融合BERT和知识图谱的文本情感分析模型  

Text Sentiment Analysis Model Based on BERT and Knowledge Graph

作　　者：刘斐瑜 俞卫琴 

机构地区：[1]上海工程技术大学数理与统计学院,上海

出　　处：《建模与仿真》2023年第4期4195-4200,共6页Modeling and Simulation

摘　　要：近年来,在线评论数与日俱增,这些在线评论涉及的范围较广,在某些特定领域里专业性较强。深度学习模型在大规模语料库上学习到了自然语言的通用表达,但在一些专业领域里对语义信息的提取还不够充分。针对该问题,本文提出一种基于BERT和知识图谱的情感分析模型。该模型在文本输入中利用知识图谱注入专家知识,再利用BERT进行词向量化,生成包含上下文语义信息的动态词向量,最后通过全连接层输出情感极性。结果表明,该文章提出的模型在几个公开数据集上提升了准确率,具有实际意义。In recent years, the number of online comments has increased, and these online reviews are widely involved, and they are professional in some specific areas. Deep learning models have learned the common expression of natural languages in large -scale corpus libraries, but for some professional fields, it is not sufficient to rely on context to extract semantic information. In response to these is-sues, the article proposes an emotional analysis model based on BERT and knowledge graph. In the input, the model uses the knowledge graph to inject expert knowledge, and then use BERT for quantification to generate a dynamic word vector containing the above semantic information, and finally output the emotional polarity through the full connection layer. The results show that the model proposed by the article has improved accuracy on several public datasets and has practical significance.

关 键 词：知识图谱 文本输入 词向量 文本情感分析 在线评论 连接层 语义信息 自然语言 

分 类 号：G63[文化科学—教育学]