A resource scheduling method, apparatus, device, medium, and product

Abstract

This application discloses a resource scheduling method, apparatus, device, medium, and product, relating to the field of computer technology. The method includes: acquiring initial resource data of a target object; performing feature standardization processing on the data to obtain a target resource feature vector; identifying the problem based on this vector and outputting a structured decision object; performing interval threshold calculations based on four constraints—inventory, category revenue, user reach, and resource allocation budget—to generate a complete parameterized allowable strategy numerical space; deriving candidate resource optimization strategies within this space boundary; filtering out non-compliant solutions through hard rule verification; selecting the target strategy using a multi-objective comprehensive scoring rule; and finally converting the target strategy into parameterized instructions and issuing them to the resource scheduling backend for execution. This method can improve the feasibility of strategy implementation.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a resource scheduling method, apparatus, device, medium and product. Background Technology

[0002] As the business scale of online resource service platforms continues to expand, these platforms generate massive amounts of multi-dimensional resource scheduling data, including user and inventory data, every day. Automating problem identification and compliant resource strategy generation through intelligent means has become a common industry requirement, and intelligent decision-making tools based on large models are gradually being applied to resource scheduling scenarios on online resource service platforms.

[0003] The current mainstream resource scheduling processing methods are divided into two categories. One category can only complete the early warning of abnormal resource scheduling data, and manually formulate resource scheduling optimization schemes based on resource experience. The other category directly calls the large model to freely generate resource strategies without combining hard constraints for boundary restrictions.

[0004] Therefore, existing resource scheduling schemes suffer from poor feasibility in implementation. Summary of the Invention

[0005] This application provides a resource scheduling method, apparatus, equipment, medium, and product that can improve the feasibility of strategy implementation.

[0006] To achieve the above objectives, this application adopts the following technical solution: Firstly, this application provides a resource scheduling method, including: Obtain the initial resource data of the target object; The initial resource data is subjected to feature standardization processing to obtain the target resource feature vector; Problem identification is performed based on the feature vector of the target resource to obtain structured decision objects; Based on the numerical constraints of material unit inventory, category revenue, user classification and access control, and resource allocation budget, interval threshold calculations are performed on structured decision objects to obtain a complete parameterized allowable strategy numerical space. The policy generation process is performed within the parameter boundaries of the fully parameterized policy numerical space to obtain candidate resource optimization policies; Candidate resource optimization strategies are subjected to hard rule verification to obtain the candidate resource optimization strategies that pass the hard rule verification; based on the multi-objective comprehensive scoring rule, the candidate resource optimization strategies that pass the hard rule verification are ranked to determine the target strategy; the target strategy is converted into parameterized instructions and the parameterized instructions are sent to the target resource scheduling backend for execution.

[0007] In some possible implementations, problem identification is performed based on the target resource feature vector to obtain structured decision objects, including: Temporal anomaly detection and hierarchical drill-down are performed on the feature vector of the target resource to obtain resource anomaly data. The resource anomaly data includes the location of the anomaly subject and the scope of its impact. The resource anomaly data is matched with a preset standard problem database to determine the problem type label. Based on the problem type label, the resource anomaly data, and the scene features corresponding to the target object, the structured decision object is determined.

[0008] In some possible implementations, interval threshold calculations are performed on the structured decision object to obtain the complete parameterized allowable policy numerical space, including: The process involves: acquiring the associated entities, product categories, inventory levels, and user classification attributes from the structured decision-making object; calculating the upper limit range of resource allocation based on the current inventory value and turnover target value of the corresponding material unit for the associated entity; calculating the lower limit threshold for resource subsidies and the lower limit threshold for electronic voucher denomination based on the minimum revenue value and break-even point value corresponding to the product category; determining the scope of interactive user groups based on user classification tags corresponding to user classification attributes; determining the upper limit of interaction times per user within a preset time period based on historical interaction frequency records corresponding to user classification attributes; determining the upper limit of total resource allocation based on the total resource allocation budget value corresponding to the structured decision-making object; determining the upper limit of resource allocation per user based on resource input-output target values; and determining the complete parameterized allowed strategy value space based on the upper limit range of resource allocation, the lower limit threshold for resource subsidies, the lower limit threshold threshold for electronic voucher denomination, the scope of interactive user groups, the upper limit of interaction times, the upper limit of total resource allocation, and the upper limit of resource allocation per user.

[0009] Among the possible implementations are: Obtain the resource indicator values ​​after the parameterized instruction is executed; calculate the contribution value based on the resource indicator values ​​to obtain the overall strategy contribution value; compare the overall strategy contribution value with the preset target threshold to obtain the comparison result; perform feedback optimization processing based on the comparison result to obtain the feedback optimization processing result; wherein, the feedback optimization processing includes at least one of the first iteration processing, the second iteration processing, and the weight calibration processing.

[0010] In some possible implementations, feedback optimization is performed based on the comparison results to obtain the feedback optimization results, including: When the comparison result shows that the overall contribution value of the strategy is greater than or equal to the preset threshold, the target strategy is stored as a new knowledge unit in the knowledge base, resulting in an updated knowledge base. The ranking of each knowledge unit in the updated knowledge base is optimized to obtain updated ranking weights and retrieval priorities. When the cumulative number of knowledge units in the knowledge base exceeds the preset sample size threshold, the mapping relationship between the inventory value of the material unit and the upper limit range of resource allocation is refitted to obtain an updated inventory allocation mapping relationship. The mapping relationship between the category revenue value and the lower limit threshold of resource subsidies is refitted to obtain an updated revenue-resource subsidy mapping relationship. The strategy generation and processing parameters are fine-tuned to obtain fine-tuned strategy generation and processing parameters. The weight coefficients of each indicator in the multi-objective comprehensive scoring rule are calibrated to obtain updated weight coefficients.

[0011] In some possible implementations, policy generation is performed within the parameter boundaries of the fully parameterized policy numerical space to obtain candidate resource optimization policies, including: Based on the problem type, product category, inventory level, and user classification attributes in the structured decision object, matching historical knowledge units are retrieved from the knowledge base to obtain a recall knowledge unit set. Based on the comprehensive contribution value of each knowledge unit in the recall knowledge unit set, a reference knowledge unit set is determined. Based on the structured decision object, the complete parameterized allowable strategy numerical space, and the reference knowledge unit set, strategy schemes are deduced within the parameter boundaries of the complete parameterized allowable strategy numerical space to obtain candidate resource optimization strategies.

[0012] Secondly, this application provides a resource scheduling device, comprising: The acquisition module is used to obtain the initial resource data of the target object; The processing and calculation module is used to perform feature standardization on the initial resource data to obtain the target resource feature vector; based on the target resource feature vector, problem identification is performed to obtain the structured decision object; based on the numerical constraints of material unit inventory, category revenue, user classification and access control, and resource allocation budget, the interval threshold calculation is performed on the structured decision object to obtain the complete parameterized allowable strategy numerical space. The decision generation module is used to generate strategies within the parameter boundaries of the fully parameterized allowable strategy numerical space to obtain candidate resource optimization strategies; perform hard rule verification on the candidate resource optimization strategies to obtain the candidate resource optimization strategies that pass the hard rule verification; based on the multi-objective comprehensive scoring rules, sort the candidate resource optimization strategies that pass the hard rule verification to determine the target strategy; convert the target strategy into parameterized instructions and send the parameterized instructions to the target resource scheduling backend for execution.

[0013] Thirdly, this application provides a computing device, including a memory and a processor; The memory stores one or more computer programs, the one or more computer programs including instructions; when the instructions are executed by the processor, the computing device performs the method as described in any one of the first aspects.

[0014] Fourthly, this application provides a computer-readable storage medium for storing a computer program for performing the method as described in any one of the first aspects.

[0015] Fifthly, this application provides a computer program product comprising one or more computer instructions, wherein when the computer instructions are executed by a computer, the computer performs the method as described in any one of the first aspects.

[0016] As can be seen from the above technical solution, this application has at least the following beneficial effects: In this application, initial resource data of the target object is obtained and feature standardization is performed to obtain a target resource feature vector. Based on this feature vector, resource problems are identified and a structured decision object is generated. Then, based on numerical constraints of material unit inventory, category revenue, user classification and access control, and resource allocation budget, interval threshold calculations are performed on the structured decision object to obtain a complete parameterized allowable strategy numerical space. Within the parameter boundaries of this space, strategy generation processing is performed to obtain candidate resource optimization strategies. Finally, hard rule verification is performed on the candidate strategies, and the target strategy is determined by multi-objective comprehensive scoring and ranking. The target strategy is converted into parameterized instructions and sent to the target resource scheduling backend for execution. Before executing strategy generation, this solution first transforms four types of constraints—inventory, revenue, user reach, and resource allocation budget—into numerical parameter boundaries through interval threshold calculations. This forms a complete parameterized numerical space for allowed strategies, ensuring that the entire strategy generation process is completed within these parameter boundaries. This guarantees from the outset that the generated candidate strategies will not violate constraints. Simultaneously, after strategy generation, a hard rule verification process is performed. Candidate strategies undergo secondary verification based on minimum resource subsidy constraints, user interaction frequency, resource supplier settlement rules, and platform compliance rules. Strategies that do not meet the constraints are filtered or corrected, forming a dual guarantee mechanism of pre-constraint limitation and post-verification filtering. Furthermore, through structured decision objects, abnormal information in the original resource data is transformed into standardized decision inputs. A multi-objective comprehensive scoring rule is used to comprehensively rank candidate strategies from multiple dimensions, including the increase rate of total transaction amount per material unit, the increase rate of user retention, the increase rate of resource supplier revenue, and the resource input-output ratio. This ensures that the final target strategy achieves optimal multi-dimensional comprehensive performance while satisfying all constraints. Ultimately, this improves the feasibility of strategy implementation.

