的领域情感词典构建方法  

作　　者：谢云高 

机构地区：[1]湖南铁道职业技术学院,湖南株洲412001

出　　处：《科技创新与应用》2025年第12期48-51,共4页Technology Innovation and Application

基　　金：湖南省教育厅科学研究项目(23C0805);湖南铁道职业技术学院校级一般课题(KJ202201)。

摘　　要：针对基于种子词扩展的领域情感词典构建方法存在词语上下文语义信息提取能力较低从而导致构建的领域情感词典领域性较差的问题,该文提出一种基于BERT的领域情感词典构建方法(BERT-SentiL)。该方法将领域情感词典的构建视为对词或短语的情感分类任务,运用BERT模型提取候选情感词的语义特征表示,然后将获取的语义特征表示输入Softamx并得到每个候选词所对应类别(正、负)的置信度(概率),最后依据置信度对候选词进行情感分类,构建领域情感词典。实验结果表明,采用BERT-SentiL方法构造的领域情感词典拥有较好的领域表达性和情绪识别有效性。Aiming at the problem that the domain sentiment dictionary construction method based on seed word expansion has low ability to extract word context semantic information,which leads to poor domainness of the constructed domain sentiment dictionary,a BERT-based domain sentiment dictionary construction method(BERT-SentiL)is proposed.This method regards the construction of a domain sentiment dictionary as an emotion classification task of words or phrases,uses the BERT model to extract the semantic feature representations of candidate sentiment words,and then inputs the obtained semantic feature representations into Softtamx to obtain the confidence(probability)of the category(positive and negative)corresponding to each candidate word.Finally,the candidate words are emotionally classified according to the confidence to build a domain sentiment dictionary.Experimental results show that the domain sentiment dictionary constructed using the BERT-SentiL method has good domain expressiveness and emotion recognition effectiveness.

关 键 词：情感词典构建 BERT 词向量 情感分析 语义特征 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]