User Behavior And Identity In The Organization

Behavioral detection methods are designed to identify user behavior and build a user's identity through behavioral data in an organization.In the face of massive data,traditional behavioral detection methods based on regular matching are difficult to im-plement;while machine learning models are also faced with problems such as scattered behavior data,difficult association between data sets and serious environmental impact.We introduce the influence factor as the core influence factor in a variety of machine learning methods.Then we propose an asset evaluation method and two models with application prospects,which effectively improve the accuracy and time performance of behavior detection.The main contributions of this dissertation are as follows:1.An asset valuation method based on enterprise data is proposed.In the compa-nies surveyed,its data collection and storage are scattered.Demand analysis between different business groups is achieved through the line of business.Therefore,we pro-pose a human-centered concept,which allocates all assets to different roles to obtain as-set distribution and then quantifies its authority through assets between different roles.The distribution of permissions obtained by this method of evaluation provides assis-tance for the experiment.2.A scheme for dynamically building an RBAC model is proposed.On the role control list generated by the model,the least privilege group generation algorithm is given and the roles with similar privilege are divided into the same min-privilege group.Based on the min-privilege group,three classification models NB,ID3 and SVM are merged by the ensemble learning algorithm stacking.In the comparative experiment,the accuracy and time performance of different models are analyzed.The experimental results show that the RBAC-Ensemble model can improve the accuracy by nearly 6%without increasing the time consumption.For the traditional model,the min-privilege groups can improve the time performance of 14%under the premise of improving the accuracy of 4%.3.A data gravity model based on the min-privilege groups is proposed.The stan-dard data gravity model discriminates the interaction force according to the difference between the data,so as to judge the correlation between the data.In order to further correlate to the user behavior data,we modified the data distance formula and the grav-ity addition method.Then,the gravitational scope is limited to the inside of the group member by introducing the min-privilege groups.Finally,the test data label is obtained by the mass weighted average.The experimental results show that the proposed model recognition accuracy is only 1%lower than the ensemble learning method and can be further improved by improving the machine performance.