Smart campus data management method and system

Abstract

The invention relates to a smart campus data management method and system. Comprising the following steps: modeling by adopting semantic consistency-oriented data behaviors; abstracting all system data as behavior event units; uniformly converting modeling results into semantic vectors, and writing the semantic vectors into a campus semantic behavior database; establishing an event standardization rule base; constructing a dynamic data flow scheduling engine by adopting a behavior link diagram; constructing a behavior path map according to the behavior event nodes, and extracting a high-frequency behavior link; a graph search algorithm is introduced to calculate a data scheduling path; a polymorphic perception and difference recognition mechanism of heterogeneous behavior states is adopted; capturing the deviation between the actual behavior event and the expected behavior link; constructing a behavior state comparison model, and identifying a potential abnormal state or a strategy failure risk; adopting a self-adaptive optimization and feedback updating mechanism of a strategy evolution graph; introducing a reinforcement learning algorithm, and carrying out strategy scoring and value return calculation on the strategy map; and adjusting and optimizing link parameters and resource scheduling logic, and notifying rhythm.

Description

Technical Field

[0001] The present invention relates to the technical field of smart campuses, and particularly relates to a smart campus data management method and system. Background Art

[0002] The existing intelligent campus data management methods and systems have promoted the construction of informatization and intelligent management in colleges and universities to a certain extent. In particular, initial achievements have been made in data integration, business collaboration, and information sharing. However, from the perspectives of the overall system architecture, data processing capabilities, behavior modeling accuracy, response flexibility, and system evolution mechanisms, there are still many deficiencies and obvious drawbacks, which restrict the development of intelligent campuses at a higher level. First, from the perspective of data management, most existing systems still remain at the primary integration level of "system docking + database aggregation", lacking a unified data semantic standard and structure mapping mechanism. The data formats, interface protocols, and update cycles of different subsystems, such as the academic affairs system, access control system, teaching evaluation system, library system, and OA platform, vary greatly, making it difficult to achieve high-quality data fusion. Even if the physical layer is connected, it is often difficult to interact at the logical layer due to the lack of semantic consistency of the data. Eventually, it is manifested as high data redundancy, serious repeated calls, weak context correlation, and the system analysis results are difficult to support accurate decision-making. Second, in terms of data scheduling, existing systems generally adopt a static interface call strategy, that is, data is pulled between systems through fixed API connections or data middleware transfers. The data call process cannot be optimized, the scheduling path cannot be selected, and the call granularity cannot be controlled. Once the system scale expands or the call request volume surges, it is extremely easy to cause performance bottlenecks and data jams. Moreover, when it comes to highly sensitive data such as grades, teaching evaluations, and personal privacy, security access control often remains at the level of permission switches and role tags, unable to achieve dynamic fine-grained evaluation according to paths or scenarios, and there are two-way problems of "over-authorization" or "insufficient permissions", seriously affecting the security and adaptability of the system. In terms of behavior modeling, most existing intelligent campus systems use single-event records or transactional data structures for behavior analysis. For example, recording whether a student signs in, evaluates teaching, or logs in, while ignoring the temporal, frequency, intensity, semantic relationships between behavior intentions and behavior objects behind the behavior. It is difficult for the system to construct a complete user behavior link or path, and it is impossible to understand the true state of students based on the "behavior process", and can only make judgments on the "behavior results", thus unable to achieve high-order behavior intelligence tasks such as trend prediction, dynamic profiling, or path recommendation. In addition, in terms of user anomaly detection, although some systems have introduced behavior anomaly recognition algorithms, they often only rely on single-variable threshold triggers, such as "not signed in for three consecutive days" and "abnormal decline in grades", which lack context and do not consider the interaction between behaviors, and cannot effectively identify potential risks, and are also prone to false alarms or missed alarms. In addition, most current intelligent campus systems lack a dynamic evolution mechanism for behavior data, that is, the system lacks self-learning ability and cannot accumulate experience, optimize paths, and improve the accuracy of strategies during operation, making the original intelligent system degenerate into an information aggregation platform with "fixed processes and static rules", deviating from the core goal of truly behavior-driven data intelligence.There are also many challenges at the technical level. For example, the data consistency guarantee mechanism is weak under high concurrency of multiple systems, the long-term scheduling delay control of the data link is imperfect, there are separation problems in the modeling of individual and group behaviors, and there is a lack of causal modeling ability between behavioral data and management responses. These problems are very likely to be magnified during the actual operation of the system. For example, during key periods such as centralized course selection at the beginning of the semester, grade query during peak periods, and teaching evaluation collection at the end of the semester, the system has problems of "high latency, low hit rate, and weak strategy" in terms of scheduling, discrimination, and response, ultimately affecting the accuracy, timeliness, and stability of the smart campus service. Summary of the Invention

[0003] The purpose of the present invention is to provide a smart campus data management method and system, so as to solve some of the drawbacks and deficiencies pointed out in the background technology.

[0004] The present invention adopts the following technical solutions to solve its above technical problems: A smart campus data management method and system, including: S1. Adopt data behavior modeling oriented to semantic consistency:

[0005] S1.1. Abstract all system data into behavioral event units; use the spatio-temporal behavior graph model to express the dynamic connections between entities;

[0006] S1.2. Uniformly convert the modeling results into semantic vectors and write them into the campus semantic behavior database; and establish an event standardization rule library to enable event recognition, fusion, and alignment;

[0007] S2. Adopt the construction of a dynamic data flow scheduling engine for the behavior link graph:

[0008] S2.1. Construct a behavior path graph according to the behavioral event nodes, and extract high-frequency behavior links including course selection → entering the classroom → after-class teaching evaluation → grade assessment; each link is assigned a flow cost and a security weight;

[0009] S2.2. Introduce a graph search algorithm to calculate the data scheduling path; activate the link data set as needed during operation to enable cross-system data linkage call and minimum redundant transfer;

[0010] S3. Adopt a polymorphic perception and difference recognition mechanism for heterogeneous behavior states:

[0011] S3.1. Capture the deviation between the actual behavioral event and the expected behavior link; conduct multi-dimensional analysis of the behavioral deviation including frequency, intensity, and associated objects, and introduce a behavioral state difference function;

[0012] S3.2. Construct a behavioral state comparison model to identify potential abnormal states or risks of strategy failure; and the output difference data is used for system adjustment, user prompt, and strategy reconstruction purposes;

[0013] S4. Adaptive optimization and feedback update mechanism using the strategy evolution graph:

[0014] S4.1. Each data scheduling and behavior response is written into the strategy evolution graph as a strategy execution event; collect user behavior feedback including changes in scores after intervention and increased resource click-through rates, and construct a causal path;

[0015] S4.2. Introduce a reinforcement learning algorithm to calculate strategy scores and value returns on the strategy graph;

[0016] S4.3. Optimize link parameters, resource scheduling logic, and notification rhythms to form a closed loop of behavior data → management response → effect evaluation → strategy evolution.

