このエントリーをはてなブックマークに追加
ID 56446
Title Alternative
Proposing an Unsupervised Approach to Evaluate Essays Using IDF on Reference Data
FullText URL
Author
Ohno, Masayuki Graduate School of Natural Science and Technology, Okayama University
Takeuchi, Koichi Graduate School of Natural Science and Technology, Okayama University
Motojin, Kota Graduate School of Natural Science and Technology, Okayama University
Obata, Yuya Graduate School of Natural Science and Technology, Okayama University
Taguchi, Masahiro Graduate School of Humanities and Social Science, Okayama University
Inada, Yoshihiko Graduate School of Education, Okayama University
Iizuka, Masaya Institute for Education and Student Services, Okayama University
Abo, Tatsuhiko Institute for Education and Student Services, Okayama University
Ueda, Hitoshi Institute for Education and Student Services, Okayama University
Abstract
大学入試において2020 年から記述式問題が導入されることから記述式の問題を自動で採点する手法の開発が求められている.本論では,エッセイタイプの小論文課題を対象に,課題に関連する参照データとWikipedia 全文から作成したidf を利用した事前採点不要な自動採点手法を提案する.先行研究において,日本語小論文を対象とした自動採点では,多くの事前採点が必要となり,実際の数百人規模の試験では利用することが難しいと考えられる.そこで本研究では,事前採点が不要な小論文採点手法を提案する.また,小論文の模擬試験を実施して小論文データを構築する.構築した小論文データに対して採点手法を用い,実験を行い評価する.また小論文データの人手による採点に対しても評価を行う.評価実験の結果neologd 辞書を利用した形態素解析器を用いて, idf 値を利用した形態素の一致数が,人手の評価値と相関が高いことを示す.
Abstract Alternative
In this paper, we describe an on-going study of developing an automatic essay-scoring system in Japanese. Essay scoring systems have already been developed and used mainly in English, while not many previous studies have been done on Japanese essay evaluations. Most of the methods and systems of automatic essay evaluation need not small number of previously human-graded essays for calibrating the parameter of regression functions or parameter of machine learning. The previous studies show the high performance for essay evaluation task, however, it must be not easy to assume large graded essays in, for example, actual tests or entrance examinations. Thus, we take a approach to evaluate Japanese essays without previously human-graded essays but with assuming reference data related to essay questions. The proposed method is a simple one, that is, evaluating the essays with co-occurrences with the reference data in their words or morphemes. In the method technical terms would be given high scores using neologd dictionary and idf values. Experimental results show that the proposed method works well in our developing Japanese mock trial writing tests. Key words automatic scoring of essays, human annotation, supporting
Keywords
自動採点 (automatic scoring of essays)
アノテーション (human annotation)
採点支援 (supporting system of essay evaluation)
idf
neologd
Published Date
2018-12
Publication Title
IEICE technical report. Natural language processing and models of communication
Volume
volume118
Issue
issue355
Publisher
電子情報通信学会
Start Page
103
End Page
108
ISSN
09135685
NCID
AA11524632
Content Type
Journal Article
language
日本語
OAI-PMH Set
岡山大学
Copyright Holders
copyright@2018
File Version
publisher