Aspect-level Sentiment Analysis Based on Knowledge Transfer and Attention Fusion

Aspect-level sentiment analysis aims to judge the sentiment polarity when a comment involves multiple aspects. Existing methods usually use aspect-level datasets to train neural network models directly, but the labeled aspect level training datasets are small, which make the model unable to fully learn and limit the performance of the model. In order to solve the problem, this paper uses the idea of transfer learning to pre-train the sentiment analysis model with large amount of document-level datasets, so as to obtain more text semantic, syntactic information and emotional features. Then, through the objective function and attention fusion method designed in this paper, the attention weight in the document-level sentiment analysis model is integrated into aspect-level sentiment analysis, so that the performance of aspect level sentiment analysis can be improved by knowledge transfer and attention fusion. The model was tested on SemEval2014 dataset, and the accuracy and F1 value of the experimental results are higher than those of the comparison model, which proves the effectiveness of the model.