Auto-POS templates and mixed metrics for recognizing terms in scientific literature

Automatic Term Recognition (ATR) is an important task for Knowledge Acquisition, which aims at acquiring formalized words which are not recorded in time in the glossary. In recent years, several statistical methods has proved to be effective, and emerging methods such as C-value, NC-Value, TermExtractor has shown great advantages on this task. However, few works have been done on the Metric mixing algorithm that combines those metrics as a whole. In this paper, we first collect part-of-speech templates from already-known terms automatically, namely Auto-POS templates, instead of artificial regular expressions, and then we match them with POS strings to acquire candidate terms. Finally we sort those candidates by metric mixing algorithm. Experimental results on IEEE2006-2007 metadata show that the metric mixing algorithm performs better than any separate metrics alone.