基于Transformer的新闻情感分析算法

News Sentiment Analysis Model Based on Transformer

目前,新闻文本情感分析大多没有关注文章结构对于判断情感极性的引导作用,且往往仅关注文章内容,导致获取信息较为局限。为此,本文提出了基于情感重点句融合知识图谱的Transformer模型。使用了多特征情感重点句抽取算法,有效地降低了文本维度;引入了知识图谱,对新闻中涉及的内容进行知识增强,丰富了文本信息;结合新闻的结构特点使用Transformer模型进行情感分析。实验结果表明,该模型性能与基线模型相比有一定提升,且融合知识图谱可以进一步有效提升模型性能,是一个值得关注的研究方向。

At present,most of the news sentiment analysis research does not pay enough attention to the guiding role of the news struc⁃ture and usually focus on content of the article,which leads to limited source of information collected for analysis.In order to improve accuracy of sentiment analysis,this paper proposes a Transformer model based on key sentiment sentence with the integration of knowledge graph.First,a key sentiment sentence extraction algorithm is used to reduce dimension of the texts,and then knowledge graph is introduced to enrich message of the texts.Finally,conduct sentiment analysis of the news texts according to the structure of the news using Transformer model.Experimental results demonstrate that the proposed model performs better compared with baseline models.In addition,knowledge graph can further improve performance of the model.