基于卷积神经网络的隐式评价对象识别

为解决课程评论中隐式评价对象识别问题，提出了一种基于文本分类的隐式评价对象的识别方法。首先通过word2vec 模型获得训练文本对应的词向量，获得短文本特征; 其次将短文本特征在TextCNN 中进一步提取高层次特征，通过K-max 池化操作后放入Softmax 分类器中进行训练得出分类模型; 最后利用训练好的分类器对隐式评价句进行分类，获取隐式评价句对应的评价对象。实验表明，基于卷积神经网络对隐式课程评论进行属性分类，课程评论的隐式评价对象识别正确率达到89． 9%，满足了课程评论中对隐式评价句对象识别的需求。

Implicit Opinion Targets Identification Based on Convolutional Neural Network

In order to solve the problem of implicit opinion targets recognition in course comments，a method of implicit opinion targets identification based on text classification was proposed in this paper． First，the word vectors corresponding to the training text were obtained by word2vec model which produced the short text features． Then，TextCNN was used to extract high-level features by obtainning classification model by pooling K-max and putting it into softmax classifier． Finally，the trained classifier was used to classify the implicit opinion sentences，and the corresponding opinion targets of the implicit opinion sentences were obtained． The experiment results show that the attribute classification of implicit curriculum reviews based on the convolutional neural network，the accuracy rate of implicit opinion targets identification in curriculum reviews is 89． 9%，which meets the needs of implicit opinion sentence in curriculum reviews．