TLN-NET: Triple Layer Norm for Aspect-based Sentiment Analysis

Aspect-based Sentiment Analysis (ABSA) is a specialized form of sentiment analysis that zeros in on pinpointing and harvesting sentiment details pertinent to distinct facets within textual content. In the field of sentiment analysis, traditional neural network models often face several challenges, including insufficient accuracy, low efficiency and poor generalization ability. Therefore, we propose a model called TLN-NET, this approach utilizes an adapted Sentic Net graph convolutional network and is designed to carry out sentiment analysis at the aspect level.First, we design a triple GLN (Graph Layer Normalization) architecture, which adds layer normalization technique to each GCN layer to maintain the stability of the gradient, thus improving the training efficiency of the model. We use the residual structure to combine the encoding output with the pooling output to enhance the expression ability of the model. Finally, we used Multi-layer Perceptron (MLP) and Softmax function to classify the final output, which improved the generalization ability of the model. We use the improved TLN-NET model to test on multiple datasets. The experimental results show that our proposed model has higher accuracy in judging aspect words, which is better than some state-of-the-art methods.