[0017] Furthermore, the method for constructing the dynamic data flow scheduling engine of the behavior link graph includes:

[0018] Abstract the original data of subsystems from different functions including the educational administration system, access control system, and teaching evaluation platform into event units of behavior characteristics; each behavior event encapsulates user behavior intentions, occurrence times, behavior locations, and system interaction states; through the abstraction method, data from different sources and different formats have a unified expression basis at the semantic layer;

[0019] Construct a multi-dimensional behavior graph with behavior events, where the nodes in the graph represent behavior events and the edges represent logical, temporal, or causal connections between events; based on the behavior graph, evaluate the scheduling overhead of each candidate path; introduce a link cost function to quantify the communication complexity, processing delay, and resource occupancy between modules during data calls. The function is defined as follows:

[0020]

[0021] Where:

[0022] Φ(l ij ) represents the value of the link scheduling cost function from event node e i to e j ; r ij represents the system resource consumption weight value, reflecting the system resources required to transmit and process data on this link; c ij represents the complexity level of the data structure in the link; t ij is the average delay time value recorded for this link in historical access; log(t ij +2) is used to compress the delay impact, and +2 is to avoid taking the logarithm of 0, which is invalid.

[0023] Furthermore, the method for constructing the dynamic data flow scheduling engine of the behavior link graph includes:

[0024] Further evaluate the security risks of data usage on each path, considering the user data permissions, information sensitivity, and historical call behaviors involved; introduce a security weight function, expressed as follows:

[0025]

[0026] Where:

[0027] Λ(l ij ) represents the security risk weight value on this link, and the higher the value, the higher the security requirement; s ij is the data sensitivity level involved in the link, and the larger the value, the more sensitive the data; u ij represents the permission level matching degree of the current user under the data type, including whether they have viewing or modification permissions; d ij is the access frequency of the link data in history; α, β are risk adjustment coefficients set in the system, used to adjust the proportion of sensitivity and permission risk in the evaluation as needed.

[0028] Furthermore, the method for constructing the dynamic data flow scheduling engine of the behavior link graph includes:

[0029] When the system receives a data scheduling request, map the request target to the termination node in the behavior path graph, and through the path search engine, comprehensively score the cost and risk of all candidate paths, and select the optimal path; introduce a path evaluation function, used to comprehensively evaluate the values of two functions and score the entire behavior path P, defined as follows:

[0030]

[0031] Where:

[0032] Θ(P) is the total evaluation score of path P, that is, the comprehensive cost of the path; Φ(l ij ) is the cost function value of each hop link in the path; Λ(l ij ) is the security weight value of each hop link in the path; γ is the security offset factor set by the system, used to balance the proportion between performance and security: when γ increases, the system pays more attention to security; when γ decreases, it pays more attention to efficiency.

[0033] Furthermore, the method for constructing the polymorphic perception and difference recognition mechanism of heterogeneous behavior states includes:

[0034] By continuously monitoring the behaviors of campus users, convert the operation records of users in different systems into standardized behavior event sequences; each event is based on a timestamp and is accompanied by behavior type, behavior object, scene source, and operation depth information; build an expected behavior link library based on historical behavior data according to user categories, time scenarios, and teaching cycles,

[0035] The link library describes the typical behavior paths that certain types of users should present and is used as a reference model for real-time behavior comparison;

[0036] After receiving a real-time behavior event, the current behavior trajectory is compared with the corresponding expected link, and the cumulative degree of behavior deviation is calculated in an integral manner; A deviation integral function is introduced to measure the degree of deviation of the overall user behavior trend, which is defined as follows:

[0037]

[0038] Where:

[0039] Ψ(u,T) represents the overall behavior deviation intensity of user u within the time window T; A u (t) is the actual behavior intensity function of the user at time t; E u (t) is the expected behavior intensity function of the same user at time t, which is generated by link library modeling; δ is the behavior deviation index, and its value range is δ>1, which is used to amplify the influence of persistent or significant behavior deviations. T is the behavior observation period, which is 1 day, 1 week, or the length of the teaching task cycle.

[0040] Furthermore, the method for constructing the polymorphic perception and difference recognition mechanism for heterogeneous behavior states includes:

[0041] Multi-dimensional feature extraction is performed on the behaviors involved in the deviation event, including frequency change, intensity fluctuation, and resource transfer information; all behavior features are mapped into standard vectors and compared dimension by dimension with the historical state vector to construct a behavior state deviation model; The following behavior state difference function is introduced:

[0042]

[0043] Where:

[0044] Ω(u) is the comprehensive deviation score of the current behavior state of user u; is the i-th behavior feature value currently collected, including the number of course clicks and reading duration; is the expected value of the i-th feature in the user's history, which is extracted from historical behaviors; η i is the degree of fluctuation of the i-th feature in historical data, which is used to avoid over-sensitive reactions to natural fluctuations; n is the total number of feature dimensions.

[0045] Furthermore, the method for constructing the polymorphic perception and difference recognition mechanism for heterogeneous behavior states includes:

[0046] Based on multiple user behavior deviation scores, through clustering modeling, outlier detection, and time-series evolution analysis, individual behavior anomalies and strategic overall failures are identified; when a large range of deviations are concentrated on a certain policy path, the system needs to determine whether it is a policy design failure, and simulate a slight policy adjustment and observe the sensitivity response of the behavior state, and define the following policy perturbation response function:

[0047]

[0048] Where:

[0049] Σ(P) is the perturbation response sensitivity index of behavior path P; ΔΩ(u) represents the change in the state difference value of user u after policy adjustment; ΔR(P,h) represents the slight policy perturbation operation implemented on path P, including fine-tuning the entrance layout and delaying the intervention prompt time; h is the perturbation amplitude factor.