[0017] It should be understood that the descriptions of technical features, technical solutions, beneficial effects, or similar language in this application do not imply that all features and advantages can be achieved in any single embodiment. Rather, it is understood that the description of a feature or beneficial effect means that a specific technical feature, technical solution, or beneficial effect is included in at least one embodiment. Therefore, the descriptions of technical features, technical solutions, or beneficial effects in this specification do not necessarily refer to the same embodiment. Furthermore, the technical features, technical solutions, and beneficial effects described in this embodiment can be combined in any suitable manner. Those skilled in the art will understand that embodiments can be implemented without one or more specific technical features, technical solutions, or beneficial effects of a particular embodiment. In other embodiments, additional technical features and beneficial effects may be identified in specific embodiments that do not embody all embodiments. Attached Figure Description

[0018] Figure 1An application environment diagram of a resource scheduling method provided in this application embodiment; Figure 2 A flowchart illustrating a resource scheduling method provided in an embodiment of this application; Figure 3 A structural diagram of a resource scheduling device provided in an embodiment of this application; Figure 4 This is a schematic diagram of a computing device provided in an embodiment of this application. Detailed Implementation

[0019] The terms "first," "second," and "third," etc., used in this application specification and accompanying drawings are used to distinguish different objects, not to limit a specific order.

[0020] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0021] This application provides a resource scheduling method. To make the technical solution of this application clearer and easier to understand, the application scenarios of the technical solution of this application are described below with reference to the accompanying drawings. Figure 1 As shown, this figure is an application environment diagram provided by an embodiment of this application.

[0022] In this application environment, the server acts as the execution entity. A communication connection is established between the server and the terminal via a wired or wireless network. The terminal is used to display the policy execution results and resource data dashboards to resource personnel. Resource personnel can view the policy execution status and resource indicator values ​​issued by the server through the terminal. A data channel is established between the server and the target resource scheduling backend through an application programming interface. The server sends parameterized instructions to the target resource scheduling backend through this interface, and the target resource scheduling backend automatically executes the instructions.

[0023] To make the technical solution of this application clearer and easier to understand, the resource scheduling method provided by the embodiments of this application will be described below in conjunction with the above application scenarios. For example... Figure 2 As shown, this figure is a flowchart illustrating a resource scheduling method provided in an embodiment of this application. The resource scheduling method includes: S201. Obtain the initial resource data of the target object.

[0024] Initial resource data refers to multi-dimensional raw data obtained by the server from multiple business subsystems of the target object, reflecting the resource status of the target object. This includes, but is not limited to, material unit data, user behavior data, resource flow record transaction data, activity data, traffic exposure data, resource supplier resource data, inventory data, and resource anomaly handling record data.

[0025] Material unit data includes, but is not limited to, material unit codes, categories, standard resource allocation benchmark amounts, resource consumption benchmark values, and inventory tags; user behavior data includes, but is not limited to, user browsing, adding to cart, searching, and page dwell path records; resource transfer record transaction data includes, but is not limited to, resource transfer record number, transaction amount, payment time, and unique material unit identifier; activity data includes, but is not limited to, activity type, discount rules, budget amount, and material unit participation list; traffic exposure data includes, but is not limited to, channel source, total exposure, and clicks; resource supplier data includes, but is not limited to, resource supplier level, performance timeliness, and complaint records; inventory data includes, but is not limited to, material unit unique identifier inventory quantity, turnover days, and replenishment cycle; resource anomaly handling record data includes, but is not limited to, refund resource transfer records, dispute reasons, and refund rate.

[0026] For example, the server's acquisition of initial resource data for the target object is divided into two acquisition modes: real-time streaming acquisition and offline batch synchronization. The server accesses three types of high-frequency data—user behavior, resource flow record transactions, and traffic exposure—in real time through a message queue, with acquisition latency controlled within minutes. Every day at midnight, the server synchronizes inventory, resource supplier resources, resource anomaly handling records, and basic information data of material units through a data warehouse to complete offline batch acquisition. The server establishes data connection channels to connect to the target object's material unit database, user behavior tracking system, resource flow record system, resource deployment backend, traffic statistics system, resource supplier management system, inventory management system, and resource anomaly handling record processing system. It pulls the corresponding category of raw data from each system. After the data is pulled, the server performs basic cleaning operations on the raw data to remove duplicate, missing, and garbled invalid data, and finally integrates them to obtain complete initial resource data.

[0027] S202. Perform feature standardization processing on the initial resource data to obtain the target resource feature vector.

[0028] Feature standardization is a unified transformation operation performed by the server on the collected multi-source heterogeneous initial resource data, including cleaning, feature extraction, and numerical normalization. It converts various business data with inconsistent formats and units into numerical fields that can be calculated uniformly. For example, the server extracts user activity over 30 days, material unit turnover days, and return on investment (ROI) of activities, and then scales the indicators in different numerical ranges to the same unit range to complete the standardization transformation.

[0029] The target resource feature vector is a unified set of numerical values ​​generated by splicing together the features after the server completes the feature standardization process. The set carries all standardized business indicators for users, material units, activities, resource suppliers, and inventory. For example, the vector simultaneously contains user stratification values, material unit revenue values, inventory turnover values, and activity ROI values.

[0030] For example, the server receives initial resource data after basic cleaning and performs feature extraction and time-series analysis operations from five dimensions: users, material units, activities, resource suppliers, and inventory. For the user dimension, the server constructs user behavior sequences, statistically analyzes user activity, spending power, category preferences, activity participation, and reach frequency data for the past 7, 14, and 30 days, and generates user segmentation tags. For the material unit dimension, the server extracts tags for the material unit's category, price range, sales volume, conversion rate, turnover days, and revenue level, and calculates the sales trend slope and volatility coefficient of the material unit. For the activity dimension, the server calculates activity exposure conversion rate, redemption rate, return on investment (ROI), user participation rate, and budget consumption progress indicators. For the resource supplier dimension, the server quantifies the resource supplier's sales health, fulfillment capability, and risk level of resource anomaly handling records. For the inventory dimension, the server identifies inventory turnover days, backlog risk level, and stockout risk level tags. After the server completes full-dimensional feature extraction, it performs normalization and standardization operations on all extracted features to eliminate the differences in units between different indicators. Finally, it concatenates and merges all standardized numerical features to generate a target resource feature vector with a unified dimension.

[0031] S203. Problem identification is performed based on the target resource feature vector to obtain structured decision objects.

[0032] One possible approach is to perform temporal anomaly detection and hierarchical drill-down on the feature vector of the target resource to obtain resource anomaly data; match the resource anomaly data with a preset standard problem database to determine problem type labels; and determine structured decision objects based on problem type labels, resource anomaly data, and scene features corresponding to the target object.

[0033] The structured decision object is a standardized structured data package generated by the server by integrating problem type tags, resource anomaly data, and scenario features. Problem type tags indicate the category of resource problems currently existing in the target object; resource anomaly data is the location result data, including the location of the anomaly subject and its impact range; scenario features indicate the current resource environment status of the target object, including activity cycle nodes, inventory levels, price competition index, and user segmentation attributes. For example, if the server identifies an anomaly in the conversion rate of the target object's material unit (unique identifier 12345), the generated structured decision object includes the problem type tag "high exposure, low conversion material unit," the anomaly subject "material unit unique identifier 12345," the impact range "material unit transaction total percentage 2.3%, anomaly lasting 4 days," and the scenario features "sufficient inventory, price competition index 0.85, flat sales period, general user segmentation."

[0034] Time-series anomaly detection is a computational operation performed by the server to determine the fluctuation of numerical indicators within the feature vector of a target resource over time. For example, the server continuously reads the standardized values ​​of the material unit conversion rate for 7 days and identifies the periods when the indicators deviate from the normal fluctuation range.

[0035] Drill-down by level is a positioning operation in which the server breaks down the range of anomalies step by step from coarse to fine. The level order is channel, category, resource supplier, material unit, and inventory holding unit (unique identifier of material unit). For example, the server first identifies that the total transaction amount of material units in all categories has decreased, then drills down to the apparel category, continues to locate the dress sub-category, and finally locks the unique identifier of the corresponding material unit.

[0036] The preset standard problem library is a standardized problem matching dataset stored locally on the server. It stores the judgment rules and problem tags corresponding to various target object resource anomalies. For example, the library has preset tags such as "high exposure and low conversion of material unit", "inventory backlog", and "mass user churn" and their corresponding judgment thresholds.

[0037] For example, the server pre-configures historical benchmark time-series samples, normal fluctuation threshold ranges, and a time-series sliding window of 7 calendar days for each time-series indicator within the target resource feature vector. The time-series indicators to be detected include standardized values ​​that are continuously updated over time, such as 7 / 14 / 30-day user activity, material unit conversion rate, inventory turnover days, activity return on investment, daily sales of resource suppliers, user click-through rate, and incremental total transaction amount of material units. The server reads the target resource feature vector, splits all numerical fields within the target resource feature vector, filters out static features that do not change over time, such as category fixed income, basic resource consumption benchmark value of material units, and resource supplier level, and removes them. It then extracts the standardized time-series values ​​for multiple consecutive days and groups them by indicator dimension to generate corresponding single-indicator time-series arrays. The server iterates through each group of time-series arrays to verify data integrity. For missing, null, or garbled daily values, it retrieves the historical average of the same period for the same indicator at the same level to fill in the missing values. At the same time, it uniformly pads the array length to the standard length of the sliding window to ensure that all time-series arrays are of consistent specifications.

