Auto-labeling Terms Based on Multi-scanning Strategy

摘要

In order to construct the term corpus of physics teaching materials for elementary education, the characters of physics terms were studied, the prediction templates for the unknown terms was built, all kinds of rules for identifying terms was extracted, and the labeling errors of maximum matching algorithm was analyzed, at last, an auto-labeling system was developed. Firstly, this algorithm scans and labels terms which match the rule templates. Secondly, it takes terms in the base glossary as anchor points, and finds out every anchor point with the maximum matching algorithm. Finally scans the context of the anchor point so as to judge whether the candidate strings is a term or not. Together with the prediction and limited function of rules, this method makes full use of the information of terms in base glossary and achieves a higher precision and recall rate. The F-index reaches about 84% in open test.