[0050] The intelligent campus data management method and system provided by the present invention have the following beneficial effects: First, it realizes the unified abstraction and semantic consistency modeling of multi-source heterogeneous data in the campus (such as teaching affairs systems, access control systems, teaching evaluation platforms, library systems, etc.), integrates the originally scattered and differently formatted data into structured behavior event units, and through the construction of spatio-temporal behavior maps and behavior link diagrams, realizes the logical, temporal, and causal associations between data, laying a foundation for subsequent intelligent scheduling and behavior analysis; Second, it proposes a link cost function and a security weight function to finely evaluate the resource consumption and security risks of each cross-system data scheduling path, and realizes a controllable optimal balance between performance and security through the path evaluation function, significantly improving the efficiency and controllability of system data calls, and reducing redundant access and risk exposure; Third, it innovatively constructs a multi-state perception and difference recognition mechanism for heterogeneous behavior states. The system can not only identify the deviation degree of single-user behavior (such as a decrease in login frequency, a reduction in learning duration, etc.), but also perform group behavior clustering and policy-level deviation recognition to timely discover potential policy design failure problems; Fourth, the system introduces a policy perturbation response function, which can accurately measure the sensitivity of user behavior response by slightly adjusting the policy path layout, rhythm, or trigger point without interrupting the original service process, so as to realize the self-learning, self-adjustment, and self-optimization of the policy; Fifth, the system forms a full-process closed-loop mechanism of "behavior data → state modeling → intelligent scheduling → policy optimization", with high real-time performance, high interpretability, and high adaptability, significantly improving the intelligent level of data management in the intelligent campus environment, promoting the transformation of the campus information system from static integration to dynamic perception, prediction-driven, and policy evolution, and providing strong technical support for colleges and universities to achieve precise teaching, personalized services, and risk warning. Description of the Drawings

[0051] Figure 1 This is a flowchart of a data management method and system for a smart campus according to the present invention.

[0052] Figure 2 This is a flowchart of a method for constructing a dynamic data flow scheduling engine for the behavior link diagram according to the present invention.

[0053] Figure 3 This is a flowchart of a method for constructing a polymorphic perception and difference recognition mechanism for heterogeneous behavior states according to the present invention. Detailed implementation manners

[0054] The following will give a detailed description of the specific implementation manners of the present invention in conjunction with the accompanying drawings.

[0055] In conjunction with the attached Figure 1 , for a data management method and system for a smart campus according to the present invention, the core lies in realizing the intelligent fusion and dynamic understanding of cross-system data. S1 adopts data behavior modeling oriented to semantic consistency, which specifically includes the following key operations: S1.1 The system uniformly abstracts the structured or semi-structured data from multiple heterogeneous subsystems such as the educational administration system, access control system, library system, and teaching platform, and does not store it in the traditional "data record" manner, but abstracts it into "behavior event units". Each event unit consists of five elements, including the behavior subject (such as students, teachers), behavior actions (such as course selection, borrowing books, signing in), occurrence time, occurrence location, and object of action (such as courses, classrooms, etc.). Then, a dynamic connection between entities is constructed using the "spatio-temporal behavior graph model". Each node in the model represents a behavior event, and the edges between the nodes represent the temporal sequence, spatial position relationship, or logical dependence of the behaviors, forming a computable, traceable, and analyzable behavior link graph; S1.2 The system further performs semantic conversion on the node and edge information in the constructed behavior graph above, converting it into the form of "semantic vectors" to support subsequent semantic calculations, similarity analysis, and policy reasoning. All semantic vectors are uniformly written into the "campus semantic behavior database" for centralized management. This database supports semantic access, semantic combination, and link backtracking of behavior events. At the same time, the system establishes a set of "event standardization rule libraries" to identify the boundaries and structures of various events, fuse and uniformly name heterogeneous behaviors from different systems, solve the problems of inconsistent naming, non-uniform granularity, and unclear behavior semantics between different systems, and realize the identification, cross-domain alignment, and behavior fusion between events, thereby laying a semantic consistency foundation for subsequent data scheduling, intelligent analysis, and behavior prediction.

[0056] S2 is constructed with a dynamic data flow scheduling engine based on the behavior link diagram, aiming to achieve intelligent, efficient, and low-redundancy invocation of cross-system data through structured modeling and dynamic calculation of the user behavior path. In S2.1, the system first constructs a "behavior path map" based on the modeled behavior event nodes. This map is a directed graph structure centered on user behavior, with event time sequences and logical relationships as edges. Through mining and frequency analysis of historical behavior data, the system extracts typical and representative high-frequency behavior links, such as educational activity links like "course selection → enter the classroom → post-class teaching evaluation → grade assessment". In the map, each link not only represents the natural order of behavior occurrence but also comes with two key attributes: one is the "flow cost", which measures the resource overhead and processing delay of data transmission, access, and processing across systems in this path; the other is the "security weight", which is used to measure the sensitivity of the data involved in the path and the access permission requirements. The system scores and ranks all links based on these two dimensions to provide parameter support for scheduling optimization; in S2.2, the system further introduces a heuristic graph search algorithm, transforming the data scheduling problem into an optimal weighted path search problem. By calculating the costs and security risks of all feasible paths, it finds the optimal data invocation path. In actual operation, the system does not mobilize all the data of the entire link at once but adopts an "activation-on-demand mechanism", that is, according to the current business requirements and user context, only activates the necessary event nodes and their related data in the link to achieve cross-system data linkage invocation and minimum redundant data flow, avoiding unnecessary data access and system overhead, thus achieving the goal of intelligent campus data invocation with high efficiency, accurate access, and security controllability.

[0057] S3 adopts a polymorphic perception and difference recognition mechanism for heterogeneous behavior states. The core purpose is to enable the system to have the ability to perceive and intelligently recognize the changes in behavior states of different users in different situations in real time, so as to promote the dynamic optimization and precise intervention of management strategies. S3.1 The system is based on an expected behavior link library built from behavior event streams and historical models, continuously comparing the actual behavior sequence of the user with the typical behavior path it should present. By calculating the deviations in time, content, and structure between behavior trajectories, "behavior deviation points" are identified, and the multi-dimensional characteristics of the deviations are quantitatively analyzed, including frequency deviation (whether the number of times the behavior appears decreases or surges), intensity deviation (whether the degree of behavior participation or the quality of completion decreases), and associated object deviation (whether there is a transfer of behavior goals or a mutation in resource access). The system constructs a "behavior state difference function" based on these characteristics. This function compares the current behavior state vector with the historical behavior state vector dimension by dimension and outputs a behavior state difference score, which is used to measure the severity and risk level of behavior deviation; S3.2 The system uses the state difference scores of multiple users to construct a "behavior state comparison model". This model combines group behavior trends, individual behavior stability, and behavior link context for clustering, classification, and anomaly recognition, so as to distinguish whether it belongs to individual occasional anomalies or systematic policy failure problems. If the system identifies that a certain behavior path has a high deviation synchronously among a large number of users and is highly sensitive to fine-tuning of system policies, it is determined that there is a risk of insufficient policy design or content mismatch for this path. The system will output a "difference data packet" on this basis, including content such as behavior deviation type, deviation intensity, associated path, and recommended response measures, as feedback input to the system, driving the adjustment of service policy parameters, generating user-side prompt information, or providing intervention suggestions for managers, so as to achieve an intelligent closed-loop management mechanism of system perception - recognition - response - reconstruction.

