Integrated marketing closed-loop management system and method

Abstract

The invention provides an integrated marketing closed-loop management system and method, and relates to the field of marketing management. Comprising the steps of capturing user behavior data in real time and generating an intention label through an intention analysis module; the resource management module calculates a weight difference value according to the difference value between the label and the label of the resource occupant, and triggers the preemption execution module when the weight difference value exceeds a threshold value; the preemption execution module migrates high-intention user data, monitors intention reduction of an original occupant to trigger resource release, and feeds back a result to the feedback learning module; the feedback learning module adjusts an intention generation rule according to the preemption success rate and the conflict rate; the degradation control module is switched to the traditional allocation logic when the transaction is overtime or the conflict exceeds the limit, the uncompleted operation is frozen, and the degradation control module rolls back to the resource snapshot; all the modules work cooperatively, efficient management and dynamic allocation of marketing resources are achieved, and the operation efficiency and stability of the system are improved.

Description

Technical Field

[0001] The present invention relates to the field of marketing management, and specifically to an integrated marketing closed-loop management system and method. Background Art

[0002] With the acceleration of the digital transformation of enterprises, the marketing scenario extends from a single channel to all online and offline touchpoints, and consumer behavior data shows fragmented and real-time characteristics; traditional marketing systems rely on offline data analysis and manual strategy adjustment, making it difficult to meet the precise resource matching requirements of high concurrency and low latency, resulting in low cross-channel collaboration efficiency and fragmented user experience; there is an urgent need in the current market for a closed-loop management system that can real-time perceive user intentions and dynamically allocate resources to cope with the complex challenges of omnichannel marketing.

[0003] Existing technical solutions mostly adopt static strategy allocation based on rule engines or historical behavior prediction models, triggering fixed marketing actions by presetting thresholds; for example, pushing coupons by stratifying customers' historical consumption amounts, or recommending similar products using collaborative filtering algorithms; such solutions rely on manual experience to define rules or offline train models, although they can achieve basic resource allocation, they cannot dynamically respond to real-time behavior changes, and lack a cross-touchpoint collaboration mechanism, resulting in frequent strategy lag and resource conflicts.

[0004] The deficiencies of the existing technologies are high response latency, inability to capture the dynamic migration of user intentions; cross-channel data isolation leads to resource allocation conflicts, such as online advertisements and offline activities not being triggered synchronously; static rules and offline models are difficult to adapt to sudden changes in the market environment, such as strategy failure when competitors promote or inventory fluctuates; in addition, traditional systems lack an abnormal degradation mechanism, and high-load scenarios are prone to service interruptions, further exacerbating the decline in user experience and resource waste. Summary of the Invention

[0005] (1) Technical Problems to be Solved Aiming at the deficiencies of the existing technologies, the present invention provides an integrated marketing closed-loop management system and method to solve the problems of lagging user intention capture caused by high response latency, resource allocation conflicts caused by cross-channel data isolation, difficulty in adapting to sudden changes in the market environment by static rules and offline models, and service interruptions caused by the lack of an abnormal degradation mechanism in high-load scenarios as mentioned in the above background art.

[0006] (2) Technical Solutions To achieve the above objectives, the present invention is realized through the following technical solutions: an integrated marketing closed-loop management system and method, including an intention analysis module, a resource management module, a preemption execution module, a feedback learning module, and a degradation control module; The intention analysis module captures behavior data in real time through edge nodes deployed at user touchpoints, generates intention tags based on the change rate of operation frequencies and the cross-touchpoint path chaos degree of the behavior data, and transmits the tags to the resource management module; The resource management module maps physical resources to virtual units, calculates the weight difference according to the tag difference between the intention tag and the resource occupant, and triggers the preemption execution module when the weight difference exceeds the threshold; The preemption execution module migrates the context data of high-intention users to the target resources, monitors the decrease of the intention of the original occupant and triggers resource release, and feeds back the preemption result to the feedback learning module; The feedback learning module adjusts the intention generation rules according to the preemption success rate and the conflict rate, and synchronizes the rules to the intention analysis module; The downgrade control module switches to the traditional allocation logic when the transaction times out or the conflict times out, freezes the unfinished operations and rolls back to the resource snapshot.

