Smartphone Based User Personality Detection And Its Application

With the rapid spread of smartphones,smartphone-based sensing technology has developed rapidly.Smart phones are equipped with a wealth of sensors,efficient pro-cessors and large storage space,which broadens the scope of research for traditional social science research:researchers use sample data from smartphones to replace tra-ditional survey sampling.Traditional social science research has made a detailed description and clinical ob-servation of personality psychology,but with the popularity of smart phones,the study of personality psychology is no longer limited to clinical observation and research.To-day,smartphone devices are able to perceive large amounts of behavioral data:GPS locations;communication data,including call text messages and contact data;motion status,motion data acquired through acceleration and gyroscope sensors,and more.In the related behavioral research of user behavior(user emotions,personality,behavior,etc.)based on behavior-aware data,there are some shortcomings in research method-s:firstly,the user's state is inferred by collecting user behavior data for a long time.The strategy of collecting user data requires users to continuously cooperate with the survey,which reduces the efficiency of data collection.Secondly,in the work relat-ed to personality prediction based on smartphone,the single task prediction method is usually adopted,that is,each personality attribute of the Big Five personality is pre-dicted separately.It does not take into account the association between the five types of personality attributes,ignoring the interrelationship between personality attributes.The specific work of this paper can be summarized as:(1)A research method of fine-grained personality trait prediction based on smart-phone snapshot data is proposed.Different from the existing methods based on mobile sensor data,the method proposed in this paper only needs to use the snapshot of the smartphone to quickly complete the Big Five personality score prediction without long-term collection of user data(smartphone snapshot refers to storage in the smartphone)Call logs,SMS logs and application usage logs),while smartphone snapshots can be instantly acquired using the API provided by the Android system.This paper extracts call text messages and application usage characteristics based on smartphone snapshot data,and analyzes the correlation between these features and self-reported personality test scores.The results of the analysis indicate that there is a significant correlation between behavioral characteristics extracted from smartphone snapshot data and self-reported personality test scores.(2)A multi-view multi-task model with attention mechanism and hybrid expert network is proposed.The attention mechanism can obtain the importance weight of different views to improve the model prediction accuracy.The hybrid expert network can input different samples.Different feature representations are used to obtain better feature representation effects;the multitasking model links the prediction tasks of five types of personality attributes,fully considering the correlation between personality attributes,and making the prediction accuracy significantly higher than the baseline model.(3)The application of personality perception model is studied.Predicting stu-dent achievement is very important for students' learning planning or career planning.Personality perception model can help and improve the accuracy of students',GPA s-core prediction.In this paper,we propose to predict student achievement based on personality-awareness models and APP usage data collected on smartphones.Based on the data collected,our analysis shows that students' GPA scores are highly correlat-ed with the usage patterns of smartphones.We further demonstrate through experiments that the student's GPA score can be predicted with high accuracy in advance through the personality perception model and the extracted application statistical features.