基于深度学习的体育评论文本情感分类研究

Research on sentiment classification of sports comment text based on deep learning

体育赛事作为公共话题,是网络舆情的来源之一。针对传统词向量语义表示质量不高,深度学习模型特征提取能力不强等问题,提出了基于ERNIE-BiSRU-AT的体育赛事评论文本分类模型。利用预训练模型ERNIE提取词的动态向量表示,BiSRU-AT模块捕获文本的上下文序列特征,并聚焦于对情感极性贡献较大的词。在真实微博女排赛事评论数据集进行实验,ERNIE-BiSRU-AT模型F1分数达到92.35%,高于实验对比的其他模型,验证了模型的有效性。

As a public topic,sports events are one of the sources of network public opinion.To address the problems of low quality of traditional word vector semantic representation and weak feature extraction ability of deep learning model,sports event comment text classification model based on ERNIE-BiSRU-AT is proposed.The pre-training model ERNIE is used to extract the dynamic vector representation of words,and the BiSRU-AT module could capture the context sequence characteristics of text,and focuses on the words that contribute greatly to the emotional polarity.The experiment on the real microblog women’s volleyball event comment data set shows that the F1 score of ERNIE-BiSRU-AT model reaches 92.35%,which is higher than other models compared in the experiment,which verifies the effectiveness of the model.