[0058] S4 adopts an adaptive optimization and feedback update mechanism for the policy evolution graph, aiming to endow the system with the ability of policy self-evolution. By driving policy updates and optimizing management methods through behavioral data, it improves the intelligent decision-making level and dynamic adaptability of the system. S4.1 The system records each data scheduling, system response, and management intervention operation triggered by user behavior as a "policy execution event" in the "policy evolution graph". This graph is a dynamically updatable causal network used to track the entire process path of "behavior - response - feedback". For example, after the system pushes academic reminders for a student's consecutive absenteeism, feedback data such as whether the student returns to school on time, whether the course grades improve, and whether resource usage rebounds will be continuously collected and written into the graph, forming a closed-loop link from the intervention action to the effect feedback. Each node in this graph represents a policy or feedback state, and each edge records the causal logical relationship between policy execution and results; S4.2 The system introduces a reinforcement learning algorithm based on the policy evolution graph to calculate policy scores and value returns for various policy paths. The system adjusts the effectiveness evaluation of policies according to the actual feedback results, evaluates the application effects of different policy combinations in different user groups through the policy benefit function, assigns a dynamically changing value score to each policy path, strengthens the policies that bring positive effects, and eliminates or adjusts the policy paths with weakened or even negative effects; S4.3 The system adaptively optimizes the entire behavior response system based on the policy scoring results, including optimizing the parameter weights of the behavior link (such as assigning a higher monitoring priority to a certain behavior), adjusting the resource scheduling logic (such as reordering the recommendation priorities of learning resources), and optimizing the notification trigger rhythm (such as intelligently pushing information according to user response habits), thus forming a systematic intelligent closed-loop of "behavior data → management response → effect evaluation → policy evolution", enabling the intelligent campus data management system to have the ability of continuous learning, real-time adaptation, and self-update, and significantly improving the personalization, precision, and dynamic response ability of educational management.

[0059] Embodiment 1

[0060] Combined with the attached Figure 2, in this embodiment, a certain university is running a new intelligent campus data management system based on this invention. The system integrates multiple subsystems such as the academic affairs system, access control system, and teaching evaluation platform, and uses "behavior events" as the core data unit to drive data scheduling optimization. Student Xiaoming is a junior student. On Wednesday morning of the tenth teaching week, he left the dormitory and entered the teaching building, swiped the access control to enter Classroom 504, and attended the "Data Structure" course. Subsequently, the system pushed a teaching evaluation reminder to him after class, and this teaching evaluation data became part of the course teaching quality analysis in the following week. The system abstracts this series of user behaviors into the following behavior event units: e1 is "selecting the course 'Data Structure'", e2 is "swiping the card to enter the teaching building", e3 is "signing in for class", e4 is "submitting the teaching evaluation after class", and e5 is the data preparation before "grading"; among them, the system constructs these events into a link in the behavior path graph: e1→e2→e3→e4→e5, and each segment involves cross-system data scheduling and interaction, such as from the course selection system to the access control system, and from the sign-in module to the teaching evaluation platform. For the link e2→e3 (that is, from "entering the teaching building" to "signing in for class"), the system needs to call the access control data and the sign-in system data for association to verify whether the student enters the classroom of the selected course on time. The resource consumption value r 23 mainly consists of two service nodes, including the access control data parsing module and the sign-in verification module, and the resource evaluation is r 23 = 3.2; at the same time, the data structure complexity involved in this link is c 23 = 9.0 (because the sign-in data contains five types of fields such as timestamp, GPS location, device ID, and course association and the nested structure is relatively deep); and the average access delay time of this link in the previous two weeks is t 23 = 1.5 seconds. Substitute the above values into the link cost function defined in this invention:

[0061]

[0062] According to the system evaluation standard, when the value of the link cost function Φ is greater than 10, it means that the call of this link has a relatively high impact on the system performance. At this time, the system scheduling engine will trigger the "optimization suggestion mechanism", such as trying to pre-load the access control data cache in advance, or transforming the sign-in verification module for parallel processing to reduce the r value and thus optimize the Φ value. Another link is e4→e5 (submitting the teaching evaluation after class to grading), which usually needs to call the teaching evaluation platform results, course numbers, teacher information, and historical teaching quality baseline data for comprehensive processing. The resource weight r 45 = 4.5, the data structure complexity c 45 = 4.0 (the data structure is relatively simple), and the average response delay of the teaching evaluation system interface is t 45 = 3.0 seconds. Substitute into the formula:

[0063]

[0064] This Φ value is significantly higher than the threshold, indicating that this cross-system call chain exerts a relatively large pressure on the data scheduling engine. In response to this situation, the system can intelligently select the "delayed non-high-frequency strategy" and schedule such teaching evaluation-related data scheduling during the low-peak data flow period with strong non-real-time performance, thereby reducing the risk of system resource congestion. In actual deployment, the system dynamically monitors and sorts the Φ values of different links, preferentially executes the path combination with the lowest Φ value, ensures that critical and highly real-time links are scheduled first, while links with high delay tolerance are processed later, so as to achieve the intelligent scheduling of the overall campus data flow. In an actual project, the system sets the range of the r value to 1.0 to 5.0, corresponding to lightweight cache-type calls to multi-level dependency-type computing calls; the c value range is generally 1 to 12, reflecting the differences in data structure complexity; the t value comes from historical interface performance monitoring and is usually between 0.5 and 5 seconds.