[0038] After data alignment is completed, the server calculates the arithmetic mean of the values ​​within the window for each time series array as the short-term benchmark mean, and simultaneously calculates the standard deviation of the window values ​​as the benchmark for normal fluctuation range. Then, it sets a floating coefficient to define the normal range in combination with the target object resource time period. The floating coefficient is 1.5 during the flat sales period and 2.5 during the pre-sale and promotion period. The normal upper limit is equal to the benchmark mean plus the product of the floating coefficient and the standard deviation, and the normal lower limit is equal to the benchmark mean minus the product of the floating coefficient and the standard deviation.

[0039] The server compares the daily indicator values ​​in the array with the normal range one by one in chronological order. Values ​​exceeding the upper limit are marked as positive anomalies, values ​​below the lower limit are marked as negative anomalies, and values ​​within the range are considered normal data. Subsequently, the server counts the number of consecutive abnormal time points under the same indicator. If an anomaly occurs on only one day, it is judged as instantaneous data jitter and the anomaly mark is cleared. If the deviation from the range occurs for two or more consecutive days, it is determined that there is a real resource time sequence anomaly. The indicator identifier, the start time of the anomaly, the number of days of the anomaly, and the difference between the indicator deviation are recorded simultaneously. The server retrieves the field mapping relationships stored when constructing the target resource feature vector, and reverse-binds the time-series indicators determined to be truly abnormal to the corresponding business entities. For example, abnormal material unit conversion rate is bound to the inventory holding unit (unique identifier of material unit) ID, abnormal total transaction amount of category material unit is bound to the category code, abnormal user activity decline is bound to the user stratification tag, and abnormal sales of resource suppliers is bound to the resource supplier ID. The abnormal indicator name, abnormal duration days, indicator deviation range, and associated business entities are integrated to form intermediate abnormal records, which are directly sent to the hierarchical drill-down process as the input data source to narrow down the abnormal range layer by layer. The entire set of time-series abnormality detection calculations is executed automatically by the server, and only the general statistical formulas of mean and standard deviation are used to complete the interval determination, without any custom special calculation formulas.

[0040] S204. Based on the numerical constraints of material unit inventory, category revenue, user classification and access control, and resource allocation budget, interval threshold calculations are performed on the structured decision object to obtain the complete parameterized allowable strategy numerical space.

[0041] The inventory value constraint of a material unit is a numerical range rule set by the server for the existing inventory and turnover target of the inventory holding unit (the unique identifier of the material unit) to limit the upper limit range of resource allocation. For example, when the inventory is sufficient, the server relaxes the upper limit range of resource allocation, allowing the generation of strategies with higher resource subsidies; when the inventory is insufficient, the server tightens the upper limit range of resource allocation to avoid excessive promotion leading to stockouts.

[0042] Category revenue constraints are profit control rules set by the server based on the category's resource consumption benchmark and break-even point. These rules define the minimum thresholds for resource subsidies for material units and electronic resource replenishment vouchers, preventing category losses from resource allocation activities. This constraint also defines two control parameters: a minimum resource subsidy limit and a minimum resource replenishment voucher amount. The minimum resource subsidy limit is the minimum allowable amount of resource subsidies that can be issued when allocating material units; the resource replenishment voucher is a targeted digital resource replenishment carrier, and its issuance amount also has a minimum control threshold. Both parameters are calculated based on the category's minimum resource revenue break-even point to prevent excessive resource allocation from causing overall losses to exceed limits.

[0043] User classification and outreach control numerical constraints are user resource control rules set by the server based on user stratification tags and historical push records. They are used to limit the user group scope that can be pushed to resource delivery activities and the maximum number of outreaches per user within a single period.

[0044] The resource allocation budget constraint is a set of financial control rules set by the server based on the total activity budget and input-output targets. It is used to limit the total amount of overall resource allocation subsidies and the maximum subsidy amount per user.

[0045] Interval threshold calculation is a batch operation in which the server reads various business parameters carried by the structured decision object, substitutes them into four types of constraints to complete numerical boundary calculations, and outputs the allowable value range of various resource deployment parameters.

[0046] Full parameterization allows the strategy numerical space to be the set of all parameter value ranges obtained by the server integrating the four types of constraints: inventory, revenue, users, and budget. This set serves as the hard numerical boundary for the subsequent generation of resource strategies.

[0047] The process of calculating the interval threshold for a structured decision object can be as follows: Obtain the associated entities, product categories, inventory levels, and user classification attributes of the structured decision object; calculate the upper limit interval value of resource allocation based on the current inventory value and turnover target value of the corresponding material unit of the associated entity; calculate the lower limit threshold value of resource subsidies and the lower limit threshold value of electronic voucher denomination based on the lower limit value of the revenue floor and the break-even point value corresponding to the product category; determine the scope of interactive user groups based on the user classification tags corresponding to the user classification attributes; determine the upper limit value of the number of interactions per user within a preset time period based on the historical interaction frequency records corresponding to the user classification attributes; determine the upper limit value of the total resource allocation budget corresponding to the structured decision object; determine the upper limit value of resource allocation per user based on the resource input-output target value; and determine the complete parameterized allowed strategy value space based on the upper limit interval value of resource allocation, the lower limit threshold value of resource subsidies, the lower limit threshold value of electronic voucher denomination, the scope of interactive user groups, the upper limit value of the number of interactions, the upper limit value of the total resource allocation, and the upper limit value of resource allocation per user. It should be noted that the equilibrium point value is the minimum resource revenue equilibrium point value for a category. The calculation logic is as follows: First, extract the standard resource allocation benchmark amount and resource consumption benchmark value of the material unit under the category to obtain the net resource revenue per unit; then divide the total fixed resource consumption of the category by the net resource revenue per unit to obtain the allocation scale equilibrium point that just covers all basic resource losses. When calculating the lower limit of resource subsidies and the minimum amount of resource replenishment vouchers, this equilibrium point is used as a constraint: if the resource subsidy is too high, causing the actual allocation scale to be lower than the equilibrium point, it will cause the overall resource loss to exceed the budget. Therefore, it is necessary to set a minimum subsidy limit based on the equilibrium point to ensure that the allocation scale is not lower than the equilibrium point and to avoid resource deficits in the category.

[0048] For example, the server reads the structured decision object and parses four fields from it: associated subject, category, inventory level, and user classification attribute. The associated subject is a field in the structured decision object that indicates the specific object where the resource problem occurs, such as the unique identifier of the material unit (12345). The category is a field in the structured decision object that indicates the category of the material unit to which the associated subject belongs, such as mobile phone accessories. The inventory level is a field in the structured decision object that indicates the current inventory level of the associated subject, such as sufficient, moderate, or insufficient. The user classification attribute is a field in the structured decision object that indicates the category to which the target user belongs, such as high-value user, potential user, or ordinary user.

[0049] Based on the associated entity, the server queries the target object's inventory management system for the current inventory value and turnover target value of the corresponding material unit. The current inventory value of the material unit is the quantity of remaining inventory recorded in the inventory management system for the associated entity; for example, the current inventory value of the material unit with the unique identifier 12345 is 500 units. The turnover target value is the preset inventory turnover target value that the associated entity should complete within a preset time period, as set in the inventory management system. The server compares the current inventory value of the material unit with the preset inventory adequacy threshold. When the current inventory value of a material unit is greater than or equal to the sufficient inventory threshold, the server determines the inventory level as sufficient and sets the upper limit range of resource allocation as the first range, for example, [0.80, 0.95] indicates that the upper limit range of resource allocation is 80% to 95% of the original value. When the current inventory value of a material unit is less than the sufficient inventory threshold but greater than or equal to the insufficient inventory threshold, the server determines the inventory level as moderate and sets the upper limit range of resource allocation as the second range, for example, [0.85, 0.95]. When the current inventory value of a material unit is less than the insufficient inventory threshold, the server determines the inventory level as insufficient and sets the upper limit range of resource allocation as the third range, for example, [0.90, 0.95]. The server outputs the determined upper limit range of resource allocation. The upper limit range of resource allocation is a numerical range containing an upper limit and a lower limit, used to limit the maximum and minimum allowable range of resource allocation in strategy generation. The sufficient inventory threshold and the insufficient inventory threshold are fixed values ​​pre-stored in the server for determining the inventory level.

[0050] The server queries the target object's material unit management system for the corresponding minimum profit margin and break-even point value for that product category. The minimum profit margin is the minimum allowable profit rate for that product category's material units, preset in the material unit management system. For example, the minimum profit margin for mobile phone accessories is 20%. The break-even point is the break-even threshold for sales volume or sales revenue calculated in the material unit management system based on the resource consumption benchmark price and standard resource allocation benchmark amount for that product category's material units. The server sets the minimum profit margin as the lower limit threshold for resource subsidies. For example, when the minimum profit margin is 20%, the lower limit threshold for resource subsidies is 8.5%, meaning that the resource subsidy level in subsequent strategy generation cannot be lower than 8.5%. The server calculates the lower limit threshold for electronic voucher face value based on the break-even point value. Specifically, the server obtains the average standard resource allocation benchmark amount for that product category's material units, subtracts the single-item resource consumption benchmark value corresponding to the break-even point value from the average standard resource allocation benchmark amount to obtain the profit margin for each material unit, and multiplies the profit margin by a preset electronic voucher ratio coefficient to obtain the lower limit threshold for electronic voucher face value. For example, when the standard resource allocation benchmark for a material unit is 99 yuan, and the break-even point corresponds to a single-item resource consumption benchmark of 85 yuan, the profit margin is 14 yuan, and the electronic voucher ratio coefficient is 1.0, then the lower limit threshold for the electronic voucher's face value is 14 yuan. The server outputs the lower limit threshold for resource subsidies and the lower limit threshold for the electronic voucher's face value. The lower limit threshold for resource subsidies is the minimum allowable value for resource subsidies, and the lower limit threshold for the electronic voucher's face value is the minimum allowable value for the face value of resource supply vouchers or electronic vouchers.

