Plagiarism Detection: Similarity Analysis Of Academic Discourse From Discourse Information Analysis And Appraisal Perspectives

This thesis has tentatively conducted similarity analysis between original andplagiarized academic papers based on discourse information theory and appraisaltheory. The objective of this paper is to extract a set of macro and micro featureseffective for such similarity analysis. In this thesis, qualitative and descriptivemethods were employed to extract the macro and micro features. Then, an experimentwas conducted to evaluate the reliability of the extracted macro features.The research findings are as follows: firstly, four major types of structuralfeatures, namely, information structure, kernel proposition, information density andinformation knots, are extracted as features to detect the similarities from macro level.As with information knot types, we have categorized them according to the frequencyand the degree between subjectivity and objectivity. Secondly, such micro features asattitudinal evaluations, graduations and information element types, including Process,entity and condition are extracted for more delicate analysis of similarity between theoriginal and the plagiarized version. Thirdly, as with the effectiveness of the macrofeatures, via the combination of all structural features with the exception ofinformation density, we have identified five of the six plagiarized experimentalabstracts.This research distinguishes itself from other similarity analyses because of itsnew angle of study, that is, the adoption of discourse information and appraisalperspectives. Theoretically and practically, this paper is significant in this area.Theoretically, it enriches the research of discourse information and appraisal theory.Practically, it improves similarity analysis from linguistic perspective, consistent withits aim of saving academic resources and avoiding academic repetition.