Traffic regulation and control method, device and equipment and readable storage medium

Abstract

The invention provides a flow regulation and control method, device and equipment and a readable storage medium, and relates to the technical field of computers, and the method comprises the steps: obtaining historical data and constraint conditions of at least one commodity; processing the constraint condition through a Lagrange relaxation method, introducing a first vector and a second vector, and constructing a regulation and control rule; obtaining an initial value of a first vector and an initial value of a second vector through iterative calculation according to historical data in combination with a regulation and control rule; collecting real-time data, and updating the first vector and the second vector in combination with the initial value; the commodity list is traversed, the target score of the at least one commodity is calculated through the regulation and control rule according to the business corresponding to the at least one commodity, and the corresponding flow is distributed to the at least one commodity, and the regulation and control rule comprises the first vector and the second vector, so that the limited flow resource is accurately guided to the commodity with higher comprehensive value, and the user experience is improved. Therefore, the overall transaction volume and resource utilization efficiency of the platform are improved, and ecological diversity and fairness are guaranteed.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a flow control method, apparatus, device, and readable storage medium. Background Technology

[0002] In existing internet platform traffic management practices, traffic control often relies on static strategies based on fixed rules or human experience. Therefore, when facing large-scale, multi-objective, and complex business scenarios, a core problem of inaccurate control is prevalent. Specifically: First, it is difficult to achieve global optimization; static rules cannot flexibly adapt to real-time changes in user preferences and product value, resulting in traffic resources not being guided to positions that generate the greatest overall benefit, leading to low resource utilization efficiency. Second, it is unable to effectively coordinate diverse business objectives and complex constraints. Third, it heavily relies on human intervention, resulting in slow response times and poor decision-making consistency, making it difficult to maintain the effectiveness of strategies in dynamic environments.

[0003] Therefore, in view of the shortcomings of existing technical solutions, the present invention provides a flow control method. Summary of the Invention

[0004] Therefore, it is necessary to provide a flow control method, apparatus, device, and readable storage medium to address the aforementioned technical problems.

[0005] On the one hand, a traffic control method is provided, which includes: acquiring historical data and constraints of at least one product; processing the constraints using the Lagrange relaxation method, introducing a first vector and a second vector, and constructing control rules; obtaining the initial values ​​of the first vector and the second vector through iterative calculation based on the historical data and the control rules; collecting real-time data and updating the first vector and the second vector based on the initial values; traversing the product list, calculating the target score of at least one product based on the business corresponding to at least one product through the control rules, and distributing the corresponding traffic to at least one product, wherein the control rules include the first vector and the second vector.

[0006] Optionally, based on historical data and in conjunction with control rules, the initial values ​​of the first vector and the second vector are obtained through iterative calculation, including: Initialize the first and second vectors; Based on historical data, the initial values ​​of the first and second vectors are initialized, and the control rules are combined to obtain the initial values ​​of the first and second vectors through iterative calculation.

[0007] Optionally, real-time data is collected and the first and second vectors are updated based on the initial values, including: Based on real-time data, calculate the violation rate and learning rate for each business. Update the first vector based on its initial value, lower bound violation, and learning rate; Update the second vector based on its initial value, upper bound violation, and learning rate.

[0008] Optionally, based on real-time data, calculate the violation rate and learning rate for each service, including: Obtain the real-time exposure of each service at the current moment and construct a real-time exposure vector; Construct a target exposure vector based on the exposure time-sharing targets for each business segment; Calculate the difference between the target exposure vector and the real-time exposure vector, and take the positive part as the lower bound violation quantity; Calculate the difference between the real-time exposure vector and the target exposure vector, and take the positive part as the upper bound violation.

[0009] Optionally, the learning rate for each service can be calculated based on real-time data, using formulas including: ; in, V is the learning rate. lower V is the lower bound violation quantity. upper Let t be the upper bound violation, and t be the number of updates.

[0010] Optionally, iterate through the product list, calculate the target score for at least one product based on the business corresponding to at least one product using control rules, and distribute the corresponding traffic to at least one product, including: If the business corresponding to a product is zero, the product's initial score will be used as the target score. The control rules are as follows, based on the requirement that each product corresponds to at least one business transaction: ; Among them, adjusted_score i For the target score of item i, s i Let λ be the initial score for item i, and λ be the updated first vector. T m is the transpose of λ i ∈{0,1} n Let m be the business row vector of product i. i,b =1 indicates that the product belongs to business category b, c b The second variable threshold set for business b represents the minimum product quality standard allowed for exposure, max(0, c b -s i This indicates that a positive penalty is generated when the product score is lower than the quality threshold.

[0011] Optionally, the method also includes: Obtain user characteristics, product characteristics, and contextual characteristics; Input user features, product features, and contextual features into the ranking model to obtain the initial score for the product.

