Abstract: In order to solve the problem of emotional ambiguity caused by the change of contextual order in traditional neural networks. A multi-channel semantic synthesis convolution neural network（SFCNN） is proposed. Firstly, the improved emotional attention mechanism focuses on the emotional weighting of word vectors. Secondly, the emotional tendency word vector is fused by multi-channel semantic synthesis layer to generate deep semantic vectors with text context semantic information, and construct sentiment classification model. Finally, the gradient descent algorithm for adaptive learning rate is used to optimize the model parameter, and to complete emotion classification task. Multiple Weibo data sets are used to verify the effectiveness of the proposed algorithm. The experimental results show that the improved affective tendency attention mechanism with the multi-channel semantic fusion CNN has better emotion classification ability, and the gradient descent algorithm of adaptive learning rate can complete the convergence of the model faster.

Key words: Convolution Neural Network（CNN）, attention, emotional classification, multi-channel

摘要： 为了解决传统神经网络由于上下文语序变化而导致的情感歧义问题，提出一种多通道语义合成卷积神经网络（SFCNN）。使用改进的情感倾向注意力机制对词向量进行情感加权操作；将情感倾向词向量进行多通道语义合成，生成带有文本上下文语义信息的深度语义向量，构建情感分类模型；使用自适应学习率的梯度下降算法对模型参数进行优化，完成行情感分类任务。为了验证改进算法的有效性，使用多种微博数据样本集在提出的模型上进行对比实验。实验结果表明，改进的情感倾向注意力机制结合多通道语义合成卷积神经网络具有较好的情感分类能力，并且自适应学习率的梯度下降算法可以更快地完成模型收敛工作。

关键词: 卷积神经网络, 注意力, 情感分类, 多通道