摘要
观点检索旨在从社交媒体文档集中找出与主题相关的观点文档,过程中包含了对主题—文档相关性特征及文档观点特征的提取.现有模型在相关性特征提取时,集中于语义概念扩展层面的研究,忽视了词语的权重问题,导致模型在语义层面的泛化能力受限,进而影响了观点检索结果的准确性.因此,该文在现有模型的基础上使用交互注意力机制提取文本中词语权重,计算出更为准确的相关性特征,将其应用于统一检索模型,实现观点检索.实验表明:与当前最好的模型相比,改进后的统一检索模型在两个Twitter公开数据集上,MAP主指标分别提升了1.5%和1.1%,很好验证了该文所提方法的有效性.
Opinion retrieval aims to find topic-related opinion documents from social media document sets,including the extraction of topic-document correlation features and document opinion features.The existing methods focus on the semantic concept extension,ignoring the weight of words,which leads to the limited generalization ability of the model at the semantic level,and then affects the accuracy of the opinion retrieval results.Therefore,on the basis of the existing models,this paper utilizes the interactive attention mechanism to extract the weight of words in the text,and calculates more accurate correlation features,which is applied to the unified retrieval model to realize the viewpoint retrieval.Experimental results show that compared with the current best model,the MAP main index of the improved unified retrieval model is increased by 1.5%and 1.1%respectively on two Twitter public datasets,which well verifies the effectiveness of the proposed method.
作者
张铭洲
ZHANG Ming-zhou(Department of Computer and Information Engineering,Li Ming Vocational University,Quanzhou 362000,China)
出处
《通化师范学院学报》
2023年第2期33-38,共6页
Journal of Tonghua Normal University
基金
福建省教育厅2022年中青年教师教育科研项目“基于深度表示学习的观点检索研究”(JAT220718)。
关键词
观点检索
注意力机制
深度表示学习
情感分析
opinion retrieval
attention mechanism
deep learning
sentiment analysis