[0007] Preferably, the intention analysis module captures the operation behaviors of users at online and offline touchpoints in real time through edge nodes deployed at user touchpoints, including clicks, scans, page stay durations, and cross-channel jump paths; the module extracts the operation timestamps and touchpoint identifiers, calculates the absolute value of the change rate of operation frequencies per unit time as the behavior density gradient. When the number of clicks on the product detail page by the user suddenly increases from 1 to 5 times within 5 seconds, the behavior density gradient is calculated as ; the module also analyzes the probability distribution differences of the user's cross-touchpoint switching paths. Assume that online price comparison is A, offline scan is B, and returning to search for competing products online is C. Through the Shannon entropy formula calculate the path chaos degree, where H is the path chaos degree, used to measure the randomness of the user's cross-touchpoint jump order; p i is the probability of the i-th path occurring, such as the probability that the user jumps from touchpoint A to touchpoint B; when the path entropy value from A to B to C by the user is 1.2, it is determined as high chaos degree; the gradient value and the path chaos degree are input into a non-linear function adjusted by dynamic parameters to generate intention tags. The parameters of the non-linear function are corrected in real time according to the competing product activity intensity and the inventory change rate. When the competing product promotion intensity increases, the parameter weight is increased by 20%; the intention analysis module encrypts the generated intention tags and transmits them to the resource management module through a dedicated channel. The encryption uses the AES-256 algorithm and attaches a timestamp signature. The non-linear function is expressed as , where S(x) is the output value, mapped to the intention intensity between 0 and 1; x is the input value, that is, the weighted sum of the behavior density gradient and the path chaos degree; k is the slope parameter, controlling the steepness of the function curve, dynamically adjusted according to the competing product activities; x0 is the offset parameter, used to calibrate the intention intensity threshold, optimized according to historical data.

[0008] Preferably, the resource management module maps advertising spaces, customer service seats and inventory resources into virtual units, each virtual unit is bound to a preset intent effectiveness interval and resource decay rate, the customer service seat is only effective for users with an intent label ≥ 4 and the resource decay rate is a 10% decrease in utility per minute; the resource management module calculates the weight difference based on the difference between the current intent label and the resource occupant's label, and the difference Δ=2 when the current user intent label is 4 and the resource occupant label is 2; when Δ exceeds the system preset threshold of 1.5, the resource management module sends a trigger signal to the preemption execution module; the module binds the resource decay rate to the hardware clock cycle, and writes the decay rate parameter through the register of the FPGA chip, so that the resource utility calculation is synchronously refreshed each time the intent label is updated; the virtualization process of the physical resources is implemented through a hash table, and each physical resource ID is mapped to a unique key value of the virtual unit.

[0009] Preferably, after receiving the trigger signal, the preemption execution module migrates the historical operation records and unfinished transactions of high-intent users to the target resources, and migrates the unpaid orders and customer service conversation records of VIP customers to the exclusive customer service seat; the module starts a timing window to monitor the intention changes of the original occupant, and if the original occupant's intention label drops from 3 to 1 within 10 seconds, the resources are forcibly released and the coupons of the store gift package are pushed to him; the module splits the preemption process into three steps: resource status update, context migration and compensation triggering. First, the ad slot status is updated to occupied, and then the user context data is migrated through the transaction log, and finally the compensation push is triggered; when any step fails, the module rolls back to the state before the preemption based on the operation serial number and timestamp, and ensures the consistency of the resource status and database records through a two-phase commit protocol.

