Research On The Method Of Categorizing Emotions In Comment Text Based On Sentence Pattern Rules And Machine Learning

Text emotion classification refers to the classification judgment of the emotional tendency of the text through the mining and analysis of subjective information such as views,opinions and emotions in the text.The review information in various fields is increasing step by step,it provides a basis for enterprises and users to make a more reasonable judgment on future choices by referring to comments.This paper studies the text emotion classification of e-commerce review text.It mainly starts from three aspects:The first is the judgment of emotion tendency based on emotion knowledge,the second is the construction of optimal emotion classifier based on emotion knowledge and machine learning algorithm,and the third is the research of emotion classification based on deep learning network model.Firstly,a new method based on position weight parameter tuning and sentence style emotion calculation rules is proposed to solve the problem of emotional orientation of long critical text.In this method,the text is decomposed into a set of clauses,which are divided into different parts and assign the corresponding position weight parameter.The emotional score is calculated based on emotional vocabulary,and the optimal weight parameters are obtained by experiments.Next,new affective computing rules are applied to long sentences in text,summarize the four related words according to the different sentence patterns,and assign the corresponding scores,combine the weight tuning experiment's optimal position weight parameter with the sentence rule algorithm,summarize the emotion calculation formula,and judge the emotion scores of each comment text in turn.Experiments show that this method is more accurate than other algorithms.Then,the machine learning algorithm is combined with emotion knowledge to construct the optimal emotion classifier.In feature extraction,all words and the two-word collocation with large amount of information are used as features,and the chi-square statistical method is used to find the optimal threshold after testing.Use scikit-learn for classification tasks,and input the obtained characteristic phrases into each classifier as independent attributes.At the same time,the emotion score obtained from the comment data based on the emotion knowledge algorithm was normalized and fused into the characteristic matrix of the constructed classifier,and the optimal emotion classifier was obtained through training and stored.Finally,on the basis of increasing the data set,deep learning algorithm is used for emotion classification task.In the process of preprocessing,in this paper,one-hot MLSTM,Character MLSTM and Word MLSTM are constructed to test the advantages of one-hot,Word vector and Word vector in model classification and verify the necessity of Word segmentation in deep learning emotional classification.The experimental results show that the classification accuracy of word vectors is higher.Also built MLSTM + Self-Attention network model,namely the length of the multilayer memory networks and the Attention mechanism model.Keras is used to construct and train a multi-layer LSTM model,and then auto-attention mechanism is added to the feature vector to pay attention to the information with large weight coefficient,and a multi-group comparison experiment was conducted with other network models,the results show that the model is stable,emotional optimal classification accuracy.