Aspect-level Sentiment Analysis Based On Attention Mechanism

Aspect-level sentiment analysis task,as a fine-grained sentiment analysis task,has been widely used in fields such as e-commerce and social media in recent years.Discovering user preferences and needs by analyzing these textual information has also attracted more and more attention from natural language processing researchers.The rapid development of deep learning technology has gradually replaced the traditional machine learning method in natural language processing tasks and has become the mainstream method for downstream tasks such as sentiment analysis.In particular,the development of attention mechanisms enables models to focus on contextual information related to aspect words.However,most of the existing attention models usually only consider a single aspect of attention information,and cannot effectively utilize all contextual information,and the attention mechanism cannot capture the position information of words.In addition,distributed word vectors usually do not contain sentiment information,which leads to a decrease in the prediction accuracy of the model.Aiming at the problems existing in the existing sentiment analysis model in natural language understanding,this paper studies the aspect sentiment analysis method around the attention mechanism,and tries to improve it in two aspects.The details are as follows:(1)A sentiment analysis model for interactive attention is designed.In the attention coding layer of the model,the attention of the context to the aspect word and the relationship within the context are extracted,and the emotional feature that combines the dependency between the aspect word and the context and the internal dependence of the word is obtained through the feature interaction layer.We conduct comparative experiments with existing models on two public datasets of sentiment analysis using two evaluation metrics,accuracy and macro-average F1,and prove that the interactive attention in this paper can make the model obtain more accurate prediction results..(2)Design an aspect sentiment analysis model that combines sentiment dictionary and attention mechanism.In this model,we combined the attention network with the sentiment dictionary to construct word vectors containing sentiment information,and compared the prediction results with the model based on word vectors,and verified that the construction method of word vectors incorporating sentiment information has Helps improve the prediction accuracy of the model.