Learning to Mine Chinese Coordinate Terms Using the Web

TL;DR

This paper presents a semi-supervised approach for extracting Chinese coordinate terms from web data, effectively handling polysemy and grouping terms into concepts, with high-quality, wide-coverage results.

Contribution

It introduces a novel semi-supervised method combining linguistic patterns and learned patterns for Chinese coordinate term extraction from web data.

Findings

High-quality extraction results

Wide coverage of coordinate terms

Effective handling of polysemy

Abstract

Coordinate relation refers to the relation between instances of a concept and the relation between the directly hyponyms of a concept. In this paper, we focus on the task of extracting terms which are coordinate with a user given seed term in Chinese, and grouping the terms which belong to different concepts if the seed term has several meanings. We propose a semi-supervised method that integrates manually defined linguistic patterns and automatically learned semi-structural patterns to extract coordinate terms in Chinese from web search results. In addition, terms are grouped into different concepts based on their co-occurring terms and contexts. We further calculate the saliency scores of extracted terms and rank them accordingly. Experimental results demonstrate that our proposed method generates results with high quality and wide coverage.