GNNTA-BiLSTM: Transformer-attention mechanism bidirectional short-term memory network-based graph neural network for English text attribute extraction

Attribute extraction is one of the sub-tasks of fine-grained sentiment analysis, its goal is to extract the attributes evaluated by users from the English comment text. The attribute classification accuracy of fine-grained sentiment analysis based on traditional methods is low. Therefore, this paper proposes a novel English text attribute extraction model based on Transformer-attention mechanism bidirectional short-term memory network (BERTAtt-BiLSTM) and graph neural network. Firstly, the model constructs the object information extraction layer based on syntactic dependency in graph neural network, and extracts attribute-view pairs. Secondly, in the word embedding layer, Transformer (BERT) module is used to implement the pre-training of word vector combined with contextual dynamic features. Then, in the feature extraction layer, attention mechanism bidirectional shortterm memory network module is integrated to reduce the dimension of feature space. Finally, in the classification layer, the attribute category of the genre-view pair is output through the activation function. Experiments are conducted on the laptop and restaurant domain datasets provided by the International Semantic Evaluation Congress in 2014 and 2016. Compared with the baseline model, the F1 value of the two English datasets improves by 2.33% and 1.44%, respectively, and the overall performance is higher than that of the current cutting-edge techniques.