Researches On Emotion Recognition Method Based On Stroke Behavior Of Smartphone

With the advent of the information age and the development of artificial intelligence technology,touch screen devices such as smartphone have become an indispensable part in people's life.Identifying the user's emotional state according to the usage of smartphone is a new way to improve the human-computer interaction and user experience.In this thesis,we present an attempt to recognize user's emotional states by using finger-stroke pattern of the smartphone.First of all,the emotional induction experiment is designed to induce three kinds of basic emotions: positive,neutral and negative.Then we analyze and process the data of stroke screen and extract the characteristics of stroke.Finally,the classification of emotion is classified by the classification method of machine learning,so as to achieve the purpose of identifying the user's emotional states through the behavior of stroke screen.At first,through the International Affective Picture System(IAPS)and the psycho-logical experiment platform E-Prime to design the emotion induction experiment,induce the positive,neutral and negative emotional states of testers.Under three basic emotional states,using Xshell and SSDroid platform to collect the data of stroke screen,and the EEG signal is collected by using EEG signal acquisition system and 64 conductive cap.After processing the EEG signal and extracting the features of frequency band associated with emotion,we get the average classification accuracy of 84.12% by using support vector machine(SVM)classification algorithm,thus the effectiveness of emotion induction is verified.Afterwards,we process and analyse the data of stroke under different emotional states,and the average value,median value,maximum value,minimum value and variance of the four indexes,including the length,time,velocity and pressure are extracted as the features of stroke screen.The feature selection algorithm of Correlationbased Feature Selection(CFS)is used for select features.Ultimately,four kinds of classification algorithms including Bayesian network(BN),back propagation neural network(BPNN),random forest(RF)and an ensemble classification algorithm based on Three-Way Decisions are used to classify the emotions from the perspective of the individual and the group.In the study of personal emotion recognition,the results of BN,BPNN,and RF classifiers are compared before and after feature selection.The results show that the classification result after feature selection is increased by about 1%.The research shows that the behavior of stroke exist significant differences between different genders,so we train classifiers for male,female and all testers after add the feature of gender.The average accuracy of three traditional classifiers on the three groups was obtained by 72.3%,74.6% and 70.8% respectively.Among three traditional classifiers,RF algorithm has the best classification effect.The recognition accuracy of the ensemble classification algorithm based on Three-Way Decisions is improved by 2.2% compared with RF.