基于CNN的方面级跨领域情感分析研究

近年来，方面级情感分析吸引了越来越多学者的关注，但方面级跨领域情感分析存在没有标注数据，难以获得好的分类结果的问题。将上下文特征与方面特征进行融合，构建基于卷积神经网络和门控单元的情感分类模型，并利用少量目标领域数据集对模型进行微调来实现迁移学习，再用迁移学习后的模型对目标领域的数据进行方面级情感分析，有效解决了训练样本不足、准确率低的问题。人工标注了适用于方面级跨领域情感分析的中、英文语料，所提出的方法在中文数据集最优的F1值达到92.19%，英文数据集最优的F1值达到了86.18%，实验结果表明基于卷积神经网络的方面级跨领域情感分析方法有效提高了目标领域的情感分类准确性。

Aspect-Level Cross-Domain Sentiment Analysis Based on CNN

In recent years, more and more researchers focus on the aspect-level sentiment analysis field. However, for aspect-level cross-domain sentiment analysis problems, no labeled data may lead to difficulty to obtain good classification results. This paper constructs a sentiment classification model based on convolutional neural networks and gated units by merging the contextual features and aspect features, and then uses a small amount of target domain data to fine-tune the model, carry out transfer learning, and analyze aspect-level sentiment of target domain, which effectively alleviates the problems of insufficient training samples and low accuracy of results. Chinese and English corpora suitable for the cross-domain aspect sentiment analysis are manually annotated. The optimal F1 value of the proposed method for Chinese data set achieves 92.19% and the optimal F1 value for the English data set achieves 86.18%. The experimental results show that the cross-domain aspect sentiment analysis model based on CNN can effectively improve the accuracy of sentiment classification in the target domain.