Multi-task Affective Computing With Graph-guided Fusion Based On Physiological Signals

With the development of science and technology,human-computer interaction technology has been applied more and more widely.On the one hand,the intelligence of the machine is reflected in the user emotion recognition,especially physiological signal takes the important position of the field of affective computing.However,in experiments and applications,we often neglect the existence of some priori knowledge between the low level features and the emotional state.In this paper,we present a new method for the statistical analysis of emotion.The main work of this paper is to build a model by adding a priori knowledge,so as to classify emotion states more accurately.The main contents are as follows:First of all,in order to get accurate and effective a priori information from the source dataset which could correctly reflect the correlation pattern between the emotional state and the physiological signal.In this paper a dataset establishment has been designed and completed.We used video as the source of inducement and invited 30 students from our school as subjects to collect and record their physiological signals in four basic emotional states,including ECG,GSR,PPG.Then,the video and audio information of the subjects were accurately recorded during the experiment to mark the collected physiological signal data.Finally,the collected emotion data are preprocessed and the statistical feature extraction and feature fusion is performed.Our goal is to find a mapping from the low-level feature to the emotional states.We use the graph model fusion L1 norm method to obtain the feature selection matrix and then use the concept of transfer learning to make feature selection to target data set for each emotion state.Finally,the classifier is used to test the model and verify our hypothesis.In this paper,we use DEAP database as target data set,and some traditional machine learning algorithms such as Support Vector Machine and Naive Bayes are used to predict.The experimental results show that the method we present is more efficient for the emotion classification and the results are compared with those traditional machine learning algorithm and other existing results.Experimental results also demonstrate that the proposed method has a better classification performance.