Using Speech Acts to Categorize Email and Identify Email Genres
Using Speech Acts to Categorize Email and Identify Email Genres
复制标题
使用言语行为对电子邮件进行分类并识别电子邮件类型
DOI：
10.1109/hicss.2006.528
复制
发表时间：
2006
期刊：
影响因子：
--
通讯作者：
R. E. Sabin
中科院分区：
文献类型：
--
作者：
Jade Goldstein;R. E. Sabin
我们定义电子邮件的体裁，以及一个子集的“言语行为”相关的电子邮件增强电子邮件的具体话语。在基于这些电子邮件行为创建了一组真实的电子邮件后，我们比较了两个分类器（随机森林和SVM-light）在识别电子邮件及其相应类型的主要通信意图方面的性能。我们的实验使用来自两个动词词汇的功能集，以及包含选定的电子邮件的特点的功能集。结果表明，更好的分类器的准确性，使用的动词词汇与更大的类的数量较少，和使用词性标记，专注于只选择动词，导致性能略有下降。单独使用电子邮件特征集比单独使用动词词典中的任一个导致更好的性能，但是使用较小的动词词典和电子邮件特征集的组合获得最佳结果。
We define genres of email as well as a subset of "speech acts" relevant to email enhanced for email specific discourse. After creating a ground truth set of emails based on these email acts, we compare the performance of two classifiers (Random Forests and SVM-light) in identifying the primary communicative intent of the email and its corresponding genre. We experiment with using feature sets derived from two verb lexicons as well as a feature set containing selected characteristics of email. Results show better classifier accuracy using the verb lexicon with the smaller number of classes over the larger, and that using part of speech tagging to focus on selecting only verbs, causes a slight drop in performance. Using the email characteristics set alone results in better performance than either of the verb lexicons alone, but the best results are obtained using a combination of the smaller verb lexicon and the email characteristics set.