[0010] Preferably, the feedback learning module receives the preemption results and subsequent conversion rate data from the preemption execution module, stores them in the database in timestamp order, and records whether a purchase is made within 30 seconds after a certain advertising position is preempted; the module adjusts the calculation weight of the behavior density gradient according to the preemption success rate, and the gradient weight is reduced by 15% when the conversion rate of a certain type of resource is lower than 10% after being preempted; the module corrects the coefficient of path confusion through conflict rate data, and increases the entropy coefficient of the high-conflict path to 1.5 times when the resource conflict rate exceeds 5%; the module synchronizes the updated weights and coefficients to the intent analysis module, and replaces the intent label generation function parameters through a hot update mechanism; the synchronization process adopts a version control mechanism, and generates a unique version number for each update.

[0011] Preferably, the downgrade control module triggers downgrade when the average transaction response time exceeds 15 milliseconds for 5 consecutive times or the resource status exception rate exceeds 5%, and starts switching when the customer service agent assignment delay reaches 20 milliseconds; the downgrade control module buckets user requests into buckets numbered 0-99 according to the hash value of the device unique identifier modulo 100; the downgrade control module gradually switches the bucket traffic to the traditional hierarchical allocation logic based on the user's historical consumption amount and access frequency, and preferentially assigns high-consumption users to the VIP channel; the module freezes all unfinished operations and rolls back to the nearest consistent version of the resource snapshot, and ensures that no new operations are written during the rollback through a distributed lock; the module maps the intent tags to static user hierarchical identifiers, maps VIP customers to the range of 4-5 of the intent tags, and stores them in an independent downgraded database table.

[0012] Preferably, the system dynamically associates the traditional hierarchical identifier mapping with a preset intent range in the downgraded mode. When a user is identified as a VIP customer, the system maps it to the range of 4-5 of the intent tags according to the preset rules; the system generates a corresponding virtual resource allocation instruction, which includes an intent level parameter and a timeout limit. For example, when allocating resources for a VIP customer, the intent level is set to 4 and the timeout is set to 10 seconds; to achieve interface compatibility and traffic tracking, the system inserts specific identification fields in the HTTP request header, including the bucket number generated based on the hash value of the user device unique identifier and the downgrade switch timestamp. For example, the bucket number "X-Degrade-Bucket: 25" and the timestamp "X-Switch-Time: 1630000000"; the bucket number is used to identify the traffic group to which the user belongs, supports the precise execution of the progressive switching strategy, and the switch timestamp records the exact moment when the downgrade operation occurs, which is used for subsequent data consistency verification and fault troubleshooting.

[0013] Preferably, the system sends simulated low-intensity preemption instructions to the preemption execution module during the protocol recovery phase, sending 10 virtual preemption requests per second; when the response time of 3 consecutive requests is lower than 20 milliseconds and the success rate is higher than 95%, the system reroutes 10% of the user traffic to the preemption protocol and gradually opens the traffic of buckets numbered 1-10; if the number of downgrade triggers accumulates to 5 times within 24 hours, the system permanently closes the preemption protocol and locks it to the traditional mode, disabling all interfaces of the preemption execution module; the system maintains the traditional allocation logic before manual intervention, and only allows administrators to reset the fuse status through the console; the persistent storage of the fuse status uses a distributed key-value database.

[0014] Preferably, the resource management module binds the resource attenuation rate to the device hardware fingerprint, and the hardware fingerprint is generated by the hash value of the device MAC address and the random offset of the system clock. The formula is ; where Hash represents a hash function used to convert the input string into a hash value of a fixed length; MAC is the MAC address of the device, which is a 48-bit unique identifier used to uniquely identify a network interface card in the network; is the modulus of the timestamp representing the system time with respect to 1000, obtaining a random offset between 0 and 999. The purpose of this operation is to increase a certain degree of randomness and prevent the hardware fingerprint from being easily forged. A confusion key is generated every 24 hours through a physical noise source, and a 256-bit key is output using a quantum random number generator; the key is synchronously refreshed with the utility decay curve in the hardware register, and the register value of the FPGA chip is updated at 0:00 every day; the system injects the confusion key into the intent path calculation and uses the key as the initial vector for entropy value calculation; the confusion key is stored in the hardware register and external access is prohibited; the physical noise source is a true random number generation device based on the principles of quantum mechanics, which generates unpredictable noise signals through quantum physical phenomena to achieve the dynamic update of the confusion key.

