| 摘 要: 摘要：现有方法在处理社交软件文本情感分析的过程中往往会出现一些问题，在捕捉其口语化、非正式表达及强上下文依赖特性时常面临挑战。为克服传统模型在语义理解深度与关键信息聚焦上的不足，本文针对抖音用户评论文本，提出并构建了一种BERT-BiLSTM-Attention (BERT-BiLSTM-Att)深度学习模型，旨在高效、准确地实现评论文本的情感倾向（积极、中立、消极）分类。该模型集合了BERT预训练模型的强大语义表征能力、BiLSTM捕捉上下文依赖关系的优势，以及聚焦关键信息的注意力机制(Attention Mechanism)。在包含超过46万条评论的数据集上，所提模型展现出了优异的性能，验证集准确率达到0.924，远高于如CNN、RNN和朴素贝叶斯等传统算法，在社交媒体的短文情感分析中充分体现了该模型的优越性。 |
| 关键词: 情感分析 BERT-BiLSTM-Att模型 文本挖掘 注意力机制 |
|
中图分类号:
文献标识码:
|
|
| Sentiment Analysis of TikTok User Reviews Utilizing a BERT-BiLSTM-Attention Model |
|
Guan Shuo
|
Guangdong Neusoft University
|
| Abstract: Abstract: Existing methods often encounter certain issues when dealing with the sentiment analysis of social software text, frequently facing challenges in capturing its colloquial, informal expressions and strong context-dependent characteristics. To overcome the shortcomings of traditional models in the depth of semantic understanding and the focus on key information, this paper proposes and constructs a BERT-BiLSTM-Attention (BERT-BiLSTM-Att) deep learning model for TikTok user comment text, aiming to efficiently and accurately classify the sentiment tendency (positive, neutral, negative) of the comment text. This model integrates the powerful semantic representation ability of the BERT pre-training model, the advantage of BiLSTM in capturing context-dependent relationships, and the Attention Mechanism for focusing on key information. On a dataset containing over 460,000 comments, the proposed model demonstrated outstanding performance, with a validation set accuracy rate of 0.924, far exceeding traditional algorithms such as CNN, RNN, and Naive Bayes, fully demonstrating the superiority of this model in the sentiment analysis of short texts on social media. |
| Keywords: Sentiment Analysis BERT-BiLSTM-Attention Model Text Mining Attention Mechanism |