[0051] The server queries the user management system for the corresponding user category tag based on the user category attribute. User category tags are categorized in the user management system based on users' historical consumption behavior, activity level, and value, such as high-value users, potential users, and ordinary users. The server then matches the user category tag with pre-defined rules for interactive user groups to determine the scope of the interactive user group. These rules are pre-stored on the server and are used to limit the scope of interactive users corresponding to different user category tags. For example, if the user category tag is "high-value user," the interactive user group includes users who have made purchases in the past 30 days; if the user category tag is "potential user," the interactive user group includes users who have browsed the page but not placed an order in the past 7 days; and if the user category tag is "ordinary user," the interactive user group includes users who have logged in in the past 30 days. The server outputs the matched scope of the interactive user group.

[0052] The server queries the target user's user behavior system for historical interaction frequency records corresponding to that user category attribute, based on the user's category attributes. These historical interaction frequency records are stored in the user behavior system showing the number of resource delivery interactions received by each user within that category within a preset historical time window. The server retrieves preset interaction frequency control rules, which are pre-stored rules used to limit the maximum number of interactions that different user categories can have within a preset time period. For example, the maximum number of interactions for high-value users is 2 times / 7 days, for potential users it's 3 times / 7 days, and for ordinary users it's 1 time / 7 days. The server matches the user category tag with the corresponding preset time period and interaction frequency limit, using this limit as the maximum number of interactions for a single user within the preset time period. The server outputs this maximum number of interactions for a single user within the preset time period. For example, a maximum interaction frequency of 2 times / 7 days for high-value users means that a single high-value user can be reached a maximum of 2 times within a 7-day time period.

[0053] The server obtains the total resource allocation budget and the resource input-output target value corresponding to the structured decision object. The total resource allocation budget is the upper limit of the total resource allocation funds allocated to the target object for this resource activity; the resource input-output target value is a preset target value in the server used to measure the efficiency of resource allocation, representing the expected return multiple per unit of resource investment. For example, a resource input-output target value of 5 means that every 1 yuan invested is expected to generate 5 yuan in return. The server determines the total resource allocation budget as the upper limit of the total resource allocation, indicating that the total amount of all resource allocation subsidies in subsequent strategy generation cannot exceed this value. The server calculates the upper limit of resource allocation per user based on the resource input-output target value. Specifically, the server obtains the average historical consumption amount and average historical conversion rate of users in the corresponding category, multiplies the average historical consumption amount by the average historical conversion rate, and then multiplies it by the resource input-output target value to obtain the upper limit of resource allocation per user. For example, when the average historical consumption amount is 100 yuan, the average historical conversion rate is 10%, and the resource input-output target value is 5, the upper limit of resource allocation per user is 100 × 10% × 5 = 50 yuan. The server outputs the total resource allocation limit and the single-user resource allocation limit. The single-user resource allocation limit represents the maximum allowed amount of resource allocation subsidies a single user can receive.

[0054] The server combines the calculated upper limit range of resource allocation, lower limit threshold of resource subsidies, lower limit threshold of electronic voucher value, range of interactive user groups, upper limit of interaction times, upper limit of total resource allocation amount, and upper limit of resource allocation per user into a complete parameterized allowable policy value space.

[0055] S205. Perform policy generation processing within the parameter boundaries of the fully parameterized policy numerical space to obtain candidate resource optimization policies.

[0056] Candidate resource optimization strategies are combinations of one or more resource deployment actions output by the server after strategy generation and processing, used to solve the current resource problems of the target object. For example, "targeted discount + details page optimization" means a combination of applying a 12% discount to a specific material unit and exposing that material unit on the details page and in the recommended position.

[0057] One possible approach involves retrieving matching historical knowledge units from a knowledge base based on the problem type, product category, inventory level, and user classification attributes of the structured decision object, thus obtaining a recall knowledge unit set; determining a reference knowledge unit set based on the comprehensive contribution value of each knowledge unit in the recall knowledge unit set; and performing strategy deduction within the parameter boundaries of the complete parameterized allowable strategy value space based on the structured decision object, the complete parameterized allowable strategy numerical space, and the reference knowledge unit set, to obtain candidate resource optimization strategies.

[0058] The knowledge base is a dataset that is persistently stored locally on the server. The dataset stores multiple complete historical knowledge units. Each complete historical knowledge unit synchronously records historical resource scenarios, past resource strategies, the effects of all resources after the strategy was implemented, and the corresponding strategy's comprehensive contribution value, which is used to provide historical cases for reference in this strategy simulation.

[0059] Historical knowledge units are structured data records created by the server that combine the resource scenarios, strategy schemes, and effect data corresponding to strategies that have been executed in the past and whose overall contribution value has reached a preset threshold.

[0060] The recalled knowledge unit set is a dataset formed by the server retrieving and filtering all historical knowledge units from the knowledge base based on matching conditions.

[0061] The overall strategy contribution value is a comprehensive quantitative evaluation value obtained by weighting and summing the values ​​of four dimensions generated by the implementation of past resource strategies corresponding to a single historical knowledge unit: the percentage increase in total transaction amount of platform material units, the percentage increase in user value, the percentage increase in revenue for resource suppliers, and the percentage of resource deployment and resource loss. For example, the overall strategy contribution value = first weight coefficient (e.g., 0.4) × percentage increase in total transaction amount of platform material units (e.g., 38%) + second weight coefficient (e.g., 0.3) × percentage increase in user value (e.g., 15%) + third weight coefficient (e.g., 0.2) × percentage increase in revenue for resource suppliers (e.g., 20%) - fourth weight coefficient (e.g., 0.1) × percentage of resource deployment and resource loss (e.g., 10%).

[0062] The reference knowledge unit set is a set of highly suitable historical knowledge units that the server selects from all historical knowledge units in the recalled knowledge set according to the strategy and the comprehensive contribution value, and retains them.

[0063] Strategy scheme derivation is a computational process in which the server integrates three types of input information and strictly limits the process to the boundary of all parameters in the fully parameterized policy numerical space, generating multiple feasible resource optimization schemes in batches.

[0064] For example, the server reads the structured decision object and parses four fields from it: problem type, product category, inventory level, and user classification attribute. The problem type is a field in the structured decision object that indicates the category of resource problems currently existing for the target object, such as "high exposure, low conversion material unit"; the product category is a field in the structured decision object that indicates the category of material units to which the problem-related entity belongs, such as "mobile phone accessories"; the inventory level is a field in the structured decision object that indicates the current inventory level of the related entity, such as "sufficient"; and the user classification attribute is a field in the structured decision object that indicates the category to which the target user belongs, such as "general".

[0065] The server constructs search criteria based on the parsed question type, product category, inventory level, and user classification attribute. The server uses question type, product category, and inventory level as primary search fields, and user classification attribute as secondary filtering fields. The server compares the primary search fields with the question type, product category, and inventory status in the resource scenario fields of each historical knowledge unit in the knowledge base, calculating the matching degree between each historical knowledge unit and the search criteria. The matching degree is a value calculated by the server based on the consistency of field values ​​between the search fields and the resource scenario fields of the knowledge units; the matching degree is highest when the values ​​of all three primary search fields match the corresponding fields of the knowledge unit. The server sorts all historical knowledge units in the knowledge base in descending order of matching degree, and determines the historical knowledge units with matching degrees higher than a preset matching degree threshold as the recall knowledge unit set. If the number of knowledge units in the recall knowledge unit set is less than a preset recall quantity threshold, the server uses the user classification attribute as a supplementary search field to expand the search scope and re-search until the number of knowledge units in the recall knowledge unit set reaches the preset recall quantity threshold.

[0066] The server retrieves the comprehensive contribution value from the performance evaluation field of each historical knowledge unit in the recalled knowledge unit set. The server sorts the historical knowledge units in the recalled knowledge unit set according to their comprehensive contribution values ​​from highest to lowest, and selects a preset number of historical knowledge units with the highest comprehensive contribution values ​​as the reference knowledge unit set. The preset number is configured locally on the server; for example, if the preset number is 3, the server selects the top 3 historical knowledge units with the highest comprehensive contribution values ​​as the reference knowledge unit set.

[0067] The server combines structured decision objects, a fully parameterized allowable strategy numerical space, and a set of reference knowledge units to construct the input data for strategy generation. Specifically, the server uses the problem type, associated subjects, scope of impact, and resource scenario characteristics from the structured decision objects as descriptive information for the problem to be solved; the server uses the upper limit range of resource allocation, the lower limit threshold of resource subsidies, the lower limit threshold of electronic voucher denomination, the range of interactive user groups, the upper limit of interaction times, the upper limit of total resource allocation, and the upper limit of resource allocation per user from the fully parameterized allowable strategy numerical space as parameter boundary constraints during strategy generation; and the server uses the resource scenario fields, strategy scheme fields, and effect evaluation fields of each knowledge unit in the set of reference knowledge units as historical experience references.

[0068] The server performs strategy deduction calculations on the input data constructed above, under the parameter boundary constraints of the fully parameterized allowed strategy numerical space. During the deduction calculation, the server selects knowledge units from the reference knowledge unit set that match the current problem type and category based on the problem type and category in the structured decision object, and extracts strategy solution fields such as strategy type, core parameters, applicable scope, and execution channel. The server compares the core parameter fields in the extracted strategy solution with each parameter boundary in the fully parameterized allowed strategy numerical space item by item. When the value of a core parameter exceeds the corresponding boundary, the server corrects the parameter value down to the boundary value; when the value of a core parameter is within the boundary range, the server retains the parameter value. The server combines the corrected or retained parameter value with the strategy type, applicable scope, and execution channel to form a candidate resource optimization strategy. The server repeats the above operation to generate a preset number of candidate resource optimization strategies according to a preset candidate strategy quantity threshold, where the parameter fields of each candidate resource optimization strategy are within the boundary range of the fully parameterized allowed strategy numerical space.