[0015] Preferably, the intent analysis module real-time statistically counts the operation frequency of the user per unit time. When the number of clicks within 5 seconds increases from 2 to 8 times, the gradient is ; and dynamically adjusts the gradient weight according to the intensity of competitor activities. During the competitor's promotion period, the gradient weight is increased to 1.5 times; then the corrected gradient value is input into a non-linear function to be mapped into an intent intensity value between 0 and 1, and the final intent label is generated in combination with the degree of chaos; the parameters of the non-linear function are optimized through the gradient descent algorithm, and the step size is adjusted based on historical data for each iteration.

[0016] Preferably, the intent analysis module statistically counts the occurrence probability of the user's cross-touchpoint switching path. When the path from A to B and then to C has an occurrence probability of 30%, the value of the Shannon entropy is calculated to be 1.5. The feedback learning module increases the coefficient of the high-chaos path according to the resource conflict rate. When the conflict rate exceeds 5%, the entropy value weight is reduced by 20%; the intent analysis module synchronizes the updated coefficient to the intent analysis module through a message queue to ensure that the subsequent chaos calculation fits the actual scenario.

[0017] Preferably, when the downgrade control module triggers a downgrade, it generates a resource snapshot image and stores it in a distributed database. The snapshot includes the resource occupancy status, user context data, and operation serial number; the module restores to the nearest consistent version according to the timestamp and serial number during rollback, and selects the complete snapshot closest to the current time; the module ensures the atomicity of rollback through a two-phase commit protocol, first locking the resource status and then writing the snapshot data; the snapshot data adopts a columnar storage format to improve the query efficiency.

[0018] (III) Beneficial Effects The present invention provides an integrated marketing closed-loop management system and method. It has the following beneficial effects: 1. By capturing real-time user full-touch behavior data and dynamically generating intent tags, the present invention accurately identifies the intensity of users' purchase intent; by binding the resource attenuation rate to the hardware clock cycle, it achieves millisecond-level response, ensuring that high-intent users can preempt high-value resources first; through an adaptive feedback mechanism, it continuously optimizes the intent generation rules and automatically adjusts strategies to cope with competitor activities and market fluctuations, increasing the cross-channel conversion rate by more than 40%; the downgrade control module seamlessly switches to the traditional hierarchical allocation mode in case of system anomalies to ensure service continuity; the hardware fingerprint binding and quantum random obfuscation key design completely block reverse engineering and form a technical barrier; ultimately, it maximizes the utilization rate of marketing resources, reduces budget waste by 35%, and seamlessly connects the user experience.

[0019] 2. Through the virtualized resource pool and dynamic preemption protocol, the present invention deeply couples the physical resource utility with the user intent, ensuring that customers in the high-intent stage can be accurately reached the optimal resources; the atomic transaction rollback and two-phase commit protocol guarantee the data consistency of the preemption operation and avoid resource conflicts and state disorders; the intent path confusion injection and columnar snapshot storage improve the system's anti-attack ability and recovery efficiency; the dual-mode redundancy design of the traditional mode and the preemption mode supports stable operation in scenarios with hundreds of millions of concurrencies; through the LRU cache and CDN preloading technologies, it maintains the smoothness of user interaction in the downgraded mode. Specific embodiments

