Aspect-based Sentiment Analysis Model Based On Deep Learning

The goal of the aspect-based sentiment analysis task is to mine and analyze the user's sentiment on the aspect words mentioned in the comment text,but existing models usually use a method of pre-trained word vectors combined with neural network models.But word vectors can not solve the problem of polysemy,Which will affect the semantics of the text captured by the model.In addition,aspect-based sentiment analysis includes two subtasks: aspect extraction and aspect-based sentiment classification.However,most existing researches assume that the aspect words mentioned in the text have been given,and only focus on the sentiment classification on a given aspect word.The aspect words need to be marked manually or in other ways,which also limits the application of these studies in practice.Therefore,this thesis designs a sequence labeling model based on the pre-trained language model in deep learning technology,which is applied to the end-to-end aspect-based sentiment analysis task.While extracting the aspect words in the user text,it predicts the user's sentiment tendency on the corresponding aspect words.Main researches in this thesis include:(1)A sequence labeling model based on a pre-trained language model is designed.The pre-trained language model calculates the text feature representation of the input user comments,and then performs the sequence through different downstream model layers such as linear layer,recurrent neural network,and conditional random field.Annotations are used for end-to-end aspect-based sentiment analysis tasks.(2)Conduct experimental research on multiple aspect-based sentiment analysis datasets,compare the design model in this thesis with other existing models,and summarize and analyze the experimental results.(3)Design and implement an aspect-based sentiment analysis system based on the design model in this thesis,which shows the practicability of this design model.Experiments show that the model designed in this thesis is significantly better than other comparative models in the end-to-end aspect-based sentiment analysis task,which verifies the effectiveness of the model designed in this thesis.