Research And Optimization Of Sentiment Classification Model Based On Specific Target

Specific target sentiment analysis is an important branch of natural language processing tasks.Most of traditional methods for solving such tasks are supervised methods based on machine learning and deep learning.In this thesis,the research work focuses on the emotional classification of specific goals,and its content is more detailed,resulting in limited training data size.The trained model is applicable to a single domain.The expression of network language is rich and varied simultaneously.It is more and more difficult for traditional methods to solve this task.Therefore,how to optimize and improve the accuracy of the classification model is a vital research task.According to the characteristics of the sentiment analysis task in this thesis,combined with the deep learning algorithms widely used in such fields.Therefore,this thesis mainly studies the combination of recurrent neural network algorithms.The main works of this thesis are as follows:1)Aiming at the problem of inaccurate recognition of emotional words in this thesis.Based on recurrent neural network,the model based on attention matrix is supplemented,optimizing attention mechanism and filling corpus.In this model,the first filling corpus is completed by manual annotation.By expanding the corpus,the Glove method is used to learn the word vector.Secondly,the word vector is used as the network input to learn the representation of the sentence.The experimental results show that the model achieves excellent accuracy under the corpus rich scene.2)Aiming at the problem of the insufficiency of feature extraction,the sentence representation is inaccurate.The model based on hierarchical feature extraction is proposed,which is based on the weight of the mark standard and introduces the attention function.The model flexibly trains weight information according to specific goals during the training process by adding an adjustable attention function.Experiments show that the model trains better attention weights in the case of less corpus,and the accuracy of classification is better than traditional research methods.3)Aiming at the problem of poor extendibility.The model based on knowledge map is proposed,using general knowledge base.The model fuses specific fact triples vectorization into the sentence vector,providing additional prior knowledge.The experimental results show that the model has achieved good performance.