CNNs and RNNs in aspect-level sentiment analysis and comparison

Abstract

Aspect-based sentiment analysis (ABSA), also known as fine-grained sentiment analysis, may offer a precise polarity for each aspect in statement aspect. CNN and RNN neural network models, the most fundamental and widely used deep neural network models, have been created by researchers using a variety of methodologies to offer reliable findings in ABSA. This paper will sort out the development process of CNN and RNN models in completing various tasks in aspect-level sentiment analysis and find out the current SOTA model. In addition, by comparing the representatives of the two models in the same environment, the advantages and disadvantages of the two models are analysed to provide guidance for choosing different baseline models. The results show that the CNN-based model has the advantages of high accuracy and fast training speed, but the design is more complicated, and the pooling layer is prone to lose position or sequence features. The RNN-LSTM based model is easier to capture sequence data, has high reliability, and has a relatively simple design structure. However, it requires significantly longer training time, the model is complex, and the computational cost is high. Considering the characteristics of the two most basic neural models, new tasks require researchers to propose better hybrid neural network models.