Research And Implementation Of Interactive Historical Emotion Analysis For Cloud Contact Center

With the rapid development of mobile devices and mobile networks,most people can access the Internet through networked devices,and people can easily express their emotional tendencies through many social platforms.When these views with subjective sentiment orientation reach a certain number,these views will have great application value and commercial value.If the number of views is huge,the emotional tendency will be judged only by manpower.Although the effect of the analysis will be better,it will cost a lot of manpower,and the sentiment analysis algorithm can judge the general tendency of these views in a very short time.This paper focuses on the emotional tendencies of chats and product reviews.Chatting records belong to partial short text,product reviews belong to long length text,product reviews tend to be more obvious than chats.In this paper,sentiment analysis based on sentiment dictionary,SVM and deep learning algorithm is applied to two kinds of data.In this paper,the same algorithm is used to analyze the emotions of two types of data and compare the effects.In addition to exploring the influence of different algorithms on the experimental results,the same algorithm is also used to explore the influence of different parameter values on the experimental results.For emotion dictionary,this paper designs the rules of emotional value calculation and tests,and finds that the effect is not very ideal.On the one hand,it is the limitation of the emotion dictionary itself.On the other hand,the number of emotional polarity makes the threshold not good enough.For SVM algorithm,this paper designs a variety of feature extraction methods,including word vector,word frequency and document feature vector,and their combination features.After testing,it is found that combination of word vector and document feature vector is a good feature extraction method.For deep learning,this paper designs six types of sentiment analysis models,which mainly use LSTM,Attention and CNN,and their combination structure.These network structures have their own advantages.LSTM is good at learning the characteristics of text sequences.Bi LSTM is an improvement of LSTM.It can learn both the above information and the following information.Attention is good at extracting important information and discard unimportant information,while CNN is good at learning local features and deep level features.After testing,Bi LSTM plus Attention found that the Attention mechanism can really improve the original model.In this paper,two kinds of neural network models of CNN and Bi LSTM are designed.These two models have achieved very good results in product reviews.In all kinds of algorithms,the accuracy of product reviews is higher than that of chats.In this paper,the defects of chats corpus are optimized and some effects have been achieved.In addition to studying the sentiment analysis algorithm,this paper applies the sentiment analysis algorithm to the cloud contact center.After the end of the traditional customer service service,it is usually done through the manual scoring of customers.There are many times that there is no evaluation after service,so that the quality of service can not be obtained.In this paper,we use emotional analysis technology to analyze customer service and customer chat records to evaluate the service satisfaction.After testing,most of the conversations achieved the desired results.The degree of satisfaction can not only monitor the quality of service but also affect the conversation search.This paper also implements a dialogue search based on the satisfaction score.When a customer queries a question through a dialogue search,the search algorithm combines the two influence factors of keyword score and satisfaction score to sort the dialogue.After testing,the effect is better than that based on keyword scoring.