基于语用交互的跨目标立场检测
Cross-target stance detection based on pragmatic interaction
任科兰 1张明书 1魏彬 1姜文 1闫法成1
作者信息
- 1. 武警工程大学密码工程学院,陕西西安 710086;武警工程大学网络与信息安全武警部队重点实验室,陕西西安 710086
- 折叠
摘要
针对缺乏足够的带标注意见数据、跨目标立场检测结果不佳且可解释性弱等问题,提出一种基于语用交互(prag-matic interaction graph convolution,PIGCN)的跨目标立场检测模型.考虑情感与立场在语义上的耦合关系,利用交互式图卷积神经网络(graphical convolutional network,GCN),增量式聚合单词在不同目标之间语用信息的相互作用,缓解目标间的信息孤岛问题.实验结果表明,该模型在平均F1值上达到了 53.4％,优于基准模型,具有更好的可扩展性和适应性,在提升模型可解释性方面具有潜力.
Abstract
To address the issues of insufficient labeled opinion data,poor performance,and weak interpretability in cross-target stance detection,a solution was proposed.A cross-target stance detection model,known as pragmatic interaction graph convolu-tion(PIGCN),was introduced.The semantic coupling between emotions and stances was taken into consideration and the graphical convolutional networks(GCN)was utilized in an interactive manner.By incrementally aggregating the pragmatic infor-mation of words between different targets,the problem of information isolation among targets was mitigated.Experimental re-sults demonstrate that the PIGCN model achieves an average F1 score of 53.4％,outperforming the baseline model.It also ex-hibits enhanced scalability and adaptability.The model shows potential in improving interpretability.
关键词
跨目标立场检测/图卷积神经网络/语用交互/词级粒度/情感词汇/可解释性/依存图Key words
cross-target stance detection/graph convolutional neural network/pragmatic interaction/word-level granularity/emotion lexicon/interpretability/dependency graph引用本文复制引用
基金项目
国家社会科学基金项目(20BXW101)
国家社会科学基金项目(18XXW015)
出版年
2024
NETL
NSTL
万方数据