Web-Based Verification on the Representativeness of Terms Extracted from Single Short Documents

Single document summarization is useful for extracting the major ideas from huge amount of daily information. However, it's a challenge to distinguish the relative importance among terms. In this paper, we propose a Web-based approach to term verification. Search-results of extracted terms are utilized as their expanded representation, and their similarity with the original document are calculated as an estimate of term representative ness. We experimented with term extraction methods on multilingual news extracts and compared the effectiveness of term verification with various Jaccard similarity measures. The experimental results show the feasibility of Web-based verification on the representativeness of extracted terms.