Research On Key Technologies Of Public Opinion Sentiment Analysis For Segmented Fields

Customer satisfaction with the service is related to the service provider's benefits.To effectively maintain existing customers and develop new customers,service providers need to analyze the emotional information contained in the customer's feedback in time and take measures to deal with it quickly,for improving the customer experience.With the rapid development of the information industry,such as the Internet and the mobile Internet,the customer satisfaction of customer with the service can be more conveniently fed back through the network platform.The Internet has replaced traditional channels as the main feedback carrier,and massive unstructured text feedback will be generated by customer groups every moment,and traditional feedback analysis methods which relies on manual processing has been difficult to meet the needs for enterprises customer relationship management;at the same time,enterprise services involve different topics,and it is necessary to accurately communicate feedback from different fields to the corresponding departments to make feedbacks properly handled.However,the text of segmented fields has its own particularity meaning in corresponding field,and the same language expression has different emotional tendency in different fields.Therefore,how to design a method which can classify massive unstructured customer feedback and analysis public opinions at same time,and with above method construct a system which can fast and automatic complete sentiment analysis of unstructured text in segmentation fields and automatically adapt to the ways of emotional expression on different fields has become an urgent problem in the maintenance of customer emotional relationship.Based on the above background,to carry out automatic analysis and processing of unstructured customer feedback texts efficiently,this paper to the demand for the sentiment analysis of the segmentation fields such as ticket service,hotel service and catering service in the life service website,and key technologies relate to above such as text classification of segmentation fields and public opinion sentiment analysis have been thoroughly studied.The main research contents and results are as follows:(1)A text classification algorithm based on attention mechanism and adversarial training is proposed.As an important task of text analysis,text classification has been widely studied,and many methods have been proposed,such as Latent Dirichlet Allocation(LDA)model text classification method,Bag-of-word model text classification method and Support Vector Machine(SVM)text classification method,etc.These methods treat words as symbols and record the presence or absence of symbols in the text and the contribution rate of the symbol to a subject(category),while ignoring the semantics represented by the word itself and the order among the words.Based on Recurrent Neural Network(RNN),this paper based on the recurrent neural network applies the attention mechanism,so that the model can better preserve the order relationship and long-distance dependence between text words,and automatically increase the weight of keywords for text classification,so that the classifier has better performance.At the same time,the adversarial training was used with generating the disturbance of word embedding in the process of model training to make the model got higher generalization ability and became robustness.Experiment results show that this method has better performance than the baseline method.(2)A sentiment analysis algorithm combining Piecewise Convolutional Neural Network(PCNN)and Generative Adversarial Network(GAN)is proposed.As one of the most important tasks in the field of public opinion monitoring,service evaluation and satisfaction analysis in the current information-society environment,text sentiment analysis needs to extract the opinions and preferences of customers in the text.Compared with traditional natural language processing and analysis tools,convolutional neural network is an effective method for automatically acquiring sentence features in deep learning,and It can learn features which have the most relevant with sentiment analysis tasks from sentences and improves the performance of sentiment analysis models.However,the original convolutional neural network model ignores the sentence structure information that is important for text sentiment analysis,and it is easy to overfit.In view of the above deficiencies,this paper adopts the piecewise strategy,which enables the deep learning-based convolutional neural network model to model the sentence structure and extracts the main features of the different structures of the sentence to analyze the sentiment tendency of the text and use the Dropout algorithm to enhance the generalization ability of the model.At the same time,the user's feedback on the service involves many different fields,and the annotation data in each field is less,to alleviate the problem of the data sparsity,this paper also uses the generated adversarial network to perform common feature extraction,so that the model can acquire the common features with feedback of different fields,and enhance the generalization ability of the model in case of less training data.Experiment results processing on different data demonstrate the effectiveness of this method.(3)An integrated sentiment analysis method based on Recurrent-Convolutional Neural Network(R-CNN)and Convolutional-Recurrent Neural Network(C-RNN)of gated unit is proposed.At present,the sentiment analysis algorithms with good effects are all based on statistical learning methods.The performance of those methods depends on the quality of feature extraction,while good feature engineering requires high expert experience and is time-consuming and Laborious,and poor portability.The neural network approach can reduce the dependency of feature engineering,and RNN can obtain context information but have semantic information bias problem;CNN-based text analysis method can obtain important features of text through pooling but it is difficult to obtain context information;For example,although sentiment analysis algorithm combining PCNN and GAN which was proposed in this paper can partially alleviate the shortage of CNN using piecewise strategy,but the model is still poor for long-distance dependency.Aiming at the above problems,this paper proposes a sentiment analysis method based on the combination of R-CNN and C-RNN based on the gating unit.Firstly,RNN and CNN are combined in different ways to alleviate the shortcomings of the two and building sub-analysis network R-CNN and C-RNN separately,finally combine the two networks through fusion gate to form the final analysis model.We performed experiments on different data sets,and the results verify the effectiveness of the method.(4)The combination of group sparse and exclusive sparse regularization terms is proposed to compress the sentiment analysis model.The sentiment analysis algorithm combining PCNN and GAN and the integrated sentiment analysis method based on R-CNN and C-RNN of gated unit proposed in this paper both use large-scale convolutional neural networks to ensure the performance of the models,resulting in larger parameters of the model.In practice,marked data is less,and resulting in that the model cannot be fully trained.At the same time,the sentiment analysis system needs to have high timeliness to be able to quickly analyze the occurrence of public opinion,respond to customer feedback in a timely manner,and effectively manage customer relationship.To solve the above problems,this paper proposes to use the group sparse and exclusive sparse regular terms to trim the model in the model pre-training process.Firstly,the sparse regularization term is used to cut off the edges with smaller weights,and the sparsely connected neuron nodes are removed.Then the model was trained after pruning.We have carried out enough experiments on different data sets to verify the effectiveness of the compression method,which improve the efficiency of the network in the prediction,while ensure that the performance of the model without greatly reduce.(5)Design and construction of customer satisfaction analysis prototype system based on the above methodBased on the above research,this paper implements the customer satisfaction analysis prototype system based on Spring-Boot framework with B/S architecture.The core functions of data preprocessing,spam filtering,segmentation fields segmenting and customer sentiment analysis are implemented,and the prototype system simulation test was carried out,which fully demonstrated the effectiveness and practicability of the method proposed by this paper.