The Immune Algorithm Of Abnormal Emotion Detection In Keystroke Behavior

People's daily life and work have been inseparable from keyboard operation,keystroke behavior has been as natural as writing habits.Similar to handwriting features,users also form their own unique keystroke features.The initial keystroke behavior is used to authenticate the user identity.By establishing the keystroke template of the user identity and comparing the difference between input and template,the input is judged whether the user is himself or not.Later,some scholars found that,in keystroke authentication,the user emotions will affect the accuracy of keystroke authentication,and gradually the MIT's laboratory and others began to study the identification of user emotions from keystrokes.Through the identification of user emotions,we can get the user feelings to enhance the user experience,and also can extract user characteristics and analyze user needs according to user emotions.Because the composition of human emotion is complex and the expression of emotion changes with time,the diversity and adaptability of emotion recognition become a hot topic.Keystroke features are mainly divided into time and pressure,and since the latter requires special equipment,the former is more widely used.The key of keystroke authentication is to extract user keystroke time features to build user templates.As the user changes,the template changes accordingly.After the user template is established,the user template needs to be updated to adapt to the changes of the user,and the composition of the template should be enriched to accommodate the user's emotional diversity.At present,there are two main methods for emotion recognition of keystroke behavior:machine learning and statistics.Although the former is flexible,it needs enough data to model and update.In reality,keystroke data about emotion is limited,so enough data is difficult to satisfy.Statistics is simple in principle and convenient to calculation,but it is not flexible enough to meet the strain.The recognition of abnormal emotion of keystroke behavior is mainly divided into two modules:detector training,detection module,how to complete the detector update while the training is completed,which is the blank of current research.Under the background of current research,the classification methods of emotion recognition are various,but lack of perfect mechanism to adapt to renewal.In order to solve the problem of diversity and adaptability updating of detection,in order to improve the accuracy of detection.Under the background of computer immunology research,this paper draws lessons from the diversity and adaptability advantage of immunity,increases the number of templates,increases template elimination,replacement,and update mechanisms and improve the method of template decision making in testing.Replace and update the mechanism to improve the scientific nature of template decision-making in testing.By increasing the number of detectors to enrich and expand the range of detection,by weight to determine the single detector in the final detection results,through the immune variation,the concept of hybridization to produce a new detector to enrich the diversity of detectors.This paper uses the idea of computer immunology for reference in solving the emotional recognition problem of keystroke behavior for the first time,and solves the adaptation problems of emotional diversity and emotional change through the updating and weight determining mechanism of multiple templates.According to the survey,students feel that the course selection experience of Wuhan University is poor,prone to emotional overreaction,and the emotion during normal time is stable.This paper takes the educational administration system of Wuhan University as the experimental platform to collect the logging data of the users under the abnormal and normal emotions during the elective and non-elective courses.After pretreatment,Euclidean distance and Manhattan distance are used as similarity distance respectively,and the effectiveness of the proposed algorithm and the K-nearest neighbor algorithm are compared experimentally to prove the effectiveness of this paper's methods.