个人言语判别特征在短文本作者鉴别中的应用

The Application of Idiolect Features to Authorship Attribution for Chinese Short Texts

目的以法律语言学为视角,通过测试语用、语篇语义以及语篇信息文本特征值对文本作者的判别能力,探究短文本作者鉴别或同一认定的方法。方法采用实验、语篇分析和统计的方法,对4位作者的28篇微博(每人7篇)共11种组合形式(二人组、三人组和四人组)逐一进行了文本特征值的测试和文本作者的判别分析。结果从语用、语篇语义学以及语篇信息领域抽取的5个特征值的不同组合对4名作者的所有11种判别组合都能进行显著区分,判别正确率达到85.7%~100%。结论基于4位作者微博文本的判别分类器已经建立并可以继续推演用于其他短文本作者的鉴别分析。

Objective From perspective of forensic linguistics,this study explores the methods of identification of short text authors by testing the features in pragmatics,discourse semantics and discourse information for authorship attribution of Chinese short texts—Microblog.Methods The blog texts used in the study include 28 Microblogs written by four authors(seven articles per person)by using experimental,textual analysis,and statistical methods.All the possible 11 combinations of the four authors are tested and attributed.Results The five different combinations of eigenvalues extracted from the fields of pragmatics,discourse semantics and discourse information can significantly distinguish all 11 discriminative combinations of the four authors.It could be concluded that the extracted features in pragmatics,discourse semantics and discourse information can significantly distinguish Microblogs of different authors,and the discrimination accuracy rate is 85.7%-100%.Conclusion Based on these results,text-based classifier of the four authors proved to be valid statistically and applicable to the authorship attribution of other types of Chinese short texts.