Research on open domain Named entity recognition based on Chinese query logs

Search engine query logs contain quantities of Named Entities. As the basic work of information extraction, traditional Named-entity extraction methods only can extract specific categories of entities. It is very difficult for them to be applied to the query log Named-entity recognition directly for their limitation. In this paper, a novel approach is proposed to extract Named Entities from user query logs. In order to avoid the dependence on large-scale tagging corpus, we annotate the data automatically by using distant supervision method. Thus the problem that the training data needs human-annotation effort is solved. Moreover, open domain Named Entities are extracted from user query logs based on the conditional random field model. Evaluation on user query logs shows the effectiveness of our approach in extracting Named Entities in open domain.