[0065] To further illustrate the part of the dynamic data flow scheduling engine of the behavior link graph proposed by the present invention regarding the assessment of data usage security risks, specifically, by introducing the security weight function Λ(l ij ) to quantitatively manage the call security of each link. In this system, in the student behavior event chain e1→e2→e3→e4→e5, each segment of the link involves data access and calls, and the data access risk no longer depends on fixed access control rules, but dynamically evaluates the sensitivity of the link, user permissions, and historical access patterns. Hereinafter, the link e4→e5 (from after-class teaching evaluation to grade assessment) will be selected as an example. This link needs to read data such as student teaching evaluation scores, course IDs, identities of teaching teachers, course content, and historical grade mapping relationships during actual operation. Some of the data, such as "teacher teaching evaluation scores", belong to evaluation-type privacy data, so their sensitivity levels are relatively high. According to the system's classification definition of data sensitivity levels: 1 represents low sensitivity (such as course names), 2 represents medium sensitivity (such as attendance data), and 3 represents high sensitivity (such as grades and teaching evaluations). The data sensitivity level involved in this link is recorded as s 45 = 3. In the current scheduling request, the system teaching analysis engine initiates the operation, rather than a single user. Therefore, the permission level is set to the highest level to match u 45 = 5, representing having viewing and analysis permissions. If it is an ordinary teacher, it is generally set to 3. According to the historical logs, the data of this link is called frequently in the system. It was called 31 times in the past week. To prevent excessive concentration and exposure of the data, the system records its access frequency index as d 45 = 31. The security weight function of the present invention is defined as follows:

[0066]

[0067] Substituting the data, we have α = 0.7 and β = 0.3. This coefficient is configured by the system administrator according to the system operation mode. Among them, the general setting range of α is 0.5 - 0.8, representing the degree of emphasis on the sensitivity level of the data itself. The general setting range of β is 0.2 - 0.5, which is used to control the influence degree of the caller's permission and historical access behavior. The sum of these two parameters is 1, which is convenient for standardized control. Substituting into the calculation, we can get:

[0068]

[0069] The calculation result Λ(l 45 ) ≈ 0.1766, indicating that the security risk of this link is relatively low. The main reason is that although the data sensitivity level is relatively high, the system caller's permission is completely matched, and the access frequency is high but controllable. Therefore, the system evaluates that this link can be normally called within the operation cycle without triggering the security delay mechanism. In contrast, if an ordinary student tries to access the e4→e5 link, that is, the permission level u 45 = 1, the system will recalculate the security weight:

[0070]

[0071] Although it doesn't decrease much compared with the system-level permission request, in the system setting, the access to this link by ordinary students defaults to trigger the permission verification logic. Even if the security weight is not high, it will be blocked due to policy restrictions, which reflects that the function of the security weight function is used for risk assessment rather than direct authorization decision. Another example is the link e2→e3 (access to the classroom through the access control → class sign-in). The data involves real-time location and course arrangement, and the sensitivity level is set as s 23 = 2, the caller is the user himself / herself, the permission level u 23 = 3, and the historical access frequency d 23 = 5. Substituting into the calculation:

[0072]

[0073] The security weight of this link is relatively slightly higher. The system will give priority to applying the cache mechanism in the scheduling to avoid unnecessary security risk control pressure caused by repeated calls of real-time data. Thus, it shows that this function can not only be used for path priority sorting but also dynamically guide the selection of the scheduling mode, such as whether to enable caching, whether to perform segmented scheduling, and whether to trigger user confirmation.

[0074] The entire process of student Xiaoming's participation in the "Data Structure" course is recorded by the system as a complete behavior chain. The event nodes are e1 (course selection), e2 (access to the teaching building through the entrance guard), e3 (class sign-in), e4 (post-class teaching evaluation), and e5 (grade assessment). The behavior path P = {e1→e2, e2→e3, e3→e4, e4→e5} is constructed as a cross-system data scheduling path. When the system receives a scheduling request, such as when the teaching quality analysis engine requests a path to obtain the data link of "the impact of course teaching evaluation scores on the final grades", the system needs to select the optimal path among all feasible paths to achieve efficient and secure data scheduling. At this time, it is necessary to introduce a path evaluation function Θ(P) for comprehensive scoring. First, the system calls the link cost function Φ(l ij ) and the security weight function Λ(l ij ). As calculated previously, the cost value of the link e2→e3 is Φ(l 23 ) ≈ 10.57, and the security weight value Λ(l 23 ) ≈ 0.2675. The cost of the link e4→e5 is Φ(l 45 ) ≈ 31.84, and the security weight is Λ(l 45 ) ≈ 0.1766. Now, after the system receives the scheduling request, it maps the path target to the e5 node, lists the candidate paths (the main path P1 in this example) through the path search engine, and substitutes them into the following path evaluation function for evaluation:

[0075]

[0076] In this system, γ is the security offset factor, which is used to dynamically adjust the emphasis on performance and security under different operating strategies. Before the exam week, the system favors stability and takes a higher value; during daily operations, it favors efficiency and takes a lower value. It is currently set to the normal teaching week, γ = 5.0, and the value range is set by the administrator to be 1.0 - 10.0, and it is dynamically adjusted according to the campus security level policy. The following are the evaluation values of each link in the set behavior path P1 (partial approximate values):

[0077] e1→e2: Φ(l 12 ) = 5.2, Λ(l 12 ) = 0.12

[0078] e2→e3: Φ(l 23 ) = 10.57, Λ(l 23 ) = 0.2675

[0079] e3→e4: Φ(l 34 ) = 7.6, Λ(l 34 ) = 0.19

[0080] e4→e5: Φ(l 45) = 31.84, Λ(l 45 ) = 0.1766

[0081] Substitute the above values into the path evaluation function for calculation:

[0082] Θ(P1) = (5.2 + 5·0.12) + (10.57 + 5·0.2675) + (7.6 + 5·0.19) + (31.84 + 5·0.1766)

[0083] Θ(P1) = (5.2 + 0.6) + (10.57 + 1.3375) + (7.6 + 0.95) + (31.84 + 0.883)

[0084] Θ(P1) = 5.8 + 11.9075 + 8.55 + 32.723 ≈ 58.98

[0085] The total evaluation score of this path is Θ(P1) ≈ 58.98. The system compares and sorts this score with other candidate paths, and selects this path as the optimal path for current data scheduling according to the principle of the lowest score; if there is a secondary path P2 with the structure e1 → e6 → e7 → e5, which calls the historical analysis cache module instead of the real-time check-in system, and its Φ value is lower overall but the security is slightly weaker, and Θ(P2) is set to be approximately 62.15. Since Θ(P1) < Θ(P2), the system still selects the P1 path to execute; and if the current period is switched to the final exam stage and γ is set to 8.0, the system's preference for security is increased, and at this time, recalculate:

