国际中文开源期刊平台

基于深度学习模型的正负向文本情感分类

Classification of Text Sentiment on Positive and Negative Tendencies via Deep Learning-based Approaches

社会科学进展 / 2024,6(6)：1397-1405 / 2024-12-27 83 117

查看全文 下载文档

• 作者： 邓依霖      徐舒文      高昕     
• 单位：

  江苏师范大学，徐州

• 关键词： 文本情感分类；自然语言处理；深度学习；情感倾向；微博
• Text sentiment classification; Natural language processing; Deep learning; Sentiment tendency; Microblog
• 摘要： 本文结合文本情感分类的核心目标和研究现状，提出了一种基于深度学习的文本情感分类方法。通过对传统机器学习方法的总结与评估，分析其在处理复杂情感表达时的局限性；运用深度学习方法，搭建了一个正负情感分类模型，可用于微博评论的数据分析与测试。该深度学习模型基于正负情感的文本分类任务，对海量文本数据实现预处理和训练测试。实验结果表明，批处理次数在1000次以内，测试集可达平均98%的准确度；对十万多条正负情感文本，可实现准确分类。与此同时，作者针对文本情感分类难题，如结构不良与讽刺文本、粗粒度情感分析、文化意识缺乏、依赖数据和注释，以及词嵌入局限性等，作了总结性评述和展望。
• This paper proposes a deep learning-based approach on text sentiment classification according to the kernel objective and current research status for text sentiment classification. By summarizing and evaluating conventional machine learning schemes, their limitations in dealing with complex emotional expressions are analyzed. A deep learning-based approach is adopted to construct a classification model on positive and negative sentiments for analyzing the data and testing microblog comments. This deep learning model is based on tasks for text sentiment classification towards positive and negative tendencies, which enables implementation of preprocessing, training, and testing massive text data. Experimental results indicate that when the batch size is within 1000 times, an average accuracy of 98% is reached in the test set, while precise classifications on more than 100,000 positive and negative text sentiment are achieved. Meanwhile, concluding remarks and outlooks are presented with respect to difficult problems on emotional text classification, i.e., ill-structured and ironic texts, coarse-grained sentiment analysis, lack of cultural awareness, dependence on data and annotations, as well as the limitations of word embeddings.
• DOI： https://doi.org/10.35534/pss.0606155
• 引用： 邓依霖，徐舒文，高昕．基于深度学习模型的正负向文本情感分类［J］．社会科学进展，2024，6（6）：1397-1405.