[0020] Next, the technical solutions in the embodiments of the present invention will be described clearly and completely. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0021] Embodiment 1: An embodiment of the present invention provides an integrated marketing closed-loop management system and method. User A browses the details page of a certain smartphone in an e-commerce APP. In the initial stage, the user clicks on the specification parameters every 10 seconds, and the behavior density gradient is calculated as 0.1. The path chaos degree is based on its single-touch behavior, that is, the Shannon entropy value of only browsing within the APP is 0.2. The system determines that the user is in the hesitation period and generates intent label 2. Subsequently, User A suddenly switches to a third-party price comparison platform and continuously refreshes the price comparison results 8 times within 5 seconds. The behavior density gradient suddenly increases to 1.6. At the same time, the probability distribution of its cross-touchpoint path from the APP details page to the price comparison platform and then back to the APP to search for competing products is significantly different. The chaos degree H calculated by the Shannon entropy formula is 1.3. The comprehensive gradient and chaos degree are input into a non-linear function for dynamic parameter adjustment, and intent label 4 is output. The intent analysis module encrypts the label and transmits it to the resource management module. The encryption process uses the AES-256 algorithm and attaches a timestamp signature to prevent tampering.

[0022] Then the resource management module scans the virtual resource pool and finds that the exclusive guided tour quota of a certain offline experience store is currently occupied by a user with intent label 2. It calculates the difference Δ = 4 - 2 = 2, which exceeds the preset threshold of 1.5, and triggers the preemption execution module. The preemption execution module immediately freezes the resource permissions of the original occupant, migrates the browsing records, unfinished orders, and historical search keywords of User A to this quota, updates the resource status to occupied, and synchronizes it to the distributed database through the two-phase commit protocol. At the same time, it starts a 10-second timing window to monitor the behavior of the original occupant. If the original user does not have the intention to upgrade within the window period, such as not triggering high-frequency operations or cross-touchpoint jumps, the system forcibly releases the resources and pushes a text message compensation about the in-store gift package to the user. After receiving the push of the exclusive guided tour plus time-limited discount, User A arrives at the offline store within 30 minutes and completes the purchase using the discount. The transaction data is transmitted back to the feedback learning module in real-time. The feedback learning module analyzes that the conversion rate of this path is 100%, increases the calculation weight of such behavior density gradient by 20%, reduces the penalty coefficient of path chaos degree by 10%, and synchronizes the new parameters to the edge node through the hot update mechanism. The degradation control module monitors the system response time throughout the process and keeps it stable in the range of 8 - 12 milliseconds, and the resource conflict rate is only 1.8%, without triggering the degradation condition. At the hardware level, the resource attenuation rate is strictly synchronized with the FPGA register clock, and the confusion key is refreshed through a quantum random number generator at 0:00 every day to ensure that the intent path calculation cannot be reversely deduced. Throughout the process, the migration of User A from low intent to high intent, resource preemption, compensation push, and parameter optimization form a closed loop, and the system maximizes resource utilization through millisecond-level response and self-learning ability.

[0023] Embodiment 2: The differences between this embodiment and Embodiment 1 are as follows: User B browses the details page of a certain brand of laptop on the e-commerce APP and clicks on the product picture every 30 seconds. The behavior density gradient is 0.03, and the path chaos degree is based on its single-touch behavior, that is, the Shannon entropy value of only browsing within the APP is 0.1, and the system generates intention label 1; Subsequently, User B scans the code to enter the brand mini-program to view the offline store inventory information. The probability distribution difference of the cross-touchpoint path from the APP details page to the mini-program inventory page is relatively low, and the Shannon entropy H is 0.5, and intention label 1 is comprehensively generated; The resource management module detects that the intention label of the occupant of a certain store inventory display position is 3, and calculates the difference Δ = 3 - 1 = 2, which exceeds the threshold of 1.5, triggering the preemption execution module; During the preemption process, due to network fluctuations, the average transaction response time exceeds 15 milliseconds continuously for 5 times, and the degradation control module immediately starts the bucket switching logic, calculates the bucket number 47 by taking the hash value of User B's device ID modulo 100, and gradually switches the traffic of bucket 47 to the traditional hierarchical allocation mode; Based on User B's historical consumption amount of 5000 yuan accumulated and monthly average access frequency of more than 10 times, the traditional logic classifies it as an intermediate user, allocates it to the ordinary inventory channel and hides the real-time discount entrance; The front-end interface preloads static page resources through CDN to ensure that the user perception delay is controlled within 180 milliseconds; The resource snapshot mirror generation module captures the current resource occupancy status, User B's context data and operation serial number, stores them in the columnar database and marks the timestamp.