[0069] S206. Perform hard rule verification on the candidate resource optimization strategies to obtain the candidate resource optimization strategies that pass the hard rule verification.

[0070] The hard rule verification includes four items: minimum constraint line verification of resource subsidies, user interaction frequency verification, settlement rule verification of resource suppliers, and compliance verification of platform resource allocation. Among them, the minimum constraint line verification of resource subsidies is used to check whether the resource subsidy value and resource supply voucher amount in the candidate scheme meet the minimum control limit set by the category revenue constraint mentioned above. If it is lower than the limit, it will be corrected to the standard lower limit to prevent the category allocation scale from falling below the resource revenue balance point and resulting in resource loss.

[0071] The resource subsidy minimum constraint line verification is a server-side operation that checks whether the resource subsidy level and electronic voucher value in the candidate resource optimization strategy meet the minimum threshold for resource subsidy and electronic voucher value. For example, if the resource subsidy level in the candidate strategy is 80% and the minimum threshold for resource subsidy is 85%, the server determines that the strategy has failed the minimum constraint line verification. The user interaction frequency verification is a server-side operation that checks whether the number of interactions per user within a preset time period in the candidate resource optimization strategy exceeds the maximum interaction limit. For example, if the number of interactions per user within 7 days in the candidate strategy is 3, and the maximum interaction limit is 2 times / 7 days, the server determines that the strategy has failed the user interaction frequency verification. The resource supplier settlement rule verification is a server-side operation that checks whether the resource supplier settlement-related parameters in the candidate resource optimization strategy comply with the target platform's resource supplier settlement rules and transaction fee rules. The platform compliance rule verification is a server-side operation that checks whether the resource placement content in the candidate resource optimization strategy complies with the target platform's advertising placement specifications, promotional activity rules, and industry regulatory requirements.

[0072] The parameter correction operation is the operation by which the server adjusts the parameters in the candidate resource optimization strategy that exceed the boundary value of the fully parameterized allowed strategy numerical space down to the boundary value. For example, when the resource subsidy intensity in the candidate strategy is 75% and the lower limit threshold of the resource subsidy is 85%, the server corrects the resource subsidy intensity parameter to 85%.

[0073] For example, the server reads a set of candidate resource optimization strategies. This set contains a preset number of candidate resource optimization strategies, each of which includes fields such as strategy type, core parameters, applicable scope, effective time, and execution channel. The core parameter fields include the resource subsidy amount, the electronic voucher value, the number of interactions per user within a preset time period, and the resource supplier settlement ratio.

[0074] The server extracts one candidate resource optimization strategy from the set of candidate resource optimization strategies one by one, and performs a minimum resource subsidy constraint line check on the candidate strategy. The specific process for verifying the minimum constraint line of resource subsidies is as follows: The server extracts the resource subsidy intensity value and the electronic voucher denomination value from the core parameter fields of the candidate strategy; the server compares the resource subsidy intensity value with the lower limit threshold of resource subsidies; when the resource subsidy intensity value is less than the lower limit threshold, the server marks the candidate strategy as failing the minimum constraint line verification of resource subsidies and corrects the resource subsidy intensity value to the lower limit threshold. For example, if the lower limit threshold of resource subsidies is 8.5% and the resource subsidy intensity of the candidate strategy is 8% of the original value, the server corrects the resource subsidy intensity to 8.5%; the server compares the electronic voucher denomination value with the lower limit threshold of electronic voucher denomination; when the electronic voucher denomination value is less than the lower limit threshold, the server marks the candidate strategy as failing the minimum constraint line verification of resource subsidies and corrects the electronic voucher denomination value to the lower limit threshold. For example, if the lower limit threshold of electronic voucher denomination is 14 yuan and the electronic voucher denomination of the candidate strategy is 10 yuan, the server corrects the electronic voucher denomination to 14 yuan.

[0075] The server performs user interaction frequency verification on candidate strategies that have passed the minimum constraint line verification for resource subsidies. The specific process of user interaction frequency verification is as follows: The server extracts the number of interactions per user within a preset time period from the core parameter fields of the candidate strategy; the server compares this number of interactions with the upper limit of interactions per user within the preset time period; when the number of interactions is greater than the upper limit, the server marks the candidate strategy as failing the user interaction frequency verification and corrects the number of interactions to the upper limit. For example, if the upper limit is 2 times / 7 days and the candidate strategy has 3 interactions / 7 days, the server will correct the number of interactions to 2 times / 7 days.

[0076] The server performs resource provider settlement rule verification on candidate strategies that have passed the user interaction frequency verification. The specific process of resource provider settlement rule verification is as follows: The server extracts the resource provider settlement ratio value from the core parameter field of the candidate strategy; the server obtains the preset resource provider settlement rules of the target object, which include constraints such as the lower and upper limits of transaction fee ratios and settlement cycles; the server compares the resource provider settlement ratio value with the lower and upper limits of transaction fee ratios; when the resource provider settlement ratio value exceeds the range of the lower and upper limits of transaction fee ratios, the server marks the candidate strategy as failing the resource provider settlement rule verification and filters the candidate strategy.

[0077] The server performs platform compliance rule verification on candidate strategies that have completed the resource supplier settlement rule verification. The specific process of platform compliance rule verification is as follows: The server extracts the strategy type field and execution channel field from the candidate strategy; the server obtains the platform compliance rule library preset by the target object, which contains a list of prohibited words under the Advertising Law, promotional activity specifications, and industry regulatory compliance standards; the server compares the strategy type and execution channel of the candidate strategy with the platform compliance rule library. If the candidate strategy contains prohibited words under the Advertising Law or violates promotional activity specifications, the server marks the candidate strategy as failing the platform compliance rule verification and filters the candidate strategy.

[0078] The server outputs candidate strategies that pass all four hard rule checks to the set of candidate resource optimization strategies that have passed hard rule checks, in the order they passed the checks. For candidate strategies that fail the checks but have already undergone parameter correction, the server outputs the corrected candidate strategy to the set of candidate resource optimization strategies that have passed hard rule checks. For filtered candidate strategies, the server records the reason for the filtering and discards the strategy, not adding it to the set of candidate resource optimization strategies that have passed hard rule checks. The server finally outputs a set of candidate resource optimization strategies that have passed hard rule checks, where each candidate strategy satisfies the minimum resource subsidy constraint, user interaction frequency constraint, resource supplier settlement rule constraint, and platform compliance rule constraint.

[0079] S207. Based on the multi-objective comprehensive scoring rules, the candidate resource optimization strategies that have passed the hard rule verification are sorted to determine the target strategy; the target strategy is converted into parameterized instructions and the parameterized instructions are sent to the target resource scheduling backend for execution.

[0080] The multi-objective comprehensive scoring rule is a pre-configured rule in the server used to comprehensively score and rank the candidate resource optimization strategies that have passed the hard rule verification. The rule includes four scoring indicators: the improvement rate of total transaction amount of material units, the improvement rate of user retention, the improvement rate of resource supplier revenue, and the resource input-output ratio. Each indicator has a corresponding dynamic weight coefficient.

[0081] The target strategy is the resource optimization scheme with the highest multi-objective comprehensive score among all candidate resource optimization strategies that have passed hard rule verification. Parameterized instructions are the server's decomposition and mapping of the textual target strategy into a set of structured fields recognizable by the resource scheduling backend. These fields include strategy type, applicable scope, various resource deployment parameters, effective time, and execution channel, and can be automatically issued and executed via an interface. The target resource scheduling backend is the business service system that carries out resource deployment configuration, material unit management, user outreach, and subsidy disbursement functions for the target object.

[0082] For example, the server breaks down the natural language description of the target strategy into a strategy type field, an applicable scope field, a core parameter field, an effective time field, and an execution channel field. The strategy type field indicates the category of resource action to which the target strategy belongs, such as "targeted discount", "issuance of resource supply vouchers", and "details page optimization"; the applicable scope field indicates the scope of material units, user groups, or resource suppliers targeted by the target strategy, such as "material unit unique identifier 12345" and "users who have browsed but not placed an order in the past 7 days"; the core parameter field indicates the key parameters when the target strategy is executed, such as "12% discount" and "50 off for purchases over 299"; the effective time field indicates the start time and duration of the target strategy, such as "effective immediately" and "effective from 00:00 on July 28, 2026"; and the execution channel field indicates the placement and reach of the target strategy, such as "material unit details page + recommendation slot" and "application push".

[0083] The server maps the policy type field to the interface name of the target resource scheduling backend. The server pre-stores a mapping table between policy types and resource backend interface names. For example, "targeted discount" maps to the "Create Material Unit Promotion" interface in the resource delivery system, "Resource Supply Voucher Issuance" maps to the "Create Resource Supply Voucher Activity" interface in the resource delivery system, and "Details Page Optimization" maps to the "Update Material Unit Details Page Configuration" interface in the material unit management system. The server converts the policy type field to the corresponding interface name based on this mapping table.

[0084] The server maps the scope field to the target object identifier of the interface. Based on the content of the scope field, the server queries the corresponding object identifier from the target object's material unit management system, user management system, or resource supplier management system. For example, when the scope field is "material unit unique identifier 12345", the server maps it to the material unit identifier corresponding to material unit unique identifier 12345 in the material unit management system; when the scope field is "users who have browsed but not placed an order in the last 7 days", the server maps it to the user group identifier that meets the condition in the user management system.

