Emotion Recognition Of Mouse Track Based On Neural Network

Emotions are a part of people’s daily lives,so the recognition of people’s emotional state is very important.For the problem of emotion recognition,the traditional methods of emotion recognition research are mainly on people’s external behavior indicators or people’s physiological signals.However,it is difficult to collect these signals,which requires the high cooperation of volunteers or the high cost of wearable devices for data collection.In the context of the rapid development of artificial intelligence,this paper proposes a variety of methods for emotion recognition of mouse track based on neural network.There is some commercial value in identifying people’s emotional states when they use computers.For Internet companies,they can provide services that make customers more satisfied.The specific research content and results of this paper are as follows:1.When building a database,this paper uses a method of volunteers for human collection.The collected data is first processed by the Excel for VBA program,and then sent to the matlab for matrix transformation processing.Finally,a standardized data set is generated.2.In the Back-propagation neural network model,this paper innovatively adopts two feature extraction methods: integral transformation and statistical analysis.The model trains and tests the collected data based on BP neural network,and calculates the correct rate of emotion recognition under different states.Experimental results show that the results of the two methods are comparable and have good results for different individuals.However,in some extreme cases,when the individual’s habit of operating the mouse is quite different,the result is unreasonable.This is also a major problem to be solved in the follow-up research.3.After completing the experiment of BP neural network,this paper also attempts to introduce a deep neural network model.This article uses three different network models,including CNN、ResNet and RNN.The model trains and tests the collected data in the Keras framework.Experimental results show that the Convolutional neural network model and residual network model can get good results.And the effect of the residual network model is the best of all the models in this paper,the test recognition accuracy can reach 98%.And the results are more stable than the BP neural network model.But for the Recurrent Neural Network,the results show that this model is unsuitable for solving the research problem in this paper.In general,the first three models of this paper can achieve the expected results.We can found that it is effective to identify the human emotional state by collecting people’s mouse motion trajectories.