that domain. We also developed a domain expansion algorithm based onthe knowledge base’s structural information, and a synonym fnding algorithm basedon the co-occurence relationships between words, to overcome the problems of incom-plete domain and low recall respectively. We evaluate our algorithm using manuallyannotated text data, both from news and microblogging. The result is promising toshow the efectiveness of our algorithms. We also developed a demo system in whichthe user can submit text to the web service and get back the linked entities.