[0086] e2 → e3: Φ = 10.57, Λ = 0.2675, and the security weighted part becomes 8·0.2675 = 2.14

[0087] e4 → e5: Φ = 31.84, Λ = 0.1766, and the security weighting is 8·0.1766 = 1.41

[0088] The total score becomes:

[0089] Θ(P1)' ≈ 5.2 + 0.96 + 10.57 + 2.14 + 7.6 + 1.52 + 31.84 + 1.41 ≈ 61.24

[0090] The security risk of the secondary path P2 is relatively high, and its security weight accumulates to 3.1, corresponding to a high score in the high-γ case. The total Θ(P2)'≈65.8. At this time, P1 still wins, but the system will remind the administrator that this path is at the edge of risk in the high-security scenario, and it is advisable to consider improving the security fault tolerance of key links. From the above examples, it can be seen that the path evaluation function Θ(P) in the present invention can achieve a comprehensive trade-off between the performance and security of data call paths, enabling the smart campus system to achieve flexible and intelligent path selection under different operation strategies. At the same time, the value ranges of various parameters in the system (such as γ = 1.0 - 10.0, r = 1.0 - 5.0, s = 0 - 3) support dynamic configuration, improving the practical applicability and popularization value of the method, and meeting the requirements of both data management efficiency and security in the real campus environment.

[0091] Embodiment 2

[0092] Combined with the attached Figure 3 , in this embodiment, it is set that in the smart campus application scenario of continuing Xiaoming's participation in the "Data Structure" course, the system realizes real-time tracking and deviation recognition of his learning behavior in different subsystems by constructing a polymorphic perception and difference recognition mechanism for heterogeneous behavior states, so as to assist in teaching intervention and strategy optimization. Xiaoming, as a junior student in the School of Computer Science, his normal behavior paths in the system include logging in to the online teaching platform five times a week, signing in once a week for each course, borrowing an average of one relevant book every two weeks, completing the teaching evaluation within 24 hours after class, and maintaining a stable mid-upper level in the class during the grading period. Based on historical big data, the system constructs an expected behavior link library for Xiaoming, a "mid- to high-performance student who attaches importance to courses", which includes nodes such as course learning, book borrowing, signing in, and teaching evaluation, as well as their time rules and behavior intensity patterns. The system uses these links as reference models for real-time behavior comparison. In the eleventh week, the system monitors and finds that Xiaoming only logged in to the platform once this week, had zero book borrowing, was late for signing in twice, and did not complete the teaching evaluation after the "Data Structure" course. His behavior was significantly abnormal; the system immediately aligned and analyzed his current behavior trajectory with the expected behavior link to generate the actual behavior intensity function A u (t) and the expected behavior intensity function E u (t), and uses the following behavior deviation integral function for quantitative evaluation:

[0093]

[0094] Among them, Ψ(u,T) represents the overall behavior deviation intensity of Xiaoming within a one-week observation window; A u (t) is the actual behavior intensity of Xiaoming at time t (such as the number of course accesses per minute, signing-in status, borrowing behavior, etc.); E u(t) is the expected behavior model generated according to its portrait type; δ is the deviation index, configured according to the behavior stability type in the system, usually 1.5 - 3.0. In this example, Xiaoming's past behaviors are highly regular, so δ = 2.5 is set to amplify the impact of his abnormal behaviors on the results; the observation period T is 7 days (the teaching week cycle). Taking the platform login behavior as an example, the expected login behavior E u (t) is 5 times a week, distributed from Monday to Friday, with an average behavior intensity of 1.0 each time. The actual behavior A u (t) only appears on Monday, with an intensity of 0.8, and the rest are 0. The system discretely samples and integrates every 24 hours. The integrated result of the login behavior deviation is:

[0095]

[0096] In the teaching evaluation behavior, Xiaoming did not complete the teaching evaluation for the "Data Structure" course in the original plan. The expected intensity E u (t) is 1.0 (the system sets that completing it within 24 hours is a full behavior), and the actual is 0.0. The integrated deviation is |0 - 1| 2.5 = 1. In the sign - in behavior, there are latenesses on Wednesday and Thursday (the behavior intensity is set to 0.5), and the rest are on time (the behavior intensity is 1.0), while the expectation is an intensity of 1.0 per day. The integrated calculation is:

[0097]

[0098] The expected number of books borrowed is 1, but the actual is 0. The integrated deviation is 1. The integrated deviation value of Xiaoming's behaviors this week is:

[0099] Ψ total = Ψ login + Ψ sign + Ψ review + Ψ library = 4.01 + 0.352 + 1 + 1 = 6.362

[0100] The system sets the threshold at 5.0. When Ψ(u,T) > 5.0, it is considered that the user's behavior state deviates from the expected pattern, and policy feedback or intervention prompts are required. Therefore, the system marks Xiaoming as "moderate behavior deviation" and pushes a personalized prompt: "Your participation in courses this week has decreased. It is recommended to complete the teaching evaluation in a timely manner and review the course content." At the same time, this difference value will be synchronized to the teaching behavior analysis module on the teacher side to prompt the teacher to pay attention to the changes in the learning state of this student and provide one-on-one tutoring if necessary. Through this mechanism, the system not only realizes behavior perception across platforms and behavior types, but also achieves unified quantification of multi-dimensional deviations through the integral function Ψ(u,T). The setting range of δ is 1.5 - 3.0, and the system can adjust parameters according to the student behavior stability type. The behavior observation period T can be selected as 1 day, 3 days, 7 days, or the entire teaching task cycle length, with extremely high adaptability, flexibility, and practical promotion value.

