An Application of Deep Learning to detect Plagiarisms in Assigned Reports based on the Style Model

2018 Fiscal Year Final Research Report

• Project/Area Number: 16K00476
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Research Field: Learning support system
• Research Institution: Kobe University
• Principal Investigator: Murao Hajime 神戸大学, 国際文化学研究科, 教授 (70273761)
• Project Period (FY): 2016-04-01 – 2019-03-31
• Keywords: 盗用発見 / 機械学習 / 深層学習

Outline of Final Research Achievements

In this study, we have applied a Variational Recurrent Auto-Encoder (VRAE) to detect plagiarisms in assigned reports. VRAE is preliminarily trained by using "expression features" of reports submitted by students. Here, "expression features" are a number of punctuations, the location of new-lines, etc. VRAE learn a map from the expression features of each students' report onto the latent space. Then, whenever a student submits a new report, we extract the expression features from it and input to the trained VRAE, and estimate the author of the report by checking where the report will map on the latent space. We test the proposed method and validate possibility.

Free Research Field

社会システム科学

Academic Significance and Societal Importance of the Research Achievements

本研究により，授業の課題レポートのように，同じテーマについて書かれており，内容に基づいた比較による盗用発見が困難な場合においても，盗用の可能性をコンピュータにより指摘することができるようになり，教員の負担軽減につながる。また，本手法を従来研究されてきた，内容に基づいた盗用発見と組み合わせることにより，さらに精度を高めることが可能となり，より広範な盗用発見に適用できる可能性がある。