中文微博作者身份识别研究

Authorship Attribution of Chinese Microblog

本文针对中文微博篇幅短小、无间隔标记等特点,建立了由词汇特征、浅层句法和深层句法特征集组成的中文微博作者文体特征模型,选取支持向量机、序列最小优化支持向量机、朴素贝叶斯和决策树算法在公开微博语料上进行算法对照实验、特征集组合实验和各组文体特征的作者身份识别实验。实验结果验证了本文模型在中文微博作者身份识别任务中的高准确率、召回率和时间效率。

In order to meet the the urgent demand of Chinese Microblog authorship attribution,we established a multidimensional stylistic features model consists of the lexical features, shallow syntactic features and deep syntactic features. This Chinese Microblog stylistic features model has been verified through control experiments and grouping experiments using LibSVM, Sequential Minimal Optimization SVM, Naive Bayesian and Decision Tree algorithm on public microblog corpus. The experimental outcome verified the contribution of each feature-dimension and the good performance of our model in the precision, recall and computing time.