An EM Algorithm for Context-Based Searching and Disambiguation with Application to Synonym Term Alignment
Jing-Shin Chang, Shih-Jay Chiou · Institutional Repositories DataBase (IRDB) · 2009
Abstract. A statistical context-based searching model and an unsupervised EM algorithm are proposed to resolve the large class of searching problems that require left and right contexts for disambiguation, in which the contexts can be synonyms. The searching problem is modeled as a machine translation problem in which pieces of contexts are accumulated to enforce the translation probability between a search result and the source query. This model is applied to the term alignment problem between traditional and simplified Chinese synonymous terms. In comparison with previous works on the same task, the EM algorithm for context-based searching and disambiguation significantly improves the term alignment accuracy by 2~48%, for technical, transliteration and common terms. The alignment accuracy ranges from 47~85 % in different domains.