[0085] The server maps core parameter fields to execution parameters of the interface. Based on the interface type corresponding to the policy type field, the server converts the parameter values ​​in the core parameter fields into the parameter format required by the interface. For example, if the execution parameters for the interface corresponding to "targeted discount" are "resource subsidy level" and "applicable material unit list," the server converts "12% discount" in the core parameter field into the resource subsidy level parameter value of 0.88.

[0086] The server maps the effective time field to the interface's scheduling time. The server converts the time description in the effective time field into a standard time format that the interface can recognize. For example, "effective immediately" is converted into the current system timestamp, and "effective from 00:00 on July 28, 2026" is converted into the corresponding date and time format.

[0087] The server maps the execution channel field to the placement location of the interface. Based on the content of the execution channel field, the server determines the specific placement parameters of the target strategy in the target resource scheduling backend. For example, "material unit details page + recommendation slot" is mapped to the details page recommendation slot identifier and the homepage recommendation slot identifier.

[0088] The server assembles the mapped policy type field (corresponding to the interface name), the scope of application field (corresponding to the target object identifier), the core parameter field (corresponding to the execution parameters), the effective time field (corresponding to the scheduling time), and the execution channel field (corresponding to the delivery location) into a structured instruction. The server then sends the structured instruction to the target resource scheduling backend via the Application Programming Interface (API), where the backend executes the instruction.

[0089] Based on the above, the resource scheduling method obtains the target resource feature vector by acquiring the initial resource data of the target object and performing feature standardization processing. Based on this feature vector, resource problems are identified and structured decision objects are generated. Then, based on the numerical constraints of material unit inventory, category revenue, user classification and access control, and resource allocation budget, interval threshold calculations are performed on the structured decision objects to obtain a complete parameterized allowable strategy numerical space. Within the parameter boundaries of this space, strategy generation processing is performed to obtain candidate resource optimization strategies. Finally, hard rule verification is performed on the candidate strategies, and the target strategy is determined by multi-objective comprehensive scoring and ranking. The target strategy is then converted into parameterized instructions and sent to the target resource scheduling backend for execution. Before executing strategy generation, this solution first transforms four types of resource constraints—inventory, revenue, user reach, and resource allocation budget—into numerical parameter boundaries through interval threshold calculations. This forms a complete parameterized numerical space for allowed strategies, ensuring that the entire strategy generation process is completed within these parameter boundaries. This guarantees from the outset that the generated candidate strategies will not exceed resource constraints. Simultaneously, after strategy generation, a hard rule verification process is performed. Candidate strategies undergo secondary verification based on minimum resource subsidy constraints, user interaction frequency, resource supplier settlement rules, and platform compliance rules. Strategies that do not meet the constraints are filtered or corrected, forming a dual guarantee mechanism of pre-constraint limitation and post-verification filtering. Furthermore, through structured decision objects, abnormal information in the original resource data is transformed into standardized decision inputs. A multi-objective comprehensive scoring rule is used to comprehensively rank candidate strategies from multiple dimensions, including the increase rate of total transaction amount per material unit, the increase rate of user retention, the increase rate of resource supplier revenue, and the resource input-output ratio. This ensures that the final target strategy achieves optimal comprehensive performance across multiple dimensions while satisfying all resource constraints. Ultimately, this improves the feasibility of strategy implementation.

[0090] Based on the above embodiments, after the parameterized instructions are executed in the target resource scheduling background, resource indicator values ​​are generated. This embodiment also provides a detailed explanation of the process of feedback optimization based on resource indicator values, specifically including: Obtain the resource indicator values ​​after the parameterized instruction is executed; calculate the contribution value based on the resource indicator values ​​to obtain the overall strategy contribution value; compare the overall strategy contribution value with the preset target threshold to obtain the comparison result; perform feedback optimization processing based on the comparison result to obtain the feedback optimization processing result; wherein, the feedback optimization processing includes at least one of the first iteration processing, the second iteration processing, and the weight calibration processing.

[0091] The feedback optimization process can be as follows: When the comparison result shows that the overall contribution value of the strategy is greater than or equal to the preset threshold, the target strategy is stored as a new knowledge unit in the knowledge base, resulting in an updated knowledge base; the ranking of each knowledge unit in the updated knowledge base is optimized to obtain the updated ranking weight and retrieval priority; when the cumulative number of knowledge units in the knowledge base exceeds the preset sample size threshold, the mapping relationship between the inventory value of the material unit and the upper limit range of resource allocation is refitted to obtain the updated inventory allocation mapping relationship; the mapping relationship between the category revenue value and the lower limit threshold of resource subsidies is refitted to obtain the updated revenue-resource subsidy mapping relationship; the strategy generation processing parameters are fine-tuned to obtain the fine-tuned strategy generation processing parameters; and the weight coefficients of each indicator in the multi-objective comprehensive scoring rule are calibrated to obtain the updated weight coefficients.

[0092] Resource index values ​​are various quantitative resource data collected by the server after the target resource scheduling background runs parameterized instructions to complete resource activities, such as the total transaction amount of platform material units, user retention data, resource supplier revenue, resource deployment and resource loss, and user interaction frequency.

[0093] The overall contribution value of the strategy is a quantitative evaluation value used to judge the effectiveness of the entire set of target strategies after the server collects resource index values ​​from all dimensions and calculates them according to a fixed weighting formula.

[0094] The preset threshold is a fixed judgment value pre-stored locally on the server, which serves as the dividing line between effective and invalid resource policies.

[0095] The comparison result is the judgment conclusion output by the server after comparing the overall contribution value of the strategy with the preset threshold, which is divided into two categories: qualified and unqualified.

[0096] Feedback optimization processing is a system self-optimization operation performed by the server based on the comparison results, which includes any one or more of the following three types of operations: first iteration processing, second iteration processing, and weight calibration processing.

[0097] The first iteration is a lightweight optimization operation that only updates the ranking of knowledge units and retrieval weights within the knowledge base, without requiring refitting of mapping relationships or fine-tuning of strategies to generate processing parameters.

[0098] The second iteration is a deep optimization operation performed after the knowledge base sample size reaches the target. It includes refitting the two types of constraint mapping relationships and fine-tuning the strategy to generate processing parameters.

[0099] Weight calibration is an optimization operation that updates the weight coefficients of each dimension within the multi-objective comprehensive scoring rule.

[0100] The preset sample size threshold is a critical value for the cumulative number of valid knowledge units in the knowledge base, pre-set by the server. The second iteration is initiated only when this value is reached. The inventory allocation mapping relationship is a correlation model that records the upper limit range of resource allocation values ​​corresponding to inventory values ​​of different material units. The revenue resource subsidy mapping relationship is a correlation model that records the lower limit threshold of resource subsidies corresponding to revenue values ​​of different product categories.

[0101] The policy generation processing parameters are the configuration parameters used by the server when executing the S205 policy generation processing, including but not limited to the low-rank adaptation matrix parameters of the large language model.

[0102] For example, after the server executes parameterized instructions in the target resource scheduling background, it collects resource indicator values ​​after the strategy execution from the target object's material unit management system, user behavior system, resource flow record system, resource supplier management system, and advertising system after a preset time period (e.g., 7 days). These resource indicator values ​​include the total transaction amount of material units after strategy execution, user retention rate, resource supplier revenue, user complaint rate, resource supplier churn rate, and resource input-output ratio.

[0103] The server obtains the total transaction amount of each material unit before strategy execution as the baseline total transaction amount, and obtains the total transaction amount of each material unit after strategy execution as the post-execution total transaction amount. The server subtracts the baseline total transaction amount from the post-execution total transaction amount to obtain the increment of the total transaction amount. Dividing this increment by the baseline total transaction amount yields the platform's percentage increase in total transaction amount. The server calculates the percentage increase in user value based on changes in user retention rate and average user spending before and after strategy execution. The server calculates the percentage increase in resource provider revenue based on changes in the total revenue of the resource provider group involved in strategy execution before and after execution. The server obtains the total resource allocation subsidy expenditure and the baseline value of traffic input resource consumption generated during strategy execution, adds them together to obtain the total baseline value of strategy resource consumption, and divides this baseline value by the baseline total transaction amount of each material unit to obtain the percentage of resource allocation resource loss.

[0104] The server obtains preset first, second, third, and fourth weighting coefficients. The server multiplies the percentage increase in the total transaction amount of platform material units by the first weighting coefficient to obtain the first product; multiplies the percentage increase in user value by the second weighting coefficient to obtain the second product; multiplies the percentage increase in resource supplier revenue by the third weighting coefficient to obtain the third product; and multiplies the percentage increase in resource deployment and resource loss by the fourth weighting coefficient to obtain the fourth product. The server adds the first, second, and third products and then subtracts the fourth product to obtain the overall strategy contribution value. For example, if the server calculates that the percentage increase in the total transaction amount of platform material units is 38%, the percentage increase in user value is 15%, the percentage increase in resource supplier revenue is 20%, and the percentage increase in resource deployment and resource loss is 10%, and the first weighting coefficient is 0.4, the second weighting coefficient is 0.3, the third weighting coefficient is 0.2, and the fourth weighting coefficient is 0.1, the server calculates the overall strategy contribution value as: 0.4 × 38% + 0.3 × 15% + 0.2 × 20%. 0.1 × 10% = 15.2% + 4.5% + 4.0% 1.0% = 22.7%.

[0105] The server obtains a preset threshold, for example, a threshold of 20 points. The server compares the overall strategy contribution value of 22.7 points with the preset threshold of 20 points. Since 22.7 is greater than 20, the comparison result is that the overall strategy contribution value is greater than the preset threshold.

