Aspect sentiment analysis based on gating convolutional network and attention weighting mechanism

Abstract

Aspects Sentiment analysis is a fine-grained text on emotional classification. Aiming at the problem that traditional attention mechanism can't effectively combine contextual meaning an spectoward with information, and single level attention can't obtain deep emotional information features, a gated convolutional network model with attention weights is proposed. Firstly, the word layer is modeled by two-way long-term and short-term memory network, and context semantic information is captured in different directions. In the meantime, different weights are assigned to context words with different positions, and then sentences are gated by convolutional network. The layers are modeled to capture the importance of different sentences, and finally the softmax regression is used for classification. The laboratory finding on the Restaurant DS and the Laptop DS in SemEval2014 indicate that the classification accuracy is better than the classification effect of GCN.