[0101] After the system identifies that there is a moderate deviation in the overall behavior trend through the integral function Ψ(u,T), it further enters the next stage of the difference identification mechanism, that is, to extract and refine the multi-dimensional characteristics of the behavior state to quantify the specific deviation degree in different behavior dimensions and provide a basis for subsequent policy recommendations. Through the multi-state perception mechanism of heterogeneous behavior states, the system extracts 5 main behavior characteristics from Xiaoming's data in the past two weeks, namely f1: the number of clicks on online courses, f2: the total video learning duration (minutes), f3: the number of borrowed professional books, f4: the classroom attendance rate, f5: the submission rate of post-course teaching evaluation. These characteristics come from heterogeneous subsystems such as the academic affairs system, the library system, the learning platform, and the teaching evaluation platform, and are uniformly transformed into structured behavior feature vectors within the system. According to historical records, Xiaoming's expected behavior values are: And during the current monitoring period, his actual behavior is: At the same time, the system calculates the volatility index based on his historical behavior floating data: η1 = 5, η2 = 60, η3 = 0.3, η4 = 0.1, η5 = 0.05. These η i values represent the natural fluctuations of the user's past behavior in each dimension, avoiding the system from overreacting to normal fluctuations. The system substitutes the above data into the behavior state difference function for calculation. The function is defined as:

[0102]

[0103] Substitute each item of data. Let the total number of feature dimensions n = 5, and calculate item by item as follows:

[0104] The first item (number of clicks): (10 - 25) 2 / (5 + 1)2 = 225 / 36 = 6.25

[0105] Item 2 (Learning duration): (120 - 300) 2 / (60 + 1) 2 = 32400 / 3721 ≈ 8.71

[0106] Item 3 (Books borrowed): (0 - 1) 2 / (0.3 + 1) 2 = 1 / 1.69 ≈ 0.59

[0107] Item 4 (Sign-in rate): (0.6 - 1) 2 / (0.1 + 1) 2 = 0.16 / 1.21 ≈ 0.132

[0108] Item 5 (Teaching evaluation rate): (0 - 1) 2 / (0.05 + 1) 2 = 1 / 1.1025 ≈ 0.907

[0109] The sum is divided by 5 to obtain:

[0110]

[0111] This value represents the comprehensive deviation score of Xiaoming's current behavior state. The system presets the following difference threshold levels: Ω < 1.5 is a mild deviation, 1.5 - 3.0 is a moderate deviation, and Ω > 3.0 is a severe deviation. Therefore, Xiaoming's current score Ω(u) ≈ 3.32 has reached the "severe behavior state deviation" level, and the system immediately pushes the following response strategies: On the one hand, it sends a warning report to the counselor, including that Xiaoming's key deviation features are a 60% decrease in click-through rate, a 180-minute reduction in learning time, and no teaching evaluation at all. At the same time, it is recommended to pay attention to whether there are factors such as learning pressure and lack of interest in courses; on the other hand, the system synchronously generates personalized pushes, such as "The relevant materials on 'Data Structure' have been updated this week. It is recommended to view the supporting videos for Chapter 4", to improve his behavior recovery rate through gentle intervention; in addition, the behavior state score will be used as a training sample for the teaching analysis platform to dynamically optimize the teaching intervention model. In terms of parameter settings, η i usually takes values in the range of 0.05 - 100, generated according to the sensitivity of the behavior type. The threshold paragraphs for the behavior index Ω(u) interval are generally set as 0, 1.5, 1.5, 3.0, 3.0, ∞, and are dynamically adjusted according to the student's semester progress and behavior stability. Through this mechanism, the system realizes a complete closed-loop perception of "behavior individual - deviation feature - dynamic portrait" in terms of data, constructs an analysis ability of behavior state differences with both granularity and strategic value, and greatly improves the teaching refinement management level and student support response efficiency in the smart campus environment.

[0112] In the case of Xiao Ming's serious behavioral state deviation in the smart campus system, this system not only conducts intervention analysis on his individual deviation, but also further initiates the most critical link of the polymorphic perception and difference recognition mechanism of heterogeneous behavioral states - group analysis and dynamic feedback optimization at the strategy level. After detecting that Xiao Ming's score Ω(u) = 3.32 reached the serious deviation level, the system did not handle it in isolation, but input the deviation result and the behavioral deviation scores of 428 students of the same major and grade into the behavioral state clustering model. The system uses the K-Means algorithm to divide students into three categories according to the behavioral deviation feature vector, of which the low deviation group (Cluster 1) accounts for about 62.1%, the moderate deviation group (Cluster 2) accounts for about 29.7%, and the high deviation group (Cluster 3) including Xiao Ming accounts for 8.2%. The system further observed that in the high deviation group, 93% of the students did not complete the post-class evaluation task of the "Data Structure" course, and this course has never had such a large-scale evaluation missing rate in the past five semesters. The system initially determined that this phenomenon was not caused by the individual will of students, but rather by a problem in the design of a key node in the current teaching strategy path, so it started the strategy perturbation sensitivity test mechanism. The system sets the current "Data Structure" after-class evaluation strategy path P to: teaching platform homepage → course information page → evaluation reminder pop-up window (path P), and performs a slight perturbation operation on it: move the evaluation reminder entrance 20px upward from the middle of the course information page, and postpone the pop-up window time from "5 minutes after class" to "10 minutes after class". The adjustment perturbation amplitude is set to h = 0.05, which belongs to the system-defined perturbation interval (value range: 0 <h≤0.1,表示相对操作或策略权重调整不超过10%)。在实施策略扰动后的第二教学周内,系统对高偏差群体重新采集行为数据,小明在新策略下完成了评教,其他受控学生中有41人完成评教,其行为状态评分平均值由Ω before =3.17 down to Ω after =2.21, the state difference change is ΔΩ(u)=3.17-2.21=0.96, the corresponding strategy disturbance is ΔR(P,h)=h=0.05, which is substituted into the strategy disturbance response function:

[0113]

[0114] This value is much higher than the response sensitivity threshold of 8.0 set by the system (the recommended value range is 5.0 - 10.0, configured according to the overall behavioral elasticity of the system), indicating that path P is highly sensitive to minor policy adjustments, with a significant behavior recovery effect, further verifying the failure problem existing in the original design of this policy. Based on this conclusion, the system is pushed to the educational affairs decision-making module, suggesting to optimize the position of the teaching evaluation reminder pop-up window to the top of the information page in all courses and uniformly set the delayed reminder time to 10 minutes. At the same time, the priority of this policy path is increased by 15% through the mechanism of enhancing the priority weight of the behavior path. In addition, the policy perturbation response score Σ(P) = 19.2 is also recorded in the policy evolution map for subsequent policy simulation of similar courses and training of the fine-tuning recommendation model. In summary, through the polymorphic perception of the behaviors of Xiaoming and the group of students and the perturbation response evaluation of the policy path, the intelligent campus data management system in the present invention can not only accurately identify individual behavior deviations, but also has the ability to detect the failure of the overall policy from perturbations, and realizes closed-loop optimization through precise calculation and the policy feedback mechanism, thus significantly improving the intelligence and dynamic adaptability of educational data governance.

[0115] The above shows and describes the basic principles, main features and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited by the above embodiments. What is described in the above embodiments and the specification only illustrates the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of the present invention claimed is defined by the appended claims and their equivalents.

