Research And Implementation Of Aspect Level Emotional Analysis Based On Deep Learning

In recent years,the development of social media platforms allows people to freely express their opinions on events,products,and more.These comments contain subjective emotions,and companies can analyze the views found in this data to understand market and social evaluations,thereby enhancing the competitiveness of their products.Aspect-level sentiment analysis is a fine-grained sentiment analysis method that can evaluate the emotional state of objects described from multiple dimensions.This paper mainly focuses on the aspect sentiment classification task in aspect-level sentiment analysis.The thesis proposes a PADGCN model based on graph convolutional networks and attention mechanisms.This model separately models the context vectors and aspect term vectors,and strengthens the semantic and syntactic relationships between aspect terms and their contexts through graph convolutional networks.By performing position encoding on the context vectors and combining the positional relationship between aspect terms and other words,their representation becomes more accurate.Finally,through the attention mechanism,the aspect term vectors and context vectors are combined,outputting the final aspect term classification results.The paper focuses on aspect sentiment classification as the main task,with aspect term extraction as an auxiliary task,and proposes the JATP multi-task model.The model constructs the aspect term extraction module using a BERT+Bi-GRU+Attention+CRF architecture,while the aspect sentiment classification module is modeled through graph convolutional networks and attention mechanisms.The two modules share the same input vectors and enhance features through a shared vector module,thereby integrating the features of both tasks and further improving the model’s generalization ability in aspect sentiment classification tasks.PADGCN and JATP models were tested in comparative experiments and ablation experiments on three public English datasets,and the experimental results proves the effectiveness of the two models in aspect sentiment classification tasks.Based on the implemented multi-task aspect sentiment analysis model JATP,the thesis designs and develops an aspect sentiment analysis system mainly to solve aspect term extraction and aspect sentiment classification.The system can analyze user input comment text and return related aspect terms and sentiment polarity to users,demonstrating a good classification effect.