摘要
研究词深度特征池化法的句子级情感分类特征表示,在进行词向量学习时,首先利用中科院分词器对语料进行分词,通过word2vec中的CBOW模型计算语料中词的深度特征词向量;在情感分类过程中,首先对词深度特征采用均值、最值等池化方法获得整句话的特征,并以此作为神经网络模型的输入,通过一个线性层、Sigmoid激活层以及线性分类标注层,来判决得到该句的情感倾向。通过在当当书评语料上进行实验,结果表明均值与最值池化拼接的特征方法取得较好的分类效果,能够更好地表征句子级情感特征。
Studies the pooling method of sentence level sentiment classification characteristic, to learn the word vector, uses ICTCLAS2016 to segments the sentences of corpus, and the depth feature vectors are calculated by CBOW model in word2 vec. In the classification of texts sentiment, gets the sentence feature by pooling the word depth features in a variety of ways, which is the input of neural network model.And then, those features will input a linear layer, sigmoid active layer and a classification linear layer to get the sentence emotional tendencies. The results of Book Review Corpus in Dangdang shows that the combination of the meaning and extreme value pooling method can achieve better classification results, which have a better express to characterize the sentence level emotional features.
出处
《现代计算机》
2016年第6期3-8,共6页
Modern Computer
基金
四川省科技支撑计划项目(No.2012GZ0091)
四川大学青年基金项目(No.2012SCU11033)
