基于Albert和句法树的方面级情感分析  

Aspect-level sentiment analysis based on Albert and syntactic tree

作　　者：王跃跃 WANG Yueyue(Business School,University of Shanghai for Science and Technology,Shanghai 200093,China)

机构地区：[1]上海理工大学管理学院,上海200093

出　　处：《智能计算机与应用》2023年第4期52-59,共8页Intelligent Computer and Applications

摘　　要：方面级情感分析是情感分析的子任务,具体目标是识别不同方面词的情感极性。先前的工作大部分采用静态词向量和循环神经网络进行这个任务的建模。然而由于自然语言表达的多样性,静态词向量不能够准确地找到修饰方面词的上下文信息,并且以往的工作在对上下文的位置信息编码时存在不足。同时发现中性标签的数据表达的不确定性,本文认为会存在一定的标签不可靠情况。所以本文提出了基于预训练模型Albert和引入句法树的模型Albert-DP,并且在损失函数中加入了标签平滑。通过该设计,本模型能够很好地表示方面词对象及其上下文,有助于情感分类。本文在公开的笔记本电脑数据集、餐馆数据集以及推特数据集上的实验表明,本文的方法优于传统的模型。Aspect-level sentiment analysis is a subtask of sentiment analysis,and its goal is to identify the emotional polarity of different aspect-level words.Most of the previous work used static word vectors and recurrent neural networks to model this task.However,due to the diversity of natural language expression,static word vectors could not accurately find the context information of the modified aspect-word,and the previous work had shortcomings in encoding the location information of the context.At the same time,the paper researches the uncertainty of the neutral label data,and it is believed that there will be some label unreliability.Therefore,the paper proposes a model Albert-DP based on pre-training model Albert and syntactic tree,and adds label smoothing to the loss function.Through this design,the model can well represent the object of aspect word and its context,which is helpful for emotion classification.The experiments on publicly available laptop datasets,restaurant datasets,and Twitter datasets show that the proposed approach outperforms traditional models.

关 键 词：情感分析 静态词向量 句法树 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]