Unsupervised Chinese Personal Name Recognition Using Search Session

Personal name recognition is an important part of named entity recognition in Web search query logs. An unsupervised method for Chinese personal name recognition in queries is proposed using search session. Based on seed personal names which are produced automatically by introducing Chinese surnames, a local expansion method is proposed by using search sessions in query logs;and by modeling the process of people establishing connections with others the authors propose a candidate filtering method using candidate contexts. P@500 of personal name recognition had reached 99% on Sogou Web search query logs, and the precision of the candidate filtering method was 81.82%. The results validate the effectiveness of the proposed personal name recognition method, and prove that the filtering method has positive impact on improving the accuracy of the recognition.