Claims

1. A smart campus data management method and system, characterized in that The following steps are involved: S1. Use semantic consistency-oriented data behavior modeling: S1.

1. Abstract all system data into behavioral event units; use spatiotemporal behavioral graph models to express dynamic connections between entities; S1.

2. Convert the modeling results into semantic vectors and write them into the campus semantic behavior database; and establish an event standardization rule base to enable event identification, fusion, and alignment; S2. Dynamic data flow scheduling engine construction using behavior link graph: S2.

1. Build a behavior path map based on behavior event nodes, extract high-frequency behavior links including course selection → entering the classroom → post-class evaluation → grade assessment; each link is assigned a flow cost and a security weight; S2.2, introduce graph search algorithm to calculate data scheduling path; activate link data set on demand at runtime to enable cross-system data linkage call and minimal redundancy flow; S3, adopting the multi-state perception and difference recognition mechanism of heterogeneous behavior states: S3.

1. Capture the deviation between actual behavior events and expected behavior links; conduct multi-dimensional analysis of behavior deviations including frequency, intensity, and associated objects, and introduce behavior state difference function; S3.

2. Build a behavior state comparison model to identify potential abnormal states or strategy failure risks; The output difference data is used for system adjustment, user prompts, and strategy reconstruction; S4, Adaptive optimization and feedback update mechanism using strategy evolution graph: S4.

1. Each data dispatch and behavior response is written into the strategy evolution map as a strategy execution event; user behavior feedback including changes in scores after intervention and increase in resource clicks is collected to construct a causal path; S4.

2. Introduce reinforcement learning algorithm to perform strategy scoring and value return calculation on the strategy map; S4.

3. Optimize link parameters, resource scheduling logic, and notification rhythm to form a closed loop of behavior data → management response → effect evaluation → strategy evolution.

2. A smart campus data management method and system according to claim 1, characterized in that The method for constructing a dynamic data flow scheduling engine of the behavior chain graph includes: The original data from subsystems with different functions, including the academic affairs system, access control system, and teaching evaluation platform, are abstracted into event units with behavioral characteristics. Each behavioral event encapsulates the user's behavioral intention, occurrence time, behavior location, and system interaction status. Through abstraction, data from different sources and in different formats have a unified expression basis at the semantic layer. The behavioral events are constructed into a multidimensional behavioral graph, in which the nodes represent behavioral events and the edges represent the logical, temporal, or causal relationships between events. Based on the behavioral graph, the scheduling overhead of each candidate path is evaluated.

3. A smart campus data management method and system according to claim 2, characterized in that The method for constructing a dynamic data flow scheduling engine of the behavior chain graph includes: The security risk of data usage is further evaluated on each path, taking into account the user data permissions, information sensitivity, and historical call behavior involved; a security weight function is introduced, expressed as follows: in: Λ(lij) represents the security risk weight value on the link. The higher the value, the higher the security requirement. sij is the data sensitivity level involved in the link. The larger the value, the more sensitive the data. uij represents the matching degree of the current user's permission level under the data type, including whether the user has the permission to view or modify. dij is the access frequency of the link data in history. α and β are the risk adjustment coefficients set in the system, which are used to adjust the proportion of sensitivity and permission risk in the assessment as needed.

4. A smart campus data management method and system according to claim 3, characterized in that The method for constructing a dynamic data flow scheduling engine of the behavior chain graph includes: When the system receives a data scheduling request, it maps the request target to the terminal node in the behavior path map, and uses the path search engine to perform a comprehensive cost and risk score on all candidate paths to select the optimal path. The path evaluation function is introduced to combine the evaluation values ​​of the two functions and score the entire behavior path P, which is defined as follows: in: Θ(P) is the total evaluation score of path P, that is, the comprehensive cost of the path; Φ(lij) is the cost function value of each hop link in the path; Λ(lij) is the security weight value of each hop link in the path; γ is the security offset factor set by the system, which is used to weigh the proportion between performance and security: when γ increases, the system pays more attention to security; when γ decreases, it pays more attention to efficiency.

5. A smart campus data management method and system according to claim 1, characterized in that The method for constructing a multi-state perception and difference recognition mechanism for heterogeneous behavior states includes: Through continuous monitoring of campus user behavior, user operation records in different systems are converted into standardized behavioral event sequences; each event is based on a timestamp and is accompanied by behavior type, behavior object, scene source, and operation depth information; based on user category, time scene, and teaching cycle, an expected behavior link library is constructed on the basis of historical behavior data. The link library describes the typical behavior path that a certain type of user should present, and is used as a reference model for real-time behavior comparison. When a real-time behavior event is received, the current behavior trajectory is compared with the corresponding expected link, and the cumulative degree of behavior deviation is calculated in an integral manner.

6. A smart campus data management method and system according to claim 5, characterized in that The method for constructing a multi-state perception and difference recognition mechanism for heterogeneous behavior states includes: Multi-dimensional features are extracted from the behaviors involved in the deviation events, including frequency changes, intensity fluctuations, and resource transfer information. All behavioral features are mapped into standard vectors and compared dimension by dimension with the historical state vectors to construct a behavioral state deviation model. The following behavioral state difference function is introduced: in: Ω(u) is the comprehensive deviation score of the current behavior state of user u; f i (a) is the behavioral feature value of the i-th course currently collected, including the number of course clicks and reading time; f i (e) is the expected value of the i-th feature in the user’s history, extracted from historical behavior; η i is the fluctuation degree of the i-th feature in historical data, which is used to avoid over-sensitivity to natural fluctuations; n is the total number of feature dimensions.

7. A smart campus data management method and system according to claim 6, characterized in that The method for constructing a multi-state perception and difference recognition mechanism for heterogeneous behavior states includes: Based on the behavioral deviation scores of multiple users, individual behavioral anomalies and overall strategic failures are identified through cluster modeling, anomaly detection, and time series evolution analysis. When a large range of deviations are concentrated on a certain strategy path, the system needs to determine whether it is a strategy design failure, simulate slight strategy adjustments, and observe the sensitivity response of the behavior state. The following strategy perturbation response function is defined: in: Σ(P) is the disturbance response sensitivity index of the behavior path P; ΔΩ(u) represents the change in the state difference value of user u after the strategy adjustment; ΔR(P,h) represents the slight strategy disturbance operation implemented on the path P, including fine-tuning the entrance layout and delaying the intervention prompt time; h is the disturbance amplitude factor.

Images