[0106] The server performs feedback optimization processing based on the comparison results. When the comparison result shows that the overall contribution value of the strategy is greater than or equal to the preset threshold, the server executes the first iteration processing. The specific process of the first iteration processing is as follows: The server combines the resource scenario, strategy scheme, and effect data corresponding to the target strategy into a new knowledge unit. The resource scenario field includes the problem type, product category, time node, inventory status, price competition index, and user classification in the structured decision object. The strategy scheme field includes the strategy type, core parameters, applicable scope, execution channel, and configuration rules of the target strategy. The effect evaluation field includes the increase rate of total transaction amount of material unit after strategy implementation, conversion rate increase rate, input-output ratio, user retention change, resource supplier revenue change, resource consumption baseline value loss, and overall contribution value. The server stores this knowledge unit in the knowledge base, resulting in an updated knowledge base.

[0107] The server optimizes the ranking weights of each knowledge unit in the updated knowledge base. The server extracts the question type, product category, inventory status, and user classification from the resource scenario field of the newly stored knowledge unit as index keys. The server calculates the weighted average of the comprehensive contribution values ​​of each knowledge unit within the same resource scenario that matches the index key value. The server then ranks each knowledge unit within the same resource scenario based on the weighted average, obtaining the updated ranking weights and retrieval priorities.

[0108] The server checks whether the cumulative number of knowledge units in the knowledge base exceeds a preset sample size threshold. The preset sample size threshold is a fixed value pre-stored on the server to trigger the second iteration, for example, a preset sample size threshold of 1000. When the cumulative number of knowledge units in the knowledge base exceeds the preset sample size threshold, the server executes the second iteration. The specific process of the second iteration is as follows: the server extracts knowledge units in the knowledge base whose overall contribution value is higher than the preset contribution threshold as the positive sample set, and extracts knowledge units in the knowledge base whose overall contribution value is lower than the preset contribution threshold as the negative sample set. The preset contribution threshold is a fixed value pre-stored on the server to distinguish between high-quality and low-quality samples, for example, a preset contribution threshold of 25 points.

[0109] Based on the correspondence between the inventory level values ​​in the resource scenario field and the resource allocation upper limit range values ​​in the strategy scheme field of each knowledge unit in both the positive and negative sample sets, the server refits the mapping relationship between the inventory level values ​​and the resource allocation upper limit range values ​​to obtain an updated inventory allocation mapping relationship. Similarly, based on the correspondence between the revenue floor value in the resource scenario field and the resource subsidy lower limit threshold value in the strategy scheme field of each knowledge unit in both the positive and negative sample sets, the server refits the mapping relationship between the revenue floor value and the resource subsidy lower limit threshold value to obtain an updated revenue-resource-subsidy mapping relationship.

[0110] The server uses a positive sample set to fine-tune the policy generation processing parameters, obtaining the fine-tuned parameters. Specifically, the server uses the resource scenario field and policy scheme field of each knowledge unit in the positive sample set as training data to perform low-rank fine-tuning operations on the low-rank adaptation matrix parameters of the large language model used in the policy generation process, thereby enhancing the output probability of the large language model to generate high-contribution policies within the fully parameterized policy numerical space.

[0111] The server performs weight calibration based on the user complaint rate, resource supplier churn rate, and resource input-output ratio values ​​from the resource metric values. The server obtains preset thresholds for user complaint rate, resource supplier churn rate, and resource input-output ratio. The server compares the user complaint rate value with the preset threshold. When the user complaint rate value is higher than the preset threshold, the server increases the dynamic weight coefficient of the user retention improvement rate in the multi-objective comprehensive scoring rule, for example, increasing the dynamic weight coefficient β of the user retention improvement rate from 0.25 to 0.35. The server compares the resource supplier churn rate value with the preset resource supplier churn rate threshold. When the resource supplier churn rate value is higher than the preset threshold, the server increases the dynamic weight coefficient of the resource supplier revenue improvement rate in the multi-objective comprehensive scoring rule, for example, increasing the dynamic weight coefficient γ of the resource supplier revenue improvement rate from 0.25 to 0.35. The server compares the resource input-output ratio (ROI) value with a preset ROI threshold. When the ROI value is lower than the preset threshold, the server increases the dynamic weight coefficient of the ROI in the multi-objective comprehensive scoring rule, for example, increasing the dynamic weight coefficient δ from 0.25 to 0.35. It should be noted that the comprehensive score calculation formula for the multi-objective comprehensive scoring rule is: Comprehensive Score = α × Material Unit Transaction Amount Improvement Rate + β × User Retention Improvement Rate + γ × Resource Supplier Revenue Improvement Rate + δ × Resource Input-Output Ratio. Where α is the fixed weight coefficient corresponding to the material unit transaction amount improvement rate, β is the dynamic weight coefficient corresponding to the user retention improvement rate, γ is the dynamic weight coefficient corresponding to the resource supplier revenue improvement rate, and δ is the dynamic weight coefficient corresponding to the resource input-output ratio.

[0112] The server applies the updated inventory distribution mapping, the updated revenue resource subsidy mapping, and the fine-tuned strategy generation parameters to subsequent strategy generation processes. The server also applies the updated dynamic weighting coefficients to the subsequent multi-objective comprehensive scoring and ranking.

[0113] In this embodiment, resource metrics are collected and quantitatively evaluated after the strategy is implemented. The knowledge base and various parameters are then updated based on the evaluation results. This adjustment method continuously standardizes the parameter boundaries and scoring criteria of the generated strategy, and continuously reduces the probability of conflicts between the generated strategy and the resource rules of the target object.

[0114] The above text combined Figures 1 to 2 The resource scheduling method provided in the embodiments of this application has been described in detail. The apparatus and equipment provided in the embodiments of this application will be described below with reference to the accompanying drawings.

[0115] This application also provides a resource scheduling device, such as... Figure 3 As shown, this figure is a structural diagram of a resource scheduling device provided in an embodiment of this application. The device includes: Module 301 is used to obtain the initial resource data of the target object; The processing and calculation module 302 is used to perform feature standardization processing on the initial resource data to obtain the target resource feature vector; to identify problems based on the target resource feature vector to obtain the structured decision object; and to perform interval threshold calculation on the structured decision object based on the numerical constraints of material unit inventory, category revenue, user classification and access control, and resource allocation budget to obtain the complete parameterized allowable strategy numerical space. The decision generation module 303 is used to generate strategies within the parameter boundaries of the fully parameterized allowable strategy numerical space to obtain candidate resource optimization strategies; perform hard rule verification on the candidate resource optimization strategies to obtain candidate resource optimization strategies that pass the hard rule verification; sort the candidate resource optimization strategies that pass the hard rule verification based on the multi-objective comprehensive scoring rules to determine the target strategy; convert the target strategy into parameterized instructions and send the parameterized instructions to the target resource scheduling backend for execution.

[0116] In some possible implementations, the processing calculation module 302 is specifically used for: Temporal anomaly detection and hierarchical drill-down are performed on the feature vector of the target resource to obtain resource anomaly data. The resource anomaly data includes the location of the anomaly subject and the scope of its impact. The resource anomaly data is matched with a preset standard problem database to determine the problem type label. Based on the problem type label, the resource anomaly data, and the scene features corresponding to the target object, the structured decision object is determined.

[0117] In some possible implementations, the processing calculation module 302 is specifically used for: The process involves: acquiring the associated entities, product categories, inventory levels, and user classification attributes from the structured decision-making object; calculating the upper limit range of resource allocation based on the current inventory value and turnover target value of the corresponding material unit for the associated entity; calculating the lower limit threshold for resource subsidies and the lower limit threshold for electronic voucher denomination based on the minimum revenue value and break-even point value corresponding to the product category; determining the scope of interactive user groups based on user classification tags corresponding to user classification attributes; determining the upper limit of interaction times per user within a preset time period based on historical interaction frequency records corresponding to user classification attributes; determining the upper limit of total resource allocation based on the total resource allocation budget value corresponding to the structured decision-making object; determining the upper limit of resource allocation per user based on resource input-output target values; and determining the complete parameterized allowed strategy value space based on the upper limit range of resource allocation, the lower limit threshold for resource subsidies, the lower limit threshold threshold for electronic voucher denomination, the scope of interactive user groups, the upper limit of interaction times, the upper limit of total resource allocation, and the upper limit of resource allocation per user.

[0118] In some possible implementations, the resource scheduling device also includes: The feedback optimization module is used to obtain the resource indicator values ​​after the parameterized instructions are executed; calculate the contribution value based on the resource indicator values ​​to obtain the comprehensive contribution value of the strategy; compare the comprehensive contribution value of the strategy with the preset target threshold to obtain the comparison result; and perform feedback optimization processing based on the comparison result to obtain the feedback optimization processing result; wherein, the feedback optimization processing includes at least one of the first iteration processing, the second iteration processing, and the weight calibration processing.

[0119] In some possible implementations, the feedback optimization module is specifically used for: When the comparison result shows that the overall contribution value of the strategy is greater than or equal to the preset threshold, the target strategy is stored as a new knowledge unit in the knowledge base, resulting in an updated knowledge base. The ranking of each knowledge unit in the updated knowledge base is optimized to obtain updated ranking weights and retrieval priorities. When the cumulative number of knowledge units in the knowledge base exceeds the preset sample size threshold, the mapping relationship between the inventory value of the material unit and the upper limit range of resource allocation is refitted to obtain an updated inventory allocation mapping relationship. The mapping relationship between the category revenue value and the lower limit threshold of resource subsidies is refitted to obtain an updated revenue-resource subsidy mapping relationship. The strategy generation and processing parameters are fine-tuned to obtain fine-tuned strategy generation and processing parameters. The weight coefficients of each indicator in the multi-objective comprehensive scoring rule are calibrated to obtain updated weight coefficients.