[0012] On the other hand, a flow control device is provided, the device comprising: a first processing module, used to acquire historical data and constraints of at least one commodity; The second processing module is used to process the constraints using the Lagrange relaxation method, introduce the first vector and the second vector, and construct the control rules. The third processing module is used to obtain the initial values ​​of the first vector and the second vector through iterative calculation based on historical data and control rules. The fourth processing module is used to collect real-time data and update the first and second vectors based on the initial values. The fifth processing module is used to traverse the product list, calculate the target score of at least one product based on the business corresponding to at least one product through the control rules, and distribute the corresponding traffic to at least one product. The control rules include a first vector and a second vector.

[0013] On another front, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it performs the following steps: acquiring historical data and constraints for at least one product; processing the constraints using the Lagrange relaxation method, introducing a first vector and a second vector, and constructing control rules; obtaining the initial values ​​of the first vector and the second vector through iterative calculation based on the historical data and the control rules; collecting real-time data and updating the first and second vectors based on the initial values; traversing the product list, calculating the target score of at least one product based on the business corresponding to at least one product through the control rules, and distributing the corresponding traffic to at least one product, wherein the control rules include the first vector and the second vector.

[0014] On another front, a computer-readable storage medium is provided, on which a computer program is stored. When executed by a processor, the computer program performs the following steps: acquiring historical data and constraints for at least one product; processing the constraints using the Lagrange relaxation method, introducing a first vector and a second vector, and constructing control rules; obtaining the initial values ​​of the first vector and the second vector through iterative calculation based on the historical data and the control rules; collecting real-time data and updating the first and second vectors based on the initial values; traversing the product list, calculating the target score for at least one product based on the business corresponding to at least one product through the control rules, and distributing the corresponding traffic to at least one product, wherein the control rules include the first vector and the second vector.

[0015] The aforementioned traffic control method, device, system, and computer equipment include the following steps: acquiring historical data and constraints for at least one product; processing the constraints using the Lagrange relaxation method, introducing a first vector and a second vector, and constructing control rules; iteratively calculating the initial values ​​of the first vector and the second vector based on the historical data and the control rules; collecting real-time data and updating the first and second vectors based on the initial values; traversing the product list, calculating the target score for at least one product based on the business corresponding to at least one product using the control rules, and distributing the corresponding traffic to at least one product, wherein the control rules include the first vector and the second vector; thus, limited traffic resources are precisely directed to products with higher overall value, thereby increasing the platform's overall transaction volume and resource utilization efficiency, ensuring ecological diversity and fairness, and enabling adaptive adjustments based on real-time changes, enhancing system stability and user satisfaction, and building an efficient, healthy, and sustainable platform traffic ecosystem. Attached Figure Description

[0016] To more clearly illustrate the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating a flow control method in one embodiment; Figure 2 This is a structural block diagram of a flow control device in one embodiment; Figure 3 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0019] It should be understood that, in the description of this application, unless the context explicitly requires it, words such as "including" or "comprising" throughout the specification should be interpreted as including rather than exclusive or exhaustive; that is, meaning "including but not limited to".

[0020] It should also be understood that the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance. Furthermore, in the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0021] It should be noted that the terms "S1," "S2," etc., are used only for descriptive purposes and do not specifically refer to the order or sequence, nor are they intended to limit this application. They are merely for the convenience of describing the method of this application and should not be construed as indicating the sequential order of the steps. Furthermore, the technical solutions of the various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed in this application.

[0022] In one embodiment, such as Figure 1 As shown, a flow control method is provided, including the following steps: S101: Obtain historical data and constraints for at least one product.

[0023] Here, historical data for products can include product and business attribute data (such as product category, merchant information, inventory depth, profit information, etc.), user behavior and context data (such as user profile, scenario information, historical conversion rate, historical click-through rate, historical gross merchandise volume (GMV) contribution, etc.), market and operational target data (such as time-sharing exposure targets, budget and cost, competitive landscape data, etc.), and system and performance data (such as the size and distribution of product sets, performance baseline of online services, etc.).

[0024] The historical data can be data from the previous day, the previous two days, or the previous N days, with the range determined by the user.

[0025] Here, constraints are mathematical expressions in the operations research optimization model that limit the range of values ​​for decision variables, reflecting various restrictions, requirements, and objectives in real-world business operations.

[0026] These constraints may include exposure targets, efficiency targets, business diversity constraints, merchant fairness constraints, user experience constraints, and operational strategy constraints.

[0027] S102: The constraints are processed by the Lagrange relaxation method, and the first and second vectors are introduced to construct the control rules.

[0028] Here, the Lagrange relaxation method is a mathematical method for solving complex optimization problems, especially suitable for integer programming or combinatorial optimization problems with constraints.

[0029] Here, the variables included in the first and second vectors are dual variables. A dual variable is a numerical value (or function) associated with a certain constraint in the original problem. It represents the marginal rate of change of the optimal value of the objective function of the original problem when the right-hand side of the constraint undergoes a small change.

[0030] The first variable can be the exposure multiplier λ∈Rn, which is used to penalize underexposure.

