Research And Implementation Of Text Emotion Prediction Method Based On Graph Model

At present,most of the neural network models that are used to predict users' emotions at the statement level on text data adopt the method of converting a piece of text data into vector format as the input of neural network model,and the corresponding emotional category of the text data as the label.Although some neural network model in the form of Long-Short Term Memory or Gated Recurrent Unit connect words in different places in the text data to make the model have "memory",the connection between words will be weakened when the place differences are big,resulted in the information contained in the text data cannot be fully mined.Moreover,because there are some words with universality and fuzziness in the text data,the information of the text data can not be expressed in a very concrete and explicit way.Therefore,it is inevitable to make emotional prediction based.When collecting data sets,the collected text data often exists in semantic environment.If each piece of collected text data is put into it,the meaning of the text data will be more concrete and clear.However,most of the current text-based data sentiment prediction models do not take the semantic environment of the text into account.To solve the above problems,a graph model of text data that represents semantic associations.is designed in this thesis.There are two kinds of graph models designed in this paper,the semantic environment graph model built on the basis of text data set and the text graph model built on the basis of a single text data.Both models take words as the vertex and the association between words as the weight of the edge.These two graph models are input into the semantic injection model,and the word meanings in the text graph model are materialized by the semantic environment graph model.In the process of constructing the text graph model,different words in the text data will be directly correlated,and the weak correlation of text context will be effectively solved.In the process of combining text graph model with semantic environment graph model,text data will be put into its semantic environment,and the problem of missing text source environment can be effectively solved.Then,the graph model is used as the feature,and the emotional category of the graph model corresponding to the text data is used as the label to train the convolutional neural network model.After the completion of the training,the convolutional neural network can accurately predict the user's emotions based on the user's textual data.In order to evaluate the effectiveness and stability of the convolutional neural network model trained by the graph model,in this thesis,a convolutional neural network model is built and the graph model is used for training,the experimental results are counted,and the performance of the model is analyzed from different aspects.In order to compare with the neural network model of word vector training,the thesis also builds the convolutional neural network model and the cyclic neural network model and uses word vector for training respectively.The three experimental models used the same training data set and test data set to calculate the experimental results and compare the performance of the three models in the task of text emotion prediction.The comparison results of the three experimental models show that the accuracy of the convolutional neural network model trained by the graph model is higher than that of the convolutional neural network model and cyclic neural network model trained by the word vector,but the training cost is higher.