JAES-MGCN:金融实体与情感极性联合获取方法  

JAES-MGCN:A Joint Acquirement Method of Financial Entity and Sentiment Polarity

作　　者：李旸 任鹏飞 王素格[3,4] 符玉杰 LI Yang;REN Pengfei;WANG Suge;FU Yujie(School of Finance,Shanxi University of Finance and Economics,Taiyuan,Shanxi 030006,China;State Grid Shanxi Electric Power Company Taiyuan Power Supply Company,Taiyuan,Shanxi 030001,China;School of Computer&Information Technology,Shanxi University,Taiyuan,Shanxi 030006,China;Key Laboratory of Computational Intelligence and Chinese Information Processing of Ministry of Education,Shanxi University,Taiyuan,Shanxi 030006,China;School of Computer Science and Technology,Taiyuan University of Science and Technology,Taiyuan,Shanxi 030024,China)

机构地区：[1]山西财经大学金融学院,山西太原030006 [2]国网山西省电力公司太原供电公司,山西太原030001 [3]山西大学计算机与信息技术学院,山西太原030006 [4]山西大学计算智能与中文信息处理教育部重点实验室,山西太原030006 [5]太原科技大学计算机科学与技术学院,山西太原030024

出　　处：《中文信息学报》2024年第11期135-145,共11页Journal of Chinese Information Processing

基　　金：国家自然科学基金(62106130,62376143,62076158,62072294);山西省基础研究计划项目(20210302124084);山西省高等学校科技创新项目(2021L284);山西省重点实验室开放课题基金(CICIP2023006)。

摘　　要：针对金融领域文本中具有实体较多、实体长度较长以及实体间存在语义关联的特性,容易导致实体和情感极性联合获取对应错误的问题,该文提出了一种基于多图卷积网络的金融实体和情感极性联合获取方法(JAES-MGCN)。该方法利用预训练模型对句子进行初始表示,构建基于多头自注意力机制的句子权重矩阵,建立基于多个图卷积网络融合的实体边界深层语义表示。在此基础上,基于多头注意力机制,建立实体与句子之间的交互信息表示,最后在解码层实现<金融实体,情感极性>二元组联合获取。在金融实体和情感极性数据集CES-data上,与已有的模型进行对比,该文所提方法在精确率和F1值上分别提升了3.66和1.42,验证了图卷积网络有利于捕获金融实体间的语义关系,进一步在公开的英文Twitter方面项情感分析数据集上验证了该方法的有效性。Due to the large number of entities,long length of entities and complex semantic relationships between entities in the financial field,it is easy to lead to wrong pairs in extracted of entities and sentiment polarity.In this paper,we propose a joint acquirement method for financial entity and sentiment polarity based on multi-graph convolutional networks(JAES-MGCN).The pre-trained model is used for the initial representation of sentences,the sentence weight matrix based on multi-head self-attention mechanism is constructed,and the entity boundary representation based on the fusion of multi-graph convolutional networks is established.On this basis,interactive information representation between entities and sentences is established based on multi-head attention mechanism,and binary groups<financial entities,emotional polarity>is acquired in decoding layer.Experiments demonstrate that the proposed method improves the precision and F1 value by 3.66 and 1.42 respectively compared to the baseline.

关 键 词：金融实体 情感极性 图卷积网络 交互信息表示 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]