基于注意力机制的双BERT有向情感文本分类研究

在计算社会科学中，理解政治新闻文本中不同政治实体间的情感关系是文本分类领域一项新的研究内容。传统的情感分析方法没有考虑实体之间情感表达的方向，不适用于政治新闻文本领域。针对这一问题，本文提出了一种基于注意力机制的双变换神经网络的双向编码表示(bi-directional encoder representations from transformers, BERT)有向情感文本分类模型。该模型由输入模块、情感分析模块、政治实体方向模块和分类模块四部分组成。情感分析模块和政治实体方向模块具有相同结构，都先采用BERT预训练模型对输入信息进行词嵌入，再采用三层神经网络分别提取实体之间的情感信息和情感方向信息，最后使用注意力机制将两种信息融合，实现对政治新闻文本的分类。在相关数据集上进行实验，结果表明该模型优于现有模型。

Dual BERT directed sentiment text classification based on attention mechanism

Understanding the emotional relationships between different political entities in political news texts is a new research topic in the text classification field in computational social science. Traditional methods of sentiment analysis cannot be applied to political news texts because they do not consider the direction of emotional expression between entities. This study proposes a dual BERT-directed sentiment text classification model based on the attention mechanism, which consists of four modules: input module, sentiment analysis module, political entity direction module, and classification module. The structure of the sentiment analysis module and the political entity direction module are identical. Both employ the BERT pretraining model to embed the input information, a three-layer neural network to extract the emotional information or emotional direction information between entities, and an attention mechanism to combine these two kinds of information to classify political news texts. Experiments on comparable data sets show that the model outperforms existing models.