[0024] During the rollback process, the resource status is locked through the two-phase commit protocol to ensure that no new operations interfere; Within the next 1 hour, User B did not trigger any high-intention behaviors, such as high-frequency refreshing or cross-channel jumping, and the system continued to allocate resources in the traditional mode to avoid ineffective preemption; After the cumulative number of degradation triggers reaches 5 times within 24 hours, the system permanently fuses the preemption protocol, disables the preemption execution module interface and locks it to the hierarchical strategy; The feedback learning module analyzes the data during the degradation period and finds that the conversion rate of such low-frequency users is only 3%, and raises the behavior density gradient threshold from 1.5 to 2.0 to reduce inefficient resource preemption; The confusion key is updated at 0:00 every day as planned, generates a new key by combining the hash value of the device MAC address and the random offset of the system clock, and synchronously refreshes the resource decay curve in the FPGA register; The binding of the hardware fingerprint and the confusion key ensures that the defense mechanism remains effective and blocks reverse engineering; Throughout the process, the system optimizes the long-term strategy while ensuring stability through elastic degradation, bucket isolation and dynamic parameter adjustment, verifying its robustness and adaptability in abnormal scenarios.

[0025] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An integrated marketing closed-loop management system, characterized in that: It includes an intention analysis module, a resource management module, a preemption execution module, a feedback learning module, and a degradation control module; The intention analysis module captures behavior data in real time through edge nodes deployed at user touchpoints, generates intention tags based on the change rate of operation frequency and the cross-touchpoint path confusion degree of the behavior data, and transmits the tags to the resource management module; The resource management module maps physical resources into virtual units, calculates the weight difference according to the difference between the intention tag and the tag of the resource occupant, and triggers the preemption execution module when the weight difference exceeds the threshold; The preemption execution module migrates the context data of high-intention users to the target resource, monitors the decrease in the intention of the original occupant and triggers resource release, and feeds back the preemption result to the feedback learning module; The feedback learning module adjusts the intention generation rule according to the preemption success rate and conflict rate, and synchronizes the rule to the intention analysis module; The degradation control module switches to the traditional allocation logic when the transaction times out or the conflict times out, freezes the unfinished operations and rolls back to the resource snapshot.

2. The integrated marketing closed-loop management system according to claim 1, characterized in that: The intention analysis module captures the operation behaviors of users at online and offline touchpoints in real time, extracts the operation timestamps and touchpoint identifiers, calculates the absolute value of the change rate of operation frequency per unit time as the behavior density gradient, and uses the Shannon entropy formula to analyze the probability distribution difference of the user's cross-touchpoint switching path as the path confusion degree, and inputs the behavior density gradient and the path confusion degree into a non-linear function with dynamically adjusted parameters to generate intention tags; then encrypts the generated intention tags and transmits them to the resource management module, and the parameters of the non-linear function are corrected in real time according to the competitive product activity intensity and inventory change rate.

3. An integrated marketing closed-loop management system according to claim 1, characterized in that: The resource management module maps physical resources into virtual units, binds a preset intention effective interval and resource attenuation rate to each virtual unit, calculates the weight difference according to the difference between the current intention tag and the tag of the resource occupant, sends a trigger signal to the preemption execution module when the weight difference exceeds the system threshold, and binds the resource attenuation rate to the hardware clock cycle to synchronize the resource utility calculation with the intention tag update, and the physical resources include advertising spaces, customer service seats, and inventory resources.

