融合情感增强与句法特征的卷烟消费者评价方面级情感分析

Aspect-level Emotion Analysis of Cigarette Consumer Evaluation Based on Emotional Enhancement and Syntactic Fcatures

为了解消费者对卷烟产品不同属性的情感信息,帮助烟草企业了解消费者评价及情感倾向,指导产品开发和市场营销决策,该文利用爬虫采集2010—2022年共18205条卷烟消费者评价数据,基于预训练模型(BERT)和双向长短时记忆网络(Bi-LSTM)对文本进行特征提取,结合句法依赖树获取语义间关系,建立融合情感增强和句法特征的方面级情感分类模型BAGCN,将BAGCN模型的分类结果与其他4种方法进行对比。结果显示,BAGCN模型在方面级情感分类效果最优,准确率和F1值达到79.49%和75.26%,BAGCN的各模块对最终的分类效果均有贡献。通过方面级情感分析发现,消费者更关注卷烟产品的价格和口感属性,对价格方面的评价多为消极情感,而口感方面的情感分布较为均衡,消费者对外观和品控方面的评价相对积极。

In order to understand consumers'emotional information about different attributes of cigarette products,help tobacco enterprises understand consumer evaluation and emotional tendency,and guide product development and marketing decisions,this paper uses crawlers to collect a total of 18205 cigarette consumer evaluation data from 2010 to 2022,extracts the features of the text based on a pre-training model(BERT)and Bidirectional Long Short-Term Memory(Bi-LSTM)network,and combines syntactic dependency tree to obtain semantic relations.An aspect-level emotion classification model BAGCN which combines affective en-hancement and syntactic features is established,and the classification results of BAGCN model are compared with the other four methods.The results show that BAGCN model has the best effect in aspect-level emotion classification,and the accuracy and F1 value reach 79.49%and 75.26%BAGCN.Through the aspect-level emotion analysis,it is found that consumers pay more atten-tion to the price and taste attributes of cigarette products,and the evaluation of price is mostly negative emotion,while the emo-tional distribution of taste is more balanced.Consumers'evaluation on appearance and quality control is relatively positive.