CN-121984761-A - Dynamic credibility maintaining method, system and storage medium

CN121984761ACN 121984761 ACN121984761 ACN 121984761ACN-121984761-A

Abstract

The application discloses a dynamic credibility maintaining method, a dynamic credibility maintaining system and a storage medium, and belongs to the technical field of computer security. The method comprises the steps of applying an irreversible projection function to multi-mode original data of user behaviors or extracted statistical features of the multi-mode original data to generate behavior state vectors, comparing the behavior state vectors with a pre-built user personal behavior baseline model to detect anomalies, locating anomaly components and anomaly parameters which cause the anomalies when the anomalies are detected, dynamically generating context response challenges according to the anomaly components and the anomaly parameters, updating a trust model representing the user credibility state according to the response of a user to the challenges, and finally dynamically adjusting system operation parameters based on the current state of the trust model. The application protects the privacy of the user through irreversible transformation and immediate destruction of data, improves the anti-cheating capability through context reaction type challenges, and realizes the balance of safety and user experience through a closed-loop dynamic adjustment mechanism.

Inventors

WANG JIANBING
CHEN QIAO
Ren Peishi
CHEN SHUFEN

Assignees

上海浩宜信息科技有限公司

Dates

Publication Date: 20260505
Application Date: 20260206

Claims (10)

1. A dynamic reliability maintenance method, comprising the steps of: Constructing a user personal behavior baseline model, namely constructing the user personal behavior baseline model based on historical behavior data of a user in advance; Periodically applying a fixed irreversible projection function to the multi-modal raw data representing the user behavior or the statistical features extracted by the multi-modal raw data on the client device to generate a behavior state vector, and clearing the multi-modal raw data from the client device after generating the behavior state vector; comparing the behavior state vector with the user personal behavior baseline model to detect whether an abnormality exists, and positioning at least one abnormal component causing the abnormality in the behavior state vector and an abnormal parameter of the abnormal component when the abnormality is detected; Generating a challenge, namely dynamically generating a context-reactive challenge according to the abnormal component and the abnormal parameter of the abnormal component; Updating a model based on a user's response to the context-reactive challenge, updating a trust model characterizing the user's trustworthiness status, and Dynamically adjusting at least one system operating parameter based on a current state of the trust model, wherein the system operating parameter is at least one of a frequency of generation of the behavior state vector, a decision threshold for detecting anomalies, and a complexity of the context-reactive challenge.
2. The method of claim 1, wherein the irreversible projection function is a weight-invariant randomly initialized neural network or a locally sensitive hash function.
3. The method of claim 1, wherein the anomaly parameters include at least one of an index of a component causing an anomaly, a direction of departure of the anomaly component, and a magnitude of departure of the anomaly component.
4. The method of claim 1, wherein the trust model is a trust entropy model comprising a trust score representing an absolute level of user trustworthiness and a trust entropy representing volatility of the trust score.
5. The method of claim 1, wherein the responding to the context-responsive challenge by the user comprises analyzing whether a newly generated sequence of behavior state vectors of the user in completing the context-responsive challenge conforms to the user personal behavior baseline model of the user.
6. The method of claim 4, wherein dynamically adjusting at least one system operating parameter comprises: And when the trust score is reduced or the trust entropy is increased, the generation frequency of the behavior state vector is increased and/or the judgment threshold value for detecting the abnormality is tightened.
7. The method of any of claims 1-6, wherein the detecting anomalies, the generating challenges, the updating model, and the dynamic adjustments are all performed on the client device.
8. A dynamic trust maintaining system for implementing the method of any one of 1-7, comprising: The baseline model construction module is used for constructing a user personal behavior baseline model based on historical behavior data of a user in advance; A behavior state vector generation module, configured on a client device, for periodically applying a fixed irreversible projection function to multi-modal raw data representing user behavior or statistical features extracted from the multi-modal raw data to generate a behavior state vector, and after generating the behavior state vector, clearing the multi-modal raw data from the client device; An anomaly detection module for receiving the behavior state vector generated by the behavior state vector generation module, comparing the behavior state vector with the user personal behavior baseline model constructed by the baseline model construction module to detect whether an anomaly exists, and locating at least one anomaly component causing the anomaly in the behavior state vector and an anomaly parameter of the anomaly component when the anomaly is detected; a challenge generation module for dynamically generating a context-reactive challenge based on the anomaly component and an anomaly parameter for the anomaly component located by the anomaly detection module; A model updating module for updating a trust model characterizing the user's trust status based on the user's response to the context-reactive challenge generated by the context-reactive challenge generating module, and And a dynamic adjustment module configured to dynamically adjust at least one system operating parameter based on a current state of the trust model updated by the model update module, where the system operating parameter is at least one of a frequency of generation of the behavior state vector, a decision threshold for detecting anomalies, and a complexity of the context-reactive challenge.
9. The system according to claim 8, wherein: The irreversible projection function is a random initialization neural network or a locally sensitive hash function with constant weight, and/or, The anomaly parameter includes at least one of an index of a component causing an anomaly, a direction of deviation of the anomaly component, a magnitude of deviation of the anomaly component, and/or, The trust model is a trust entropy model comprising a trust score representing an absolute level of user trustworthiness, and a trust entropy representing a volatility of the trust score, and/or, The response of the user to the context-responsive challenge in the model update module includes analyzing whether a newly generated behavior state vector sequence of the user in completing the context-responsive challenge meets the user's personal behavior baseline model, and/or, The dynamic adjustment module dynamically adjusts at least one system operating parameter including increasing the frequency of generation of the behavior state vector and/or tightening the decision threshold for detecting anomalies when the trust score decreases or the trust entropy increases, and/or, The anomaly detection module, the challenge generation module, the model update module, and the dynamic adjustment module are all configured on the client device.
10. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the method of any of claims 1 to 7.

