Development Of Automatic Empathy Testing System Based On Eye Movement

Purpose : To solve the problems of subjectivity and interference of social expectations,based on research results in the field of machine learning,using standardized visual materials,collecting eye movement data of the subject's perceptual visual materials,by extracting the global and local eye movement features of the subject's eye movements,automatic classification learning using public machine learning platforms,thus forming a system for the machine to automatically determine the level of empathy.research method: This article takes a combination of psychological experiments and machine learning,first collected the eye movement information of 166 college students watching semi-projected affectionate pictures through SMI RED500 telemetry infrared sensor eye tracker and collect the IRI-C Interpersonal Response Index Scale.According to the scoring criteria of the scale developers,the empathy scores of the participants were obtained.Then the subjects 'eye movement characteristics were extracted as independent variables,the original data of eye movements include coordinate data,pupil diameter,eye movement events(only blink time)and their corresponding time series.After a series of feature extraction,screening,dimensionality reduction and other processing and the subjects' IRI-C scores were used as dependent variables.Using Leave one method,use the models in existing machine learning platform scikit-learn(Linear Regression,K-Nearest Neighbor KNN,Random Forest,Support Vector Machine Regression SVR)for eye movement feature extraction and learning,ability to automatically predict individual empathy.Research result: This study uses the original eye movement characteristics obtained by viewing semi-projected pictures as the basis,and obtains the highest performance prediction system based on SVR.On the 166 samples,the empathy prediction score obtained by the leave-one-out method was significantly correlated with the empathy score in the real IRI-C questionnaire,and the Pearson correlation coefficient was 0.95.The prediction error in the four dimensions of IRI-C and the total score is between 6.5% and 10.5%.Although random forest and KNN are integrated models of multiple decision tree fusion,which can be used for regression tasks and are widely used in emotion recognition,however,in this study,the Pearson coefficient r = 0.63~0.70 between the predicted value and the observed value of the empathy score.That is,the results show that different models have large differences in the learning of eye movement data,but at the same time,it is found that it is feasible to use machine learning to predict empathy based on eye movement data.At the same time,it also shows that eye movement characteristics such as time series dynamic indicators and heat maps may have a relationship with empathy.Conclusion: The prediction of empathy based on eye movement data has a good effect and high accuracy.The measurement of empathy is no longer totally dependent on the empathy scale,and machine learning can be used to construct a more objective,accurate,efficient,and true measurement method.