运用CNN-LSTM混合模型的短文本分类

Short text classification using CNN-LSTM hybrid model

针对分类短文本时卷积神经网络(CNN)只提取局部特征和长短时记忆网络(LSTM)学习计算量大、处理时间长且随着短文本文字量增加与上下文的联系会减弱的问题,给出了基于CNN-LSTM混合模型算法.该算法融合CNN对短文本的特征提取能力,降低了文本数据量;利用LSTM的记忆能力,充分学习短文本的全局特征,进而对短文本进行更加有效地分类.实验结果表明,CNN-LSTM混合模型对短文本的分类效果远远好于CNN模型和LSTM模型.

The convolutional neural network (CNN) only extracts local features;the long-term and short-term memory network (LSTM) has a large amount of learning computation and a long processing time;the context connection will weaken with the increase of short text volume. In order to solve the problems above, this paper proposes a hybrid model algorithm based on CNN-LSTM. The proposed algorithm integrates the feature extraction ability of CNN to short text, reduces the amount of text data, utilizes the memory ability of LSTM to fully learn the global features of short text, and then classifies short text more effectively. Experimental results show that this CNN-LSTM hybrid model is much better than the CNN model and LSTM model in terms of short text classification.