Description

Dynamic credibility maintaining method, system and storage medium Technical Field The present application relates to the field of computer security technologies, and in particular, to a method, a system, and a storage medium for maintaining dynamic credibility. Background With the popularity of tele-office, online education and examination, how to continuously verify the identity authenticity and behavior compliance of remote users has become a key technical challenge. In the prior art, schemes for continuous identity verification using behavioral characteristics have emerged. According to the scheme, behavior data such as mouse operations and touch screen gestures of a user are collected, characteristics of the behavior data are extracted to generate behavior characteristic fingerprints of the user, and a baseline model of personal behaviors is established. In the running process, the system continuously compares the real-time behavior characteristics of the user with the baseline model, and when the deviation exceeds a threshold value, the system judges that the user is abnormal, and can trigger an alarm or require the user to perform secondary identity verification. However, such prior art still suffers from the following drawbacks: 1. The contradiction between privacy and safety is that the system still needs to collect and transmit the behavior characteristic data of the user for comparison. These data, although not the original biometric, may indirectly reveal the user's behavioral habits and risk being intercepted and analyzed during transmission and storage; 2. Verification and exception decoupling when behavioral exceptions are detected, the verification tasks triggered by the system are typically preset, standardized, e.g., requiring the user to complete a face recognition. The verification has no context correlation with the specific abnormal behavior causing the verification, so that an attacker can easily predict the verification content and bypass the verification in the modes of prerecorded video or automatic script and the like; 3. The existing reliability evaluation model mostly adopts simple linear score addition and subtraction and fixed threshold judgment, so that accidental errors and continuous cheating attempts of users cannot be intelligently distinguished, the natural evolution of a user behavior mode is difficult to adapt, and the normal users are frequently disturbed or detected by high-definition cheaters in an evading manner. Disclosure of Invention The technical problems to be solved by the application are that the prior art has the defects of incomplete privacy protection, easy prediction and avoidance of verification links, and poor rigidity and adaptability of a system regulation mechanism in continuous behavior authentication. In order to solve the technical problems, the application provides a dynamic credibility maintenance method, which comprises the steps of constructing a user personal behavior baseline model, constructing a user personal behavior baseline model in advance based on historical behavior data of a user, generating a behavior state vector, periodically applying a fixed irreversible projection function to multi-mode raw data representing user behaviors or statistical features extracted from the multi-mode raw data on a client device to generate the behavior state vector, clearing the multi-mode raw data from the client device after the behavior state vector is generated, detecting an abnormality, comparing the behavior state vector with the user personal behavior baseline model to detect whether an abnormality exists or not, and locating at least one abnormal component causing the abnormality in the behavior state vector and an abnormal parameter of the abnormal component when the abnormality is detected, generating a context-reactive challenge according to the abnormal component and the abnormal parameter of the abnormal component, updating the model, representing the user credibility state according to the response of the user to the context-reactive challenge, and adjusting the behavior state of the system based on at least one of the dynamic state and the dynamic state of the trust state model, wherein the system is used for judging that the dynamic state is at least one of the dynamic state and the dynamic state is in the complex, and the dynamic state is adjusted based on the at least one of the dynamic state. Optionally, the irreversible projection function is a random initialization neural network or a locally sensitive hash function with constant weight. Optionally, the anomaly parameter comprises at least one of an index of a component causing an anomaly, a direction of departure of the anomaly component, a magnitude of departure of the anomaly component. Optionally, the trust model is a trust entropy model comprising a trust score representing an absolute level of user trustworthiness and a trust entropy representing the volatility of the trust s