基于分段卷积神经网络的文本情感极性分析

Analysis of Text Emotion Polarity Based on Piecewise Convolution Neural Network

传统文本情感分析方法存在文本信息分析准确度较差、召回率偏低且时间成本较高问题,提出基于分段卷积神经网络的文本情感极性分析方法。卷积层使用滤波器完成局部特征提取,获得卷积核函数运算法产生的特征图,将其输入至下采样层,由下采样层输出局部最优特征。利用分化池操作,将卷积向量中的最大极性值融合成向量,并使用非线性函数对该向量计算,获得文本句子的极性特征。运用softmax分类器与Dropout算法,随机将原始输入数据按照一定比例置0,没有置0的则进行运算与连接,最后对文本向量和网络参数进行优化完成情感极性分析。仿真结果证明,所提方法准确性较高、召回效果较好,且时间成本更低,对比传统方法更具有较好的应用前景。

Traditionally, the text sentiment analysis method has some problems, such as poor classification accuracy, low recall rate and high time cost. Therefore, a method to analyze text sentiment polarity based on piecewise convolution neural network was proposed. In the convolution layer, the filter was used to extract the local feature, and then feature map generated by convolution kernel operation method was obtained. After that, the feature was input to the lower sampling layer, and the local optimal feature was output from the lower sampling layer. Moreover, the maximum polarity in convolution vectors was fused into a vector by piecewise pool. Meanwhile, the nonlinear function was used to calculate this vector, so as to obtain the polarity feature of sentences. In addition, SoftMax classifier and Dropout algorithm were used to randomly set the original input data to 0 at a certain percentage. If it is not set to 0, the operation and connection were carried out. Finally, the text vector and network parameters were optimized to complete the analysis of sentiment polarity. Simulation results prove that the proposed method has higher accuracy, better recall effect and lower time costs, so it has better application prospects than traditional methods.