【摘要】 中文短文本分类最重要的是提取和表示短文本特征,传统的表示方法静态训练词向量,存在特征稀疏、语义不足等问题,对此提出一种基于改进词性信息和ACBiLSTM的中文短文本分类模型。用BERT模型动态训练词向量,引入词性因子优化,将得到的具有词性信息的词向量作为输入层信息;经过卷积神经网络初步提取局部特征,通过双向长短时记忆网络进一步得到具有上下文时序信息的中间向量表示;结合注意力机制,突显关键信息,获得最终文本特征表示。在FudanNews、SougouNews及THUCNews数据集上将该模型与其他模型进行对比,实验结果表明,其准确率和F1值显著提高。更多还原

【Abstract】 The most important part of Chinese short text classification is how to extract and represent short text features. Traditional representation methods train the word vectors statically and has problems such as sparse features and insufficient semantics. To this end, a Chinese short text classification model based on improved part of speech information and ACBiLSTM is proposed. The BERT model was used to dynamically train the word vector, while introducing optimized parts of speech to improve it, and the obtained word vector based on part-of-speech information was used as the information of the input layer; the word vector was initially extracted through the convolutional neural network, and the bidirectional long-short-term memory network further obtained the intermediate vector representation with context time series information; it combined the attention mechanism to highlights the key information, and obtained the final text feature representation. Comparing with other models on the FudanNews, SougouNews, and THUCNews datasets shows that the accuracy and F1 value of the experiment are significantly increased.更多还原