Open AccessAspect Based Sentiment Analysis of catering field reviews via RoBERTa-AOA model
Author(s) -
Bo Li,
F Pan,
Zhaoyu Shou,
Huibing Zhang
Publication year - 2021
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1848/1/012064
Subject(s) - softmax function , sentence , computer science , sentiment analysis , field (mathematics) , artificial intelligence , natural language processing , artificial neural network , mathematics , pure mathematics
The emotional information included in catering field reviews has a great influence on the ordering behaviour of users. The traditional Aspect Sentiment Analysis Model based on Neural Network often has the problem of neglecting Aspect Words. This paper proposes an aspect sentiment analysis model that combines RoBERTa and Attention-over-Attention (AOA). Firstly, the model uses RoBERTa to model sentences and aspects respectively. Then, the deep semantic features of sentence and aspect were extracted by Bi-LSTM. Finally, AOA is adopted to learn the weight of aspect to sentence and sentence to aspect respectively, and the results are input into SoftMax layer for classified output. Experimental results on real Meituan catering data sets show that the accuracy of RoBERTa-AOA is about 1.7%-3.7% higher than that of other models.