Chinese Personal Name Recognition in Web Queries via Bootstrapping

The bootstrapping method of recognizing names in query logs is useful in some ways, but it is affected by noisy data, which comes from irrelevant queries and irrelevant template matching. We propose a trend selection method which aims to select templates more relevant to personal names rather than other categories. In order to eliminate noisy data when matching, the boundaries of candidates are relocated by the presented method named forward-backward keyword matching based on the corpus from People´s Daily. Experimental results on Sogou corpus indicate that the trend selection method is better while compared to other template selection method. And the forward-backward keyword matching method is helpful for boundary demarcation and recognition.