基于有效距离和注意力机制的方面级情感分析  

Aspect-level Sentiment Analysis Based on Effective Distance and Attention Mechanism

作　　者：刘跃 张琨[1] 朱浩华 江浩俊 方自正 LIU Yue;ZHANG Kun;ZHU Haohua;JIANG Haojun;FANG Zizheng(Nanjing University of Science and Technology,Nanjing 210094)

机构地区：[1]南京理工大学,南京210094

出　　处：《计算机与数字工程》2025年第3期637-642,707,共7页Computer & Digital Engineering

摘　　要：目前基于位置的方面级情感分析主要是根据方面词和其他单词的语义位置来进行研究,然而该方法容易导致单词权重分配不合理,降低了情感分析的准确率。因此,通过对句子的语法方面进行研究,提出了一种新的情感分析模型ED-BERT-BIGRU(EDBB),该模型首先将三种文本分别输入到BERT模型中获得对应的词向量,利用双向GRU捕获上下文表示,然后利用依赖语法树中每个词与方面词的相对位置,结合计算有效距离CED算法来获得每个单词与方面词的有效距离,同时结合局部文本动态权重赋值LDAW算法和自注意力机制为每个单词赋予不同的权重,从而提高情感分析的准确率。与其他深度神经网络模型相比,该模型在SemEval-2014的Restaurant和Laptop这两个公开数据集上取得了更高的Accuracy和F1值,模型训练结果更好。At present,location-based aspect sentiment analysis mainly studies aspect words and other words based on semantic position,but this method is prone to unreasonable word weight allocation,which reduces the accuracy of sentiment analysis.Therefore,a new emotion analysis model,ED-BERT-BIGRU(EDBB),is proposed by studying the grammatical aspects of sentences.The model firstly inputs three kinds of texts into BERT model to obtain corresponding word vectors,and uses bidirectional GRU to capture context representation.Then it uses rely on syntax tree in each word and the word of the relative position,combining with the calculating CED algorithm to obtain the effective distance each word and the word of the effective distance,at the same time,combined with local text weight LDAW dynamic assignment algorithm and the attention mechanism for each word gives different weights,so as to improve the accuracy of sentiment analysis.Compared with other deep neural network models,this model achieves higher Accuracy and F1 values on the Restaurant and Laptop public data sets of Semeval-2014,and the training results of the model are better.

关 键 词：相对位置 有效距离 CED算法 LDAW算法 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]