BRCNN与语义信息结合的跨领域方面词抽取  

Cross-domain Aspect Word Extraction Combining BRCNN and Semantic Information

作　　者：王登雄 李卫疆[2] WANG Dengxiong;LI Weijiang(Faculty of Information Engineering and Automation,Kunming University of Science and Technology,Kunming 650500,China;Yunnan Key Laboratory of Artificial Intelligence,Kunming University of Science and Technology,Kunming 650500,China)

机构地区：[1]昆明理工大学信息工程与自动化学院,昆明650500 [2]昆明理工大学云南省人工智能重点实验室,昆明650500

出　　处：《小型微型计算机系统》2024年第12期2936-2943,共8页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(62066022)资助。

摘　　要：方面词抽取是方面级情感分析的关键步骤.当训练数据和测试数据来自同一领域时,用于该任务的现有方法已经可以得到令人满意的结果.然而,当训练数据与测试数据分别来源于不同领域时,这些方法呈现出的效果就急剧下降.为了解决这一缺乏可扩展性和鲁棒性的问题,本文提出了一种新的BRCNN方法结合语义信息来弥合源域与目标域的差距.该方法利用不同领域之间的语义相似性作为枢轴信息,从而降低了源域与目标域之间的差异性实现了方面词的跨领域抽取.同时,本文探究了BRCNN模型分别使用句法信息,语义信息,句法和语义信息相结合的知识结构作为枢轴信息弥合源域与目标域差距的性能比较,最终在基准数据集上展现出了比较好的性能.Aspect term extraction is a key step in aspect-level sentiment analysis.Existing methods for this task have achieved satisfactory results when the training data and test data come from the same domain.However,when the training data and test data come from different domains,the performance proposed by these methods starts to decline.To address this lack of scalability and robustness,this paper proposes a new BRCNN method that combines semantic information to bridge the source and target domains.This method utilizes the semantic similarity between different domains as the pivot information.thereby reducing the difference between the source domain and the target domain,and realizing the cross-domain extraction of aspect term.At the same time,the paper discusses the performance comparison of the BRCNN model using syntactic information,semantic information,and a combination of syntactic and semantic information as the pivot information to bridge the gap between the source domain and the target domain,and finally performed better on the benchmark dataset.performance.

关 键 词：方面词抽取 领域适应 卷积神经网络 语义相似性 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]