在数字化信息高速传播的当下,实现对新闻评论精准且具备可解释性的情感分析,对舆情洞察与舆论引导极为关键。本研究运用RoBERTa模型,结合多尺度卷积神经网络(MSCNN),深度剖析新闻评论中的情感倾向。同时,采用局部可解释的模型无关解释方...在数字化信息高速传播的当下,实现对新闻评论精准且具备可解释性的情感分析,对舆情洞察与舆论引导极为关键。本研究运用RoBERTa模型,结合多尺度卷积神经网络(MSCNN),深度剖析新闻评论中的情感倾向。同时,采用局部可解释的模型无关解释方法(LIME),实现对模型预测结果的深度解释,可视化展示模型在处理新闻评论时对词汇和短语的关注重点,为模型决策提供清晰依据。实验结果表明,RoBERTa-MSCNN在新闻评论情感分析任务上取得了更优的性能,准确率达到83.34%,精确率为82.6%,召回率为84.67%,F1值提升至83.62%。同时,可解释性分析为用户理解模型输出提供了清晰的视角,有助于新闻媒体更有效地进行舆情监测与引导,为相关领域的研究与应用提供了有力支持。In the current era of high-speed digital information dissemination, accurate and interpretable emotional analysis of news comments is crucial for public opinion insight and guidance. In this study, RoBERTa model and multi-scale convolutional neural network (MSCNN) are used to analyze the emotional tendency of news commentary. At the same time, Local Interpretable Model-agnostic Explanations (LIME) is used to realize the in-depth interpretation of the model prediction results and visually display the model’s focus on words and phrases when processing news comments, providing a clear basis for the model’s decision-making. The experimental results show that RoBERTa-MSCNN has achieved superior performance in the task of sentiment analysis of news comments. Its accuracy rate reaches 83.34%, the precision rate is 82.6%, the recall rate is 84.67%, and the F1 score has been increased to 83.62%. At the same time, interpretability analysis provides a clear perspective for users to understand the model output, helps news media to monitor and guide public opinion more effectively, and provides strong support for research and application in related fields.展开更多
随着网络信息的爆炸式增长,威胁情报分析作为军事情报分析与战略决策的重要组成部分,其面临着来源多样化和信息结构复杂化的挑战。传统的人工信息提取方法在处理这些大量结构化及非结构化信息时效率低下,准确性有限。文中针对这一挑战,...随着网络信息的爆炸式增长,威胁情报分析作为军事情报分析与战略决策的重要组成部分,其面临着来源多样化和信息结构复杂化的挑战。传统的人工信息提取方法在处理这些大量结构化及非结构化信息时效率低下,准确性有限。文中针对这一挑战,提出了一种结合RoBERTa、BiLSTM和条件随机场(Conditional Random Fields,CRF)的命名实体识别新算法。此算法通过Ro-BERTa模型深入挖掘文本的语义特征,BiLSTM模型捕捉序列上下文信息,CRF层用于精确的实体标记,从而有效提升信息提取的准确率和效率。本文基于开源情报语料库构建了一个涉及导弹发射事件的命名实体识别数据集,并在此基础上进行了实验,结果表明,该方法在精确率、召回率及F1值等关键指标上相较于主流深度学习方法表现出显著的性能提升,其中F1值高达94.21%。展开更多
文摘在数字化信息高速传播的当下,实现对新闻评论精准且具备可解释性的情感分析,对舆情洞察与舆论引导极为关键。本研究运用RoBERTa模型,结合多尺度卷积神经网络(MSCNN),深度剖析新闻评论中的情感倾向。同时,采用局部可解释的模型无关解释方法(LIME),实现对模型预测结果的深度解释,可视化展示模型在处理新闻评论时对词汇和短语的关注重点,为模型决策提供清晰依据。实验结果表明,RoBERTa-MSCNN在新闻评论情感分析任务上取得了更优的性能,准确率达到83.34%,精确率为82.6%,召回率为84.67%,F1值提升至83.62%。同时,可解释性分析为用户理解模型输出提供了清晰的视角,有助于新闻媒体更有效地进行舆情监测与引导,为相关领域的研究与应用提供了有力支持。In the current era of high-speed digital information dissemination, accurate and interpretable emotional analysis of news comments is crucial for public opinion insight and guidance. In this study, RoBERTa model and multi-scale convolutional neural network (MSCNN) are used to analyze the emotional tendency of news commentary. At the same time, Local Interpretable Model-agnostic Explanations (LIME) is used to realize the in-depth interpretation of the model prediction results and visually display the model’s focus on words and phrases when processing news comments, providing a clear basis for the model’s decision-making. The experimental results show that RoBERTa-MSCNN has achieved superior performance in the task of sentiment analysis of news comments. Its accuracy rate reaches 83.34%, the precision rate is 82.6%, the recall rate is 84.67%, and the F1 score has been increased to 83.62%. At the same time, interpretability analysis provides a clear perspective for users to understand the model output, helps news media to monitor and guide public opinion more effectively, and provides strong support for research and application in related fields.
文摘随着网络信息的爆炸式增长,威胁情报分析作为军事情报分析与战略决策的重要组成部分,其面临着来源多样化和信息结构复杂化的挑战。传统的人工信息提取方法在处理这些大量结构化及非结构化信息时效率低下,准确性有限。文中针对这一挑战,提出了一种结合RoBERTa、BiLSTM和条件随机场(Conditional Random Fields,CRF)的命名实体识别新算法。此算法通过Ro-BERTa模型深入挖掘文本的语义特征,BiLSTM模型捕捉序列上下文信息,CRF层用于精确的实体标记,从而有效提升信息提取的准确率和效率。本文基于开源情报语料库构建了一个涉及导弹发射事件的命名实体识别数据集,并在此基础上进行了实验,结果表明,该方法在精确率、召回率及F1值等关键指标上相较于主流深度学习方法表现出显著的性能提升,其中F1值高达94.21%。