4. An integrated marketing closed-loop management system according to claim 3, characterized in that: The resource management module binds the resource attenuation rate to the device hardware fingerprint, which is generated by the hash value of the device MAC address and the random offset of the system clock, generates a confusion key through the physical noise source at regular intervals and synchronizes and refreshes it with the hardware register, and injects the confusion key into the intention path calculation to block the reverse engineering to deduce the behavior rule.

5. An integrated marketing closed-loop management system according to claim 1, characterized in that: After receiving the trigger signal, the preemption execution module migrates the historical operation records and unfinished transactions of high-intention users to the target resource, and at the same time starts a timing window to monitor the intention change of the original occupant. If the intention of the original occupant continues to decrease within the window period, it forcibly releases the resource and pushes a compensation strategy to it, and splits the preemption process into three steps: resource status update, context migration, and compensation trigger. When any step fails, it rolls back to the preemption-before state based on the operation serial number and timestamp.

6. The integrated marketing closed-loop management system according to claim 1, wherein: The feedback learning module receives the preemption results and subsequent conversion rate data of the preemption execution module, stores them in the database in timestamp order, adjusts the calculation weight of the behavior density gradient according to the preemption success rate, corrects the coefficient of the path confusion degree through the conflict rate data, and synchronizes the updated weight and coefficient to the intention analysis module to overwrite its original generation rules.

7. An integrated marketing closed-loop management system according to claim 1, characterized in that: The downgrade control module triggers downgrade when the transaction average response time exceeds a set threshold continuously for multiple times or the resource status exception rate reaches a certain proportion. It buckets user requests by device unique identifier, gradually switches the bucket traffic to the traditional hierarchical allocation logic based on the user's historical consumption amount and access frequency. During the switching process, all unfinished operations are frozen, rolled back to the nearest consistent version of the resource snapshot, and the intention label is mapped to a static user hierarchical identifier.

8. An integrated marketing closed-loop management system according to claim 7, characterized in that: In the downgraded mode, the traditional hierarchical identifier is mapped to a preset intention interval, converted into a virtual resource allocation instruction, the bucket number and switching timestamp are embedded in the HTTP request header to maintain interface compatibility, the function entry dependent on real-time preemption in the front-end interface is hidden, and the user perception delay is controlled within a fixed time through the global cache of static pages.

9. An integrated marketing closed-loop management system according to claim 7, characterized in that: In the protocol recovery stage, a simulated preemption request is sent to the preemption execution module. When the response time of three consecutive requests is lower than another set threshold and the success rate exceeds the set proportion, one-tenth of the user traffic is re-routed to the preemption protocol. If the number of downgrade triggers accumulates to a certain number within a day, the preemption protocol is permanently closed and locked to the traditional mode until manual intervention.

10. An integrated marketing closed-loop management method proposed according to an integrated marketing closed-loop management system as claimed in claim 1, characterized in that: The intention analysis module captures user behavior data in real time at the edge nodes of user touchpoints, generates intention labels based on the change rate of operation frequency and cross-touchpoint path confusion degree of the behavior data, and transmits the labels to the resource management module; The resource management module maps physical resources to virtual units, calculates the weight difference according to the difference between the intention label and the label of the resource occupant, and triggers the preemption execution module when the weight difference exceeds the threshold; The preemption execution module migrates the context data of high-intention users to the target resources, monitors the decrease in the intention of the original occupant and triggers resource release, and feeds back the preemption results to the feedback learning module; The feedback learning module adjusts the intention generation rules according to the preemption success rate and conflict rate, and synchronizes the rules to the intention analysis module; The downgrade control module switches to the traditional allocation logic when the transaction times out or conflicts, freezes the unfinished operations and rolls back to the resource snapshot.