Unsupervised Word Sense Discrimination Improves Construction of the Wordnets

This paper reports an unsupervised approach we adopted for Wordnet construction. We combine ways of CBC word sense discrimination and automatic mapping measures to remove redundant senses of adjectives in Chinese Concept Dictionary. We optimize the similarity measure of adjective concepts and put forward two novel strategies to automatically evaluate the model’s performance. The experimental results show a rather fine conformance to the work based on human judgments.