Research On Aspect-Based User Reviews Sentiment Analysis Based On Deep Learning

In recent years,with the rapid development of social media and e-commerce platforms,a large number of comments which have rich emotional information are generated every day.If we can mine emotional tendency based on these comments,it will have great influence on both platforms and users.The field of text sentiment analysis is mainly divided into document level,sentence level and aspect level.Aspect-based sentiment analysis is more applicable to the characteristics of comments.Therefore,the thesis focuses on the extraction of aspect terms in comments and the discrimination of sentiment polarity for given aspect terms.The main work in this thesis is as follows:(1)In the area of aspect-based sentiment analysis,the identification of aspect terms in comments is first required.Therefore,the thesis is based on a BERT-Bi LSTM-CRF model to achieve the extraction of aspect terms from comments.For the aspect recognition task,the model adopts BERT for semantic representation and uses Bi LSTM to learn contextual information features,and then learns the dependencies between neighbouring tags through the CRF layer to obtain the best annotation sequence to achieve the extraction of aspect terms.Meanwhile,the ablation experiments of this model are designed,and the results demonstrate the effectiveness of learning contextual features by Bi LSTM through BERT for semantic representation.Finally,the model is experimentally compared with the baseline model on datasets,and the results demonstrate the effectiveness of the BERT-Bi LSTM-CRF-based model in the aspect recognition task.(2)This thesis proposes a Bi Att-GCN model incorporating auxiliary information(AI-Bi AttGCN)for sentiment analysis of a given aspect term.The model incorporates contextual features with part-of-speech tag information as well as positional encoding information in the aspect representation features,and computes a new aspect feature vector through an attention mechanism,and then constructs a graph structure of aspect features using the relative position of the aspect terms in the comments as the input to a graph convolutional network to finally make sentiment polarity discriminations for the aspect terms.AI-Bi Att-GCN model and baseline model are compared on datasets.The results show that this method can effectively deal with aspect term emotion analysis tasks.Then,ablation experiments were designed based on the AI-Bi Att-GCN model to show the importance of part-of-speech tag information,position information and graph convolutional network to the model,and to explore the effects of the number of GCN layers and the number of aspect terms in the comments on the model performance.(3)Based on the implemented aspect term recognition model and aspect sentiment analysis model,the thesis completes the development and implementation of a prototype aspect-based user comment sentiment analysis system with the help of a common web development framework,and details the system requirements,system architecture and the design and implementation of the core functional modules.The system requirements,system architecture and core functional modules are elaborated and implemented.Finally,relevant web pages are displayed to verify the usability of the system.