A MEMs-based Labeling Approach to Punctuation Correction in Chinese Opinionated Text

摘要

This paper presents a maximum entropy models based approach to punctuation prediction and correction for Chinese opinionated texts. This study involves three parts. First, we conduct a survey of punctuation errors in Chinese opinionated texts based on a corpus of online product reviews. Then, we propose a maximum entropy sequence labeling approach to Chinese punctuation prediction. Finally, we perform punctuation error detection and correction by comparing automatically-predicted punctuations with the relevant original punctuations in opinionated texts. Our experimental results show our system is effective for most punctuation errors in Chinese opinionated texts.