基于特征融合分段卷积神经网络的情感分析

Sentiment analysis based on piecewise convolutional neural network combined with features

为解决文本情感分析任务中传统卷积神经网络模型存在没有考虑句子结构和过度依赖于所输入的词向量问题,提出一种基于特征融合的分段卷积神经网络模型(PF-CNN)。考虑句子的结构,分段池化提取句子的主要特征;利用词性特征与词向量融合的方法,解决词向量无法区分同义词的问题。实验结果表明,与传统的文本卷积神经网络相比,PF-CNN模型在情感分析任务上,准确率、召回率和F1值等指标都有显著提升。

To solve problems that traditional convolutional neural network model does not consider the structure information of sentences,and results of the model are too dependent on input word vector,which appear in sentiment analysis research,sentiment analysis based on piecewise convolutional neural network combined with features(FP-CNN)was proposed.The structure of sentences was taken into consideration by adopting the piecewise convolution strategy to extract the main features of sentences.The problem that word vector cannot distinguish the synonyms was solved after combining lexical feature with word vector.Experimental results show that compared with traditional text CNN,PF-CNN has significant improvement in the performance of sentiment analysis task in terms of precision,recall,F1 score and so on.