基于“DiaASQ+RoBERTa”的对话式股吧评论方面级情感分析  

Aspect-Based Sentiment Analysis of Conversational Stock Forum Comments Using “DiaASQ+RoBERTa”

作　　者：段一奇 吴江[1] 程征 Duan Yiqi;Wu Jiang;Cheng Zheng(School of Information Management,Wuhan University,Hubei Wuhan 430072;AVIC Securities Co.,Ltd,Guangdong Shenzhen 518057;School of Information Management,Central China Normal University,Hubei Wuhan 430079)

机构地区：[1]武汉大学信息管理学院,湖北武汉430072 [2]中航证券有限公司,广东深圳518057 [3]华中师范大学信息管理学院,湖北武汉430079

出　　处：《情报理论与实践》2025年第4期163-173,共11页Information Studies:Theory & Application

基　　金：国家自然科学基金面上项目“数据融合视角下的区间型深度学习预测技术及其应用研究”的成果,项目编号:72171183。

摘　　要：[目的/意义]针对以往金融评论情感分析研究中多数研究局限于单一评论文本为研究对象,方面级金融情感分析中提取情感元素不够丰富等问题,提出了一个新的对话式股吧评论方面级情感分析任务。[方法/过程]基于该任务,构建了一个两阶段式的对话式股吧评论方面情感四元组提取框架,利用“DiaASQ+RoBERTa”方法分阶段提取对话式股吧评论中的方面类别、方面术语、观点术语及情感极性,经过情感元素的组合映射得到方面情感四元组,然后以东方财富股吧为数据来源构建对话式股吧评论数据集进行实证研究。[结果/结论]实验结果表明,与同一数据集上的最佳基线模型相比,“DiaASQ+RoBERTa”方法在F1值方面提升了约16.18%,显著优于基线方法。该研究不仅拓宽了金融评论情感分析领域的研究范围,还进一步完善了方面级金融情感元素提取的丰富度,为投资决策提供更多参考依据。[Purpose/significance]In previous studies on financial sentiment analysis,most research has been limited to analyzing individual comment texts,and the extraction of sentiment elements in aspect-based financial sentiment analysis has often been insufficiently comprehensive.To address these issues,we propose a new task of aspect-based sentiment analysis of conversational stock forum comments.[Method/process]Based on this task,we construct a two-stage framework for aspect sentiment quadruple prediction in conversational stock forum comments.The“DiaASQ+RoBERTa”method was applied to sequentially extract aspect categories,aspect terms,opinion terms,and sentiment polarities from the comments.Aspect sentiment quadruples were then obtained through a combination and mapping of these sentiment elements.An empirical study was conducted on a dataset of conversational stock forum comments derived from the Eastmoney Stock Forum.[Result/conclusion]Experimental results demonstrate that the“DiaASQ+RoBERTa”method achieved approximately a 16.18%improvement in F1 score over the best baseline model on the same dataset,highlighting its superior performance.This research not only expands the scope of sentiment analysis in the financial domain but also enhances the richness of aspect-based sentiment element extraction,providing more comprehensive references for investment decision-making.

关 键 词：方面级情感分析 DiaASQ RoBERTa 对话式股吧评论 金融评论 

分 类 号：G63[文化科学—教育学]