Bag-of-Embeddings-for-Text-Classification课件.pptxVIP

Peng Jin(1), Yue Zhang(2) , Xingyuan Chen*(1) , Yunqing Xia(3) (1) School of Computer Science, Leshan Normal University (2) Singapore University of Technology and Design, Singapore (3) Search Technology Center, Microsoft, Beijing China;精品文档;Typical workflow for supervised text classification;词形的问题：无法区分下面的情况 多义现象 同一个含义在不同类中上下文也可能不同： ;词向量的特征词;词向量如何用于文本分类？;如何利用词向量进行文本分类？ ;Trianing Model;Bag-of-Embedding Model;Training ;分类决策（BoE）;Dataset twenty newsgroup (20NG) [Lang, 1995] ， which consists of 11,314 training instances and 7,532 test instances, which are partitioned (nearly) evenly across the classes. Reuters-21578 Lewis [1995] , a imbalanced dataset. We choose this benchmark dataset as follow Debole and Sebastiani [2005] and perform the R8 evaluation . Baseline SVM: linear SVM using the package LIBSVM (Wang et al. 2013), 10-fold cross-validation, TWE: Learning topic embeddings and word embeddings for text classification (Liu et al. 2015]) GoW: Three graph-of-words approaches for text classification which capture the distance dependency between words (Rousseau et al. 2015) Evaluation Metrics AC: the standard accuracy MF1: DescLDA#2: the macro-average F1- score;Our model outperforms the state-of-the art models TWE and GoW on the balanced and the imbalanced data sets.;这个图形放映了BoE模型在文本分类时是如何起作用的。目标词对所在的环境敏感。这是BOE模型能够取得好的效果的原因。我们的方法能利用上下文信息！;反映了类向量是描述的所在类的上下文信息，不同类之间差异巨大。;We built a text classifier by using a Naive Bayes model with bag-of-embeddings probabilities. Our model is conceptually simple, with only parameters being embedding vectors. Experiments on two standard datasets showed that our model outperforms state-of-the-art methods for both balanced and imbalanced data. ;