深度学习框架下类别不平衡数据情感分析

Sentiment Analysis of Class Imbalance Data Under the Framework of Deep Learning

[目的/意义]将类别不平衡学习方法与深度学习相结合,提升数据不平衡分布下情感分析的性能。[方法/过程]提出LSTM深度学习框架下自适应不平衡数据情感分析方法。不平衡程度低时,对少数类样本进行过采样,然后利用LSTM进行深度学习训练;不平衡程度高时,对多数类样本进行多组均衡化欠采样,然后分别对每组训练数据学习LSTM模型,最后通过集成学习方法,获得最终情感分类结果。[结果/结论]在网络商品评论语料库上的实验结果显示,本方法可以提升类别不平衡数据的深度学习性能,实现不平衡数据的自适应情感分析。

[Purpose/Significance]This paper tries to improve the performance of sentiment analysis under imbalance data distribution by combining class imbalance learning with deep learning.[Method/Process]An adaptive imbalance data sentiment analysis method under the LSTM deep learning framework was proposed.When the degree of imbalance was low,the samples of minority class were oversampled and then LSTM was used for deep learning training.When the degree of imbalance was high,multiple groups of equalization undersampling was carried out for the samples of majority class,and then LSTM model was learned for each group of training data respectively.Finally,the final sentiment classification result was obtained through ensemble learning.[Result/Conclusion]The experimental results on the online commodity comment corpus showed that the method can improve the deep learning performance of class imbalance data and realize the adaptive sentiment analysis of imbalance data.