[0120] In some possible implementations, the decision generation module 303 is specifically used for: Based on the problem type, product category, inventory level, and user classification attributes in the structured decision object, matching historical knowledge units are retrieved from the knowledge base to obtain a recall knowledge unit set. Based on the comprehensive contribution value of each knowledge unit in the recall knowledge unit set, a reference knowledge unit set is determined. Based on the structured decision object, the complete parameterized allowable strategy numerical space, and the reference knowledge unit set, strategy schemes are deduced within the parameter boundaries of the complete parameterized allowable strategy numerical space to obtain candidate resource optimization strategies.

[0121] The resource scheduling apparatus according to the embodiments of this application can correspondingly execute the method described in the embodiments of this application, and the other operations and / or functions of each module / unit of the resource scheduling apparatus are respectively for implementing Figure 2 For the sake of brevity, the corresponding processes of each method in the illustrated embodiments will not be described in detail here.

[0122] This application also provides a computing device. For example... Figure 4 As shown in the figure, this is a schematic diagram of a computing device provided in an embodiment of this application. The computing device 400 includes a bus 401, a processor 402, a communication interface 403, and a memory 404. The processor 402, the memory 404, and the communication interface 403 communicate with each other via the bus 401.

[0123] Bus 401 can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. Buses can be categorized as address buses, data buses, control buses, etc. For ease of representation, Figure 4 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0124] Processor 402 can be any one or more of the following processors: central processing unit (CPU), graphics processing unit (GPU), microprocessor (MP), or digital signal processor (DSP).

[0125] Communication interface 403 is used for communication with external devices.

[0126] Memory 404 may include volatile memory, such as random access memory (RAM). Memory 404 may also include non-volatile memory, such as read-only memory (ROM), flash memory, hard disk drive (HDD), or solid state drive (SSD).

[0127] The memory 404 stores executable code, and the processor 402 executes the executable code to perform the aforementioned resource scheduling method.

[0128] Specifically, in achieving Figure 3 In the case of the illustrated embodiment, and Figure 3 When the modules or units of the resource scheduling device described in the embodiment are implemented by software, the execution... Figure 3 The software or program code required for the functions of each module / unit can be partially or wholly stored in memory 404. Processor 402 executes the program code corresponding to each unit stored in memory 404 and performs the aforementioned resource scheduling method.

[0129] This application also provides a computer-readable storage medium. The computer-readable storage medium can be any available medium that a computing device can store, or a data storage device such as a data center containing one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state drive). The computer-readable storage medium includes instructions that instruct the computing device to execute the resource scheduling method described above.

[0130] This application also provides a computer program product comprising one or more computer instructions. When the computer instructions are loaded and executed on a computing device, all or part of the processes or functions described in this application are generated.

[0131] The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, or data center to another website, computer, or data center via wired (e.g., coaxial cable, fiber optic) or wireless (e.g., infrared, wireless, microwave, etc.) means.

[0132] When the computer program product is executed by a computer, the computer performs any of the aforementioned resource scheduling methods. The computer program product can be a software installation package; when any of the aforementioned resource scheduling methods is required, the computer program product can be downloaded and executed on the computer.

[0133] The descriptions of the processes or structures corresponding to the above figures each have their own emphasis. For parts of a process or structure that are not described in detail, please refer to the relevant descriptions of other processes or structures.

[0134] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions within the technical scope disclosed in this application should be covered within the scope of protection of this application.

Claims

1. A resource scheduling method, characterized in that, The method includes: Obtain the initial resource data of the target object; The initial resource data is subjected to feature standardization processing to obtain the target resource feature vector; Problem identification is performed based on the feature vector of the target resource to obtain structured decision objects; Based on the numerical constraints of material unit inventory, category revenue, user classification and access control, and resource allocation budget, interval threshold calculations are performed on structured decision objects to obtain a complete parameterized allowable strategy numerical space. The policy generation process is performed within the parameter boundaries of the fully parameterized policy numerical space to obtain candidate resource optimization policies; Candidate resource optimization strategies are subjected to hard rule verification to obtain the candidate resource optimization strategies that pass the hard rule verification; based on the multi-objective comprehensive scoring rule, the candidate resource optimization strategies that pass the hard rule verification are ranked to determine the target strategy; the target strategy is converted into parameterized instructions and the parameterized instructions are sent to the target resource scheduling backend for execution.

2. The method according to claim 1, characterized in that, The process of identifying problems based on target resource feature vectors to obtain structured decision objects includes: Temporal anomaly detection and hierarchical drill-down are performed on the feature vector of the target resource to obtain resource anomaly data; the resource anomaly data includes the location of the anomaly subject and the scope of its impact. The abnormal resource data is matched with a preset standard problem database to determine the problem type label; Based on problem type labels, resource anomaly data, and the scenario characteristics corresponding to the target object, the structured decision object is determined.

3. The method according to claim 1, characterized in that, The step of calculating interval thresholds for the structured decision object to obtain the complete parameterized allowable policy numerical space includes: Obtain the associated entities, product categories, inventory levels, and user classification attributes from the structured decision-making objects; Based on the current inventory value and turnover target value of the material unit corresponding to the associated entity, the upper limit range of resource allocation is calculated. Based on the minimum revenue threshold and break-even point value corresponding to the respective product category, the minimum threshold for resource subsidies and the minimum threshold for electronic voucher denomination are calculated. The scope of interactive user groups is determined based on user category tags corresponding to user category attributes; the upper limit of the number of interactions for a single user within a preset time period is determined based on historical interaction frequency records corresponding to user category attributes. Based on the total resource allocation budget corresponding to the structured decision-making object, determine the upper limit of the total resource allocation; based on the resource input-output target, determine the upper limit of resource allocation per user. Based on the upper limit range of resource allocation, the lower limit threshold of resource subsidies, the lower limit threshold of electronic voucher denomination, the range of interactive user groups, the upper limit of interaction times, the upper limit of total resource allocation, and the upper limit of resource allocation per user, a complete parameterized allowable strategy value space is determined.

4. The method according to claim 1, characterized in that, The method further includes: Obtain resource indicator values ​​after the parameterized command is executed; The contribution value is calculated based on the resource indicator values ​​to obtain the overall contribution value of the strategy. The overall contribution value of the strategy is compared with the preset threshold to obtain the comparison result; Based on the comparison results, feedback optimization processing is performed to obtain the feedback optimization processing result; wherein, the feedback optimization processing includes at least one of the first iteration processing, the second iteration processing, and the weight calibration processing.

5. The method according to claim 4, characterized in that, The step of performing feedback optimization processing based on the comparison results to obtain feedback optimization processing results includes: When the comparison result shows that the overall contribution value of the strategy is greater than or equal to the preset threshold, the target strategy is stored as a new knowledge unit in the knowledge base, resulting in an updated knowledge base. The ranking of each knowledge unit in the updated knowledge base is optimized to obtain updated ranking weights and retrieval priorities. When the cumulative number of knowledge units in the knowledge base exceeds the preset sample size threshold, the mapping relationship between the inventory value of the material unit and the upper limit range of resource allocation is refitted to obtain an updated inventory allocation mapping relationship. The mapping relationship between the category revenue value and the lower limit threshold of resource subsidies is refitted to obtain an updated revenue-resource subsidy mapping relationship. The strategy generation and processing parameters are fine-tuned to obtain fine-tuned strategy generation and processing parameters. The weight coefficients of each indicator in the multi-objective comprehensive scoring rule are calibrated to obtain updated weight coefficients.

6. The method according to claim 1, characterized in that, The process of generating a policy within the parameter boundaries of the fully parameterized policy numerical space to obtain candidate resource optimization policies includes: Based on the problem type, product category, inventory level, and user classification attributes in the structured decision object, the matching historical knowledge units are retrieved from the knowledge base to obtain the recall knowledge unit set; A reference knowledge unit set is determined based on the overall contribution value of each knowledge unit in the recalled knowledge unit set; Based on structured decision objects, a complete parameterized allowable strategy numerical space, and a set of reference knowledge units, strategy schemes are deduced within the parameter boundaries of the complete parameterized allowable strategy numerical space to obtain candidate resource optimization strategies.

7. A resource scheduling device, characterized in that, The device includes: The acquisition module is used to obtain the initial resource data of the target object; The processing and calculation module is used to perform feature standardization on the initial resource data to obtain the target resource feature vector; based on the target resource feature vector, problem identification is performed to obtain the structured decision object; based on the numerical constraints of material unit inventory, category revenue, user classification and access control, and resource allocation budget, the interval threshold calculation is performed on the structured decision object to obtain the complete parameterized allowable strategy numerical space. The decision generation module is used to generate strategies within the parameter boundaries of the fully parameterized allowable strategy numerical space to obtain candidate resource optimization strategies; perform hard rule verification on the candidate resource optimization strategies to obtain the candidate resource optimization strategies that pass the hard rule verification; based on the multi-objective comprehensive scoring rules, sort the candidate resource optimization strategies that pass the hard rule verification to determine the target strategy; convert the target strategy into parameterized instructions and send the parameterized instructions to the target resource scheduling backend for execution.

8. A computing device, characterized in that, Including memory and processor; The memory stores one or more computer programs, the one or more computer programs including instructions; when the instructions are executed by the processor, the computing device performs the method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store a computer program for performing the method as described in any one of claims 1 to 6.

10. A computer program product, characterized in that, The computer program product includes one or more computer instructions that, when executed by a computer, perform the method as described in any one of claims 1 to 6.

Images