面向汽车领域对象级知识增强情感分析模型研究
Research on Object-level Knowledge-enhanced Sentiment Analysis Model in the Automotive Field
骆仕杰 1韩抒真 1金日泽 2汪剑鸣 2李轩冰3
作者信息
- 1. 天津工业大学网络安全和信息化办公室,天津 300387
- 2. 天津工业大学计算机科学与技术学院,天津 300387
- 3. 天津中医药大学图书馆,天津 301617
- 折叠
摘要
面对汽车评论中对各项指标情感分类分析的需求,本研究提出了两部分任务:一是识别汽车评价对象并抽取情感要素;二是进行基于情感知识增强的情感分类分析.本文借助点互信息方法(PMI),探究对象词语与情绪词语的联系,进一步运用文本情感要素分析方法,构建了一种基于情感知识增强的汽车评论对象级情感分析模型(OLSCA).该模型首先采用PMI方法确定汽车评价关键指标与情绪词语极性的关系,接着通过自动情感词语掩盖及情感对象预测分析,形成词语、词语极性、对象级情感关系三部分的预测目标,生成针对标记对象的情感分类结果.实验证明,OLSCA在汽车评价领域对短文本评论进行情感分类分析,相较于传统文本语义情感分析有更大实际价值,有助于根据用户评价意图,全面构建汽车综合评价体系.
Abstract
In response to the need for sentiment classification analysis across various indices in automobile reviews,this study intro-duces two tasks:identifying the objects of car evaluations and extracting emotional components,and conducting sentiment classification analysis augmented by emotional knowledge.Leveraging the Point Mutual Information(PMI)method,the study elucidates the relation-ship between object words and emotional words,further utilizing textual emotional component analysis to establish an emotion-knowl-edge-enhanced Object-Level Sentiment Analysis Model(OLSCA).Initially,the model employs PMI to discern the relationship be-tween key evaluation indicators and the polarity of emotional words.Subsequently,it generates predictive targets for words,word polar-ity,and object-level emotional relations via automated emotional word masking and emotional object predictive analysis,yielding senti-ment classification outcomes for tagged objects.Experimental results demonstrate that OLSCA,when executing sentiment classification analysis on succinct reviews in the car evaluation domain,offers substantial practical utility compared to traditional text semantic senti-ment analysis,facilitating the comprehensive development of a car evaluation framework grounded in user evaluation intent.
关键词
PMI/对象级情感分析/情感知识增强/用户评价意图/汽车评价体系Key words
PMI/object-level sentiment analysis/emotion knowledge enhancement/user evaluation intent/vehicle evaluation system引用本文复制引用
出版年
2025
NETL
NSTL
万方数据