Automatic Indexing System with Compound Noun Decomposition and Correction of Word Boundary Error in Real Korean Text

摘要

For the effective automatic indexing of documents, proper keywords should be extracted from documents in which they occur. In Korean text, the keyword extraction has such three key issues as noun extraction, compound noun decomposition, and word-boundary error handling. In most previous researches, each of the issues has been so independently handled using a different method that it would be difficult to effectively solve the problems in a single integrated system. In this paper we suggest an integrated method to handle the three issues, in which the correction of word boundary errors plays an essential role because it can make the other problems easier. The proposed indexing method is based on the CYK algorithm for segmentation and the Viterbi algorithm for word boundary checking. And also, instead of corpus-based statistical information, it just relies on some heuristics for high accuracy in a real text. The experimental evaluation showed a promising result with the indexing recall and precision of 94.44% and 92.83%, respectively.