| 王思丽,杨恒,刘巍.基于混合机器学习的网络舆论情感识别方法研究[J].情报工程,2024,10(2):011-026 |
| 基于混合机器学习的网络舆论情感识别方法研究 |
| Method of Online Public Opinion Sentiment Recognition Based on Hybrid Machine Learning |
| |
| DOI：10.3772/j.issn.2095-915X.2024.02.002 |
| 中文关键词: 混合机器学习；深度学习；网络舆论；情感识别；特征嵌入 |
| 英文关键词: Hybrid Machine Learning; Deep Learning; Online Public Opinion; Sentiment Recognition; Feature Embedding |
| 基金项目:甘肃省哲学社会科学规划项目“基于大数据技术提升新闻媒体舆论监督能力研究”（2021YB158）；甘肃省自然科学基金项目“甘肃省医疗健康大数据资产管理模式与再利用机制研究”（23JRRA581）。 |
| 作者 | 单位 | | 王思丽 | 1. 中国科学院西北生态环境资源研究院文献情报中心 兰州 730000；2. 甘肃省知识计算与决策智能重点实验室 兰州 730000 | | 杨恒 | 1. 中国科学院西北生态环境资源研究院文献情报中心 兰州 730000；2. 甘肃省知识计算与决策智能重点实验室 兰州 730000 | | 刘巍 | 1. 中国科学院西北生态环境资源研究院文献情报中心 兰州 730000；2. 甘肃省知识计算与决策智能重点实验室 兰州 730000 |
|
| 摘要点击次数: 928 |
| 全文下载次数: 1068 |
| 中文摘要: |
| [目的/意义]提高网络舆论情感识别的效率和准确性，为决策者评估舆情倾向提供有效技术方法支持。[方法/过程]综合利用机器学习和深度学习的优点，将情感极性向量Sentimet_Embedding、预训练词向量Word2Vec等多通道特征嵌入方法，双向长短期记忆网络BLSTM、卷积神经网络CNN等深度神经网络模型，以及随机失活Dropout、批标准化BN等技术策略有机结合，构建了融合文本情感极性和预训练语义特征的基于混合机器学习的网络舆论情感识别模型，并通过收集社交媒体评论文本数据集对模型的可行性与有效性进行了验证。[局限]方法及模型性能尚未达到最优，未来仍有许多可以改进的空间。[结果/结论]研究结果表明，通过多通道特征嵌入方法及混合叠加神经网络模型能够显著提高网络舆论情感识别模型的性能；基于混合机器学习的网络舆论情感识别模型比传统机器学习或单一深度学习分类模型的识别精度高。 |
| 英文摘要: |
| [Objective/Significance] Improve the efficiency and accuracy of online public opinion sentiment recognition, and provide effective technical support for decision-makers to evaluate public opinion tendencies. [Methods/Processes] This study comprehensively utilizes the advantages of machine learning and deep learning, and organically combines multi-channel feature embedding methods such as Sentimet_Embedding and Word2Vec, deep neural network models such as BLSTM and CNN, as well as technical strategies such as random dropout and batch normalization, constructs a hybrid machine learning based online public opinion sentiment recognition model that integrates text sentiment polarity and pre trained semantic features. Finally, the feasibility and effectiveness of the model are verified by collecting social media public opinion data. [Limitations] The performance of the method and model has not yet reached its optimal level, and further improvements are needed in the future. [Results/Conclusions] Research results indicate that the performance of online public opinion sentiment recognition model can be significantly improved through multi-channel feature embedding methods and hybrid overlay neural networks. The online public opinion sentiment recognition model based on hybrid machine learning has higher accuracy than traditional machine learning classification models or single deep learning classification models. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
