基于RoBERTa和BiGRU-AT的微博评论情感分类模型
A sentiment classification model for Weibo comments combining RoBERTa and BiGRU-AT
曾孟佳 1杨卓 2黄旭1
作者信息
- 1. 湖州师范学院信息工程学院,湖州 313000;湖州学院电子信息学院,湖州 313000;湖州市城市多维感知与智能计算重点实验室,湖州 313000
- 2. 湖州师范学院信息工程学院,湖州 313000
- 折叠
摘要
针对传统静态词向量如glove无法表示多义词的缺陷,以及现有微博情感分类模型对于隐式评论文本特征提取能力不足等问题,提出了一种结合RoBERTa和BiGRU-AT的微博评论情感分类模型.用预训练模型RoBERTa得到融合句子语境的动态词向量;然后采用BiGRU-AT模块的双向门控循环单元提取文本序列特征、注意力机制捕获文本序列中的关键情感信息;最后利用归一化指数函数输出情感倾向结果.实验结果显示,该模型与现有常用经典模型相比,精确率和F1值均取得了较好的效果,具有较好的实用价值.
Abstract
Aiming to address the limitations of traditional static word vectors,such as glove,in representing polysemous words,as well as the insufficient capability of existing Weibo sentiment classification models in extracting implicit textual features from comments,a novel Weibo comment sentiment classification model that combines RoBERTa and BiGRU-AT is proposed.The pre-trained model RoBERTa is utilized to obtain dynamically contextualized word vectors that incorporate sentence context.Subse-quently,the BiGRU-AT module,consisting of bidirectional gated recurrent units,is employed to extract textual sequential features,while an attention mechanism is employed to capture crucial information within the text sequence.Finally,sentiment classification is performed using Softmax.Experimental results demonstrate that compared to commonly used conventional models,the proposed model achieves superior performance in terms of precision and F1 score,indicating its practical utility.
关键词
文本情感分类/RoBERTa/BiGRU/注意力机制Key words
text emotion classification/RoBERTa/BiGRU/attention mechanism引用本文复制引用
基金项目
教育部人文社会科学一般项目(20YJCZH005)
浙江省湖州市工业攻关项目(2018GG29)
国家级大学生创新创业训练项目(202313287007)
出版年
2024
NETL
NSTL
万方数据