Semantic Classes Discovery Based on Soft Pattern

TAN Hong-ye,ZHAO Tie-jun
DOI: https://doi.org/10.3321/j.issn:0367-6234.2007.11.024
2007-01-01
Abstract:This paper presents a method of using soft pattern to discover semantic classes. In the method, patterns are obtained in a bootstrapping cycle, and then the soft pattern is generated based on the hard patterns, finally the named entities of the interested semantic classes are discovered with the fuzzy match of the soft pattern. Experiments on Chinese corpus of People’s Daily show that the lowest recall can achieve 60.1%, which suggests that soft pattern performs well in ensuring high recall cause of its integration of rich information, flexible format and its supporting to the fuzzy match.
What problem does this paper attempt to address?