Fine-grained Sentiment Analysis Based On Deep Memory Networks

Fine-grained sentiment analysis is a scientific issue closely watched by academics and industry in recent years.It has been widely used in many fields such as e-commerce commentary,social network,and public opinion analysis.Fine-grained sentiment analysis can predict the emotional polarity of a particular object in a sentence,and explore the user's emotional inclinations,and thus has received widespread attention.The existing methods are mainly based on deep learning to construct a neural network model,which often requires a large number of manual annotation data,and faces the poor accuracy of emotional prediction.In view of the defects and deficiencies of previous methods,this paper focuses on the core issues of fine-grained semantic analysis.Firstly,combining the traditional feature engineering and deep representation learning,a fine-grained sentiment analysis model of composite memory network based on semantic features is proposed.Three feature compositing strategies are proposed to integrate the representation of word vector and semantic features in deep memory networks.The model achieves high accuracy over the previous methods on the SemEvel2014 datasets.Secondly,combined with the latest development of attention mechanism,a composite fine-grained sentiment analysis model for domain transfer learning is proposed.Utilizing the characteristics of semantic features,it implements model sharing across different fields.Experiment results show that the improved model has better migration capabilities and has achieved good performance in two sentiment analysis tasks in the SemEvel2014,SemEvel2015 and SemEvel2016 data sets.In order to verify the effect of our model,the experiment was carried out on the test datasets from the actual e-commerce comment,and the effect of the attention mechanism in the model was deeply analyzed.Investigate the importance of the attention mechanism to the problem of fine-grained sentiment analysis,and conduct detailed controlled trials and specific case studies for specific models.Compared with the previous model,the composite memory network can focus more attention on the key emotional words in the context.