Abstract: This paper first points out that shallow parsing problems can be converted into categorizaion problems，then we show how to cope with shallow parsing with random forests.Next，one variant of random forests is given，namely Base+Bootstrap.The experiment result shows that Fβ can reach the value of 92.25% under the model of Base+Bootstrap.

Key words: shallow parsing, random forests, decision trees, Bootstrap

摘要： 文章首先阐述浅层句法分析可以转化为一个分类问题，然后论述了如何用随机森林的方法来完成这个分类任务。接下来对随机森林算法进行了改进，即基本模型+Bootstrap方式。实验结果显示，针对CoNLL2000提出的浅层句法分析任务，基本模型+Bootstrap方式的Fβ值可以达到92.25%，较基本模型有明显提高。

关键词: 浅层句法分析, 随机森林, 决策树, Bootstrap