Semi-supervised approach for Persian word sense disambiguation

Abstract

Word-sense disambiguation is one of the key concepts in natural language processing. The main goal of a language is to present a specific concept to the audience. This concept is extracted from the meaning of words in that language. System should be able to identify role and meaning of words in order to identify the concepts in texts properly. This issue becomes more problematic if there are words that take different meanings because of their surrounding words. Regarding that different practical programs have been developed in Persian language, it is vital now to find a solution for word-sense disambiguation in Persian language. Lack of training data is the biggest challenge in the course of word-sense disambiguation in Persian language. In order to face this problem, machine learning approach with minimal supervision is employed in this research. The applied method tries to disambiguate word senses by considering defined features of target words and applying collaborative learning method. Extracted corpus from published news by news agencies is used as the reference corpus. Evaluating the program by the available corpus on three considered ambiguous words, the implemented method has been able to properly identify the meaning of 5368 documents with 88% recall, 95% precision and 93% accuracy rate.