基于Keras框架下的航空公司评论数据的情感分析

Emotional analysis of airline comments data based on Keras framework

长短时记忆网络(LSTM)模型可有效改善循环神经网络梯度消失,但在评论文本情感分类中未能达到一定普适性。针对此问题,文章基于Keras框架,采用模型融合的方法,对LSTM网络和反向循环神经网络结构进行融合和改进,即D-LSTM、BRNN以及M-BRNN。以航空公司评论数据集为例,采用改进的模型对文本进行分类,研究不同参数对模型性能的影响。经仿真对比分析结果表明,该模型较已有的文本分类模型鲁棒性更好,准确率比传统的方法提高了3.7%。

LSTM network model can effectively improve the gradient disappearance of recurrent neural network,but it fails to achieve certain universality in the emotion classification of comment texts.To address this problem,based on the Keras framework,this paper adopted the method of model fusion to fuse and improved the structure of LSTM network and reverse recurrent neural network,namely,D-LSTM,BRNN and M-BRNN.Taking the data set of airline comments as an example,an improved model was used to classify the text,and the influence of different parameters on the model performance is studied.In comparison,the model is more robust than the existing text classification model,and the accuracy rate increases by 3.7% compared with the traditional method.