Writing style change detection(WSCD)is the task of automatically identifying transitions in the writing style of a text,which may include changes in authorship.While traditional WSCD research has focused on style chan...Writing style change detection(WSCD)is the task of automatically identifying transitions in the writing style of a text,which may include changes in authorship.While traditional WSCD research has focused on style changes between human authors,the increasing use of AI-generated text introduces a new challenge:detecting transitions between human-written and AI-generated content.This survey,to the best of our knowledge,is the first to address WSCD in both human-human and human-AI collaborative contexts.We begin by providing a systematic review that investigates task descriptions,and analyzes,categorizes,and compares the existing approaches,methodologies,and datasets commonly used in WSCD research.Additionally,we highlight the challenges and limitations in the current literature including the lack of comprehensive and realistic datasets,limited language coverage,and insufficient focus on domains where textual integrity is crucial,such as education.Furthermore,we point out emerging research directions in WSCD,including the potential use of federated learning(FL)to enable the use of sensitive data sources.We highlight the intersection between traditional human-human style change detection and machine-generated text detection,aiming to facilitate knowledge transfer between these do-mains that can benefit emerging studies focused on human-AI collaborative writing.展开更多
基金the Natural Sciences and Engineering Research Council of Canada(NSERC)(No.RGPIN-2020-06482).
文摘Writing style change detection(WSCD)is the task of automatically identifying transitions in the writing style of a text,which may include changes in authorship.While traditional WSCD research has focused on style changes between human authors,the increasing use of AI-generated text introduces a new challenge:detecting transitions between human-written and AI-generated content.This survey,to the best of our knowledge,is the first to address WSCD in both human-human and human-AI collaborative contexts.We begin by providing a systematic review that investigates task descriptions,and analyzes,categorizes,and compares the existing approaches,methodologies,and datasets commonly used in WSCD research.Additionally,we highlight the challenges and limitations in the current literature including the lack of comprehensive and realistic datasets,limited language coverage,and insufficient focus on domains where textual integrity is crucial,such as education.Furthermore,we point out emerging research directions in WSCD,including the potential use of federated learning(FL)to enable the use of sensitive data sources.We highlight the intersection between traditional human-human style change detection and machine-generated text detection,aiming to facilitate knowledge transfer between these do-mains that can benefit emerging studies focused on human-AI collaborative writing.
