A Study On Authorship Attribution For A Chinese Online Suicide Note

In some covert assassinations which are often disguised as suicide with a forged suicide note,suicide notes are vital evidence to prove whether the dead took his or her own life or was slayed by others.The evidence is usually obtained by authorship attribution of the suicide note.Therefore,identifying the author of the suicide note involved in cases is crucial in determining the case’s nature and disclosing criminal activities.As the most significant concern of authorship attribution is searching for reliable discriminant features,this study endeavors to test the discriminating power of those features based on formal Chinese syntax and discourse information theory to unveil more useful discriminant features for authorship attribution of a Chinese suicide note.Specifically,this study is to grip with the following three research questions: 1)What are the distinct features for differentiating different candidate authors of the Chinese suicide note? 2)How do the distinct features perform in attributing different candidate authors of the Chinese suicide note?3)Why these features are powerful in discriminating between or among different groups of candidate authors? Nine people were selected as the candidate authors of the suicide note.In order to train a fit model for authorship prediction of the suicide note,twenty training texts were collected from each candidate author,each text containing 600-750 words.Discriminant analysis was employed to test the discriminating and predictive power of selected features in all the possible combinations between or among authors.The results demonstrate that 1)features selected from formal Chinese syntax and from the micro perspective of discourse information theory(such as Adjective,Determiner,nounpredicate sentence,State,Quality,agent,elaboration)could discriminate different-author combinations for Chinese suicide note.2)all the three feature sets give gratifying performance in discriminating between or among different authors.When comparing cross-genre author combination and similar-genre author comparison,it is found that feature set 1 and feature set2 present a slightly better result in cross-genre comparisons than in similar-genre comparisons,while feature set 3 poses to the opposite result.Additionally,when comparing sociolinguistically different and sociolinguistically similar author combinations,sociolinguistically different author combinations presents more satisfactory results than sociolinguistically similar author combinations when using feature set 2 and feature set 3,while feature set 1 presents higher original and cross-validated accuracy rates in sociolinguistically similar author combinations than in sociolinguistically different author combinations.3)Even in five-author combinations,each feature set successfully attribute the query suicide note to its author,which proves that each author has their own idiolect style in terms of the features.An in-depth qualitative analysis of some representative examples selected from each author illuminates that each author shows distinct habitual use of lexis,sentence and information elements.It is suggested that culprits like working memory capacity,gender,text genre account for linguistic variation among authors.Chinese-oriented linguistic features can also achieve a decent result,which implicates more Chinese-oriented linguistic features for authorship attribution of Chinese texts should be sought in future research.