基于注意力机制的混合神经网络模型的文本分类
首发时间:2019-04-22
摘要:文本分类是自然语言处理(NLP)中的重要任务之一。在文本分类中,句子建模是至关重要的。在已存在的工作中,卷积神经网络(CNN)能有效捕捉空间的局部相关性,循环神经网络(RNN)的变体双向长短期记忆神经网络模型(BiLSTM)能够从历史信息和未来信息中学习长期依赖性。针对它们的不同建模能力优势,本文提出了一种基于注意机制的混合神经网络模型。该模型首先基于BiLSTM引入一个贡献率来调整历史信息和未来信息的不同影响,并应用注意力机制将CNN与BiLSTM结合起来,运用注意力权重突出关键信息来缓解文本信息的丢失。实验结果表明,引入注意机制和贡献率可以有效提高文本分类的性能。
关键词: 文本分类 卷积神经网络 循环神经网络 长短记忆神经网络模型 双向长短记忆神经网络模型 注意力机制 贡献率
For information in English, please click here
Text Categorization Using Hybrid Neural Network Model Based on Attention Mechanism
Abstract:Text categorization is one of important tasks in natural language processing(NLP). In text categorization, sentence modeling is critical. In exsisting schemes, local correlation of spatial can be effectively captured by convolutional neural network (CNN), and long-term dependencies, historical information and follwing information can obtained by BiLSTM a variant of recurrent neural network (RNN). For their different modeling capabilities, this paper proposes a hybrid neural network model based on the attention mechanism. It first introduces a contribution rate based on BiLSTM to adjust the different influences on historical information and following information and applies the attention mechanism to combine CNN with BiLSTM to figure out the weight of key information for the relief of the loss of text information. The experimental results show that the introduction of attention mechanism and the contribution rate can effectively improve the accuracy of text categorization.
Keywords: Text categorization Convolutional Neural Network Recurrent Neural Network Long Short-Term Memory Bidirectional LSTM Attention mechanism Contribution rate
基金:
引用
No.****
同行评议
勘误表
基于注意力机制的混合神经网络模型的文本分类
评论
全部评论0/1000