Research On Personality Recognition And Personality Privacy Protection Text-Based In Social Networks

Personality is a high-level abstract of individual human characteristics and it is also a scientific quantitative criteria for distinguishing the differences between people.Personality is not only closely related to human behavior in the real world,but also highly related to social network behavior in the virtual world.At present,a lot of research work has been done on user personality analysis and recognition in social networks at home and abroad.The user personality analysis and recognition model based on big data analysis methods and machine learning methods is currently a more effective and accurate method.At the same time,personality is a stable and important psychological feature,which has been widely used in research in the security field and it is also an important factor in the use of social engineering attacks.The current personality research in social engineering faces two major challenges:First,due to the particularity of the personality datasets,the existing personality analysis and recognition methods still have defects in data processing and model innovation and cannot extract deeper and more complete feature information.The accuracy of personality analysis and recognition based on artificial intelligence methods is still low.Second,personality as a part of the user’s individual privacy needs to be effectively protected,but the research on personality privacy protection is almost in a blank state.There is a lack of research on personal privacy protection methods and a complete set of analysis,identification and protection system framework of personality in social engineering.This thesis comprehensively utilizes the relevant characteristic attributes of user text data and artificial intelligence learning methods to study a complete set of personality recognition and protection methods.The main research results are as follows:1.Personality recognition is an important basis for recommendation system,political prediction,psychological research and social engineering safety protection.This thesis studies personality recognition based on data processing.In order to improve the accuracy of personality recognition,this thesis proposes a personality recognition methods based on PSO-SMOTETomek technology.Aiming at the problem of serious unbalanced data distribution,this thesis firstly proposes the method of combining over-sampling and under-sampling to achieve data balance.Then,to solve the problem of incomplete extraction of existing personality features,this thesis adds two categories of mental lexicon TF-IDF and text style feature in feature extraction,and applies dimension optimization to all the features.Finally,shallow machine learning algorithms are used for personality classification.Experimental results show that,compared with the latest personality recognition model,the accuracy of personality recognition on the two open and authoritative datasets is increased by 3-10 percentage points and 4-8 percentage points respectively,achieving the purpose of improving the accuracy of personality recognition.2.The personality recognition methods based on shallow machine learning algorithms needs to extract a lot of features manually,which consumes a lot of manpower and time.Besides,the use of five groups of independent characteristics to classify the five personalities doesn’t take the correlation among user personality traits into account.This study aims to construct an end-toend personality recognition model using a deep learning-based framework.In order to solve the above problems,this thesis proposes a personality recognition model based on graph convolutional neural network.Firstly,a large heterogeneous personality graph is constructed based on the relationships among users,documents,words and word co-occurrence of the whole personality corpus.Then,under the supervised learning of category tags with known users and documents,the feature automatic learning is realized by embedding three types of nodes in the graph.Finally,by sharing a set of personality characteristics,this thesis carries out multi-classification tasks for personality based on the full connection layer,integrating the correlation between five personality traits into personality recognition and improving the accuracy of personality recognition in automatic personality recognition.Experimental results on two public and authoritative benchmark datasets show that the proposed model is superior to the latest approach.The average accuracy of personality recognition is improved by 2.75-3.17 percentage points and the average F1-score is improved by 2.4-9.2 percentage points.3.The protection of personality privacy in social engineering is an important basis for security protection research in the fields of cyberspace security,social engineering security and cognitive domain security in the future.However,the research on the protection of personality privacy is almost blank.In order to preliminarily realize the protection of personality privacy,this thesis proposes a framework of personality privacy protection based on adversarial text generation.This thesis,based on the adversarial method of gradient descent and cosine similarity calculation to generate adversarial text,proposes a personality privacy protection framework that combines BERT algorithm,fast gradient projection method and cosine similarity calculation,so that the personality recognition model can not accurately identify personality traits.Simulation results on two benchmark datasets show that the accuracy of the model is reduced by 25 percentage points and 56 percentage points respectively,which achieves the purpose of protecting personality privacy.