| With the rapid development of the Internet,social networks and major platforms have become the gathering place of data and information.How people mine the information and value empowerment they need from massive data is also a challenge.Emotional analysis is an important research content in the field of natural language processing.It is widely used in marketing,politics,finance and other fields,and has important scientific research value and social value.At present,the mainstream method of emotion analysis is carried out under the framework of deep learning.It focuses more on the emotional semantics related to views.It needs to use the word embedding information based on sentences and the construction of models to effectively mine emotional information.In addition,social network texts and product reviews often exhibit characteristics such as short space,non-standard grammatical structures,and richer emotional expressions.Therefore,compared to traditional machine learning algorithms,deep learning is more suitable for the abstract and complex characteristics of text.This paper focuses on the emotional analysis of short text from the following aspects.(1)Convolutional memory network text sentiment analysis based on rough data reasoning.In order to solve the problem that the existing emotion classification algorithm lacks the use of semantic association rules in feature extraction,and generates a large number of words unrelated to emotion prediction after word segmentation,resulting in the unrepresentative features mined.A convolutional memory network emotion analysis model based on rough data reasoning is proposed.The emotional word set Word2 Vec word vector representation of the text is obtained by rough data reasoning through context information,and the feature vector embedding layer is improved by fusing the Fast Text word vector.Secondly,convolutional neural network(CNN)is used to splice bi-directional long short-term memory(Bi LSTM)to extract deeper emotional features.Finally,the Attention mechanism is added to calculate the weight and screen significant and important features.The experimental results on two data sets show that the accuracy rate of emotion classification and F1 value of the model can reach 84.66% and 85.1% respectively,2.04% and 3.1% higher than the highest value in the baseline model,effectively improving the prediction ability of emotion classification.(2)Emotional analysis of Chinese text in two-channel attention network based on mixed word embedding.Solving the problem that the traditional static word vector embedding method cannot effectively deal with the polysemy problem in Chinese text,and it is difficult to mine the contextual emotional features and internal semantic association structure.Firstly,the emotion elements related to the text are integrated into Word2 Vec and Fast Text word vectors by rough data reasoning in a channel,and CNN extracts the local features of the text;Secondly,another channel uses BERT for word embedding supplement,and Bi LSTM obtains the global features of the text.Finally,the attention calculation module is added for the deep interaction of dual-channel features.The accuracy of the experiment on three Chinese data sets reached 92.43%,0.81% higher than the highest value of the benchmark model.The selected dataset is only for coarse-grained emotion classification modeling,and has not considered the experiment in fine-grained field.According to the output results,the emotion category is determined,which effectively improves the performance of the model for Chinese text emotion classification.(3)Aspect-level emotion analysis based on GCN and aspect region representation based on rough data reasoning.At present,most aspect-level affective analysis studies have ignored the modeling of the interaction between the aspect words and the context,and the graph convolution model based on the dependency tree is highly dependent on the parsing quality of the dependency tree.To solve this problem,a GCN based on rough data reasoning and an aspect level emotion analysis model(AAS-RGCN)based on aspect region representation are proposed.First,Bi LSTM is used to extract aspect word features,then GCN is used to construct text graph structure based on rough data reasoning to extract context features that have semantic relationship with aspect words,and finally,attention mechanism for learning interactive information optimize classification output between aspect features and context features.Regarding the dataset,it’s shown that its accuracy and macro F1 value have achieved better classification results than the benchmark model. |