[0031] Here, different business functions correspond to different first variables, and the first variables corresponding to different business functions can form a first vector. The first vector is an n-dimensional vector, representing the exposure multiplier for each business function.

[0032] The second variable can be a quality threshold c∈Rn, used to penalize low-quality exposures.

[0033] Different services correspond to different second variables, and these second variables can form a second vector. The second vector is an n-dimensional vector representing the quality threshold for each service.

[0034] Here, the control rule is a rule that takes into account multiple constraints to obtain an adjusted score.

[0035] Specifically, based on the original constraint optimization problem, constraints are introduced, and the Lagrange relaxation method is used to transform the constraints into penalty terms. By constructing and decomposing the Lagrange function, the comprehensive control rules for the commodity are derived.

[0036] S103: Based on historical data and combined with control rules, the initial values ​​of the first vector and the second vector are obtained through iterative calculation.

[0037] Here, iterative computation is a computational method that gradually approximates the target solution by repeatedly performing the same or similar computational steps.

[0038] Specifically, today's request data is written to disk and used as training data for offline operations. The control parameters for each business are calculated through control rules, and the same control parameters for each business are combined into a vector as the initial value.

[0039] S104: Collect real-time data and update the first and second vectors based on the initial values.

[0040] The period for collecting real-time data can be set manually, such as collecting data once per minute or once per hour.

[0041] Specifically, by collecting real-time data, the correction of the first vector and the second vector is determined, and the first vector and the second vector are updated based on the initial values ​​of the first vector and the second vector.

[0042] S105: Traverse the product list, calculate the target score of at least one product based on the business corresponding to at least one product through the control rules, and distribute the corresponding traffic to at least one product, wherein the control rules include a first vector and a second vector.

[0043] Here, the business corresponding to a product refers to the business category or business scenario to which the product belongs from different dimensions. A product can belong to zero, one, or more business dimensions simultaneously, and each business dimension has its own independent traffic control target.

[0044] The business can include category business (such as electronic products, apparel, etc.), brand business, marketing activity business (such as major promotions, new product launches and clearance sales, etc.), user segmentation business (such as recommendations for high-end users, recommendations for cost-effective users and students, etc.), and strategic support business (such as high-profit products, cross-border products and environmentally friendly products, etc.).

[0045] The higher the target score of a product, the more traffic it will receive.

[0046] Specifically, for online requests, the product list is traversed, and the target score for each product is calculated using control rules based on the business to which the product belongs. The products are then sorted in descending order according to their target scores, and traffic is distributed to the corresponding products.

[0047] It should be noted that this application can accurately direct limited traffic resources to products with higher overall value, thereby increasing the platform's overall transaction volume and resource utilization efficiency, ensuring ecological diversity and fairness, and can adaptively adjust according to real-time changes, enhancing system stability and user satisfaction, and building an efficient, healthy, and sustainable platform traffic ecosystem.

[0048] In some specific implementations, based on historical data and in conjunction with control rules, the initial values ​​of the first vector and the second vector are obtained through iterative calculation, including: Initialize the first and second vectors; Based on historical data, the initial values ​​of the first and second vectors are initialized, and the control rules are combined to obtain the initial values ​​of the first and second vectors through iterative calculation.

[0049] Here, historical data can be historical request data.

[0050] Specifically, historical request data is stored in storage space (such as HDFS / S3), and first and second vector samples are constructed through feature extraction. The optimization objective is defined as maximizing the correlation between the adjusted score and the actual performance. Numerical optimization methods (such as BFGS) are used to iteratively solve for the first and second vectors to obtain the optimal first and second vectors.

[0051] In one embodiment, regularization terms, time decay considerations, and A / B testing verification can also be introduced to improve the accuracy of the initial values.

[0052] In this way, offline operations research can utilize a whole day's worth of request data to optimize traffic allocation from a global perspective. Compared with online real-time computing, offline computing can take into account the traffic characteristics of different time periods and different user groups, thereby making more comprehensive decisions and avoiding local optima that may be caused by online real-time decisions.

[0053] In some specific implementations, real-time data is collected and the first and second vectors are updated based on initial values, including: Based on real-time data, calculate the violation rate and learning rate for each business. Update the first vector based on its initial value, lower bound violation, and learning rate; Update the second vector based on its initial value, upper bound violation, and learning rate.

[0054] Here, violation amount refers to the difference between the actual exposure amount and the time-sharing target exposure amount.

[0055] Here, the lower bound violation indicates the degree to which the actual exposure is lower than the target exposure, reflecting the situation of insufficient exposure.

[0056] Here, the upper bound violation indicates the degree to which the actual exposure exceeds the target exposure, reflecting the situation of overexposure.

[0057] The real-time data collection frequency can be one minute.

[0058] Specifically, the exposure multipliers of all services form the first vector. Using the initial value as the base value, the learning rate multiplied by the lower bound violation is used as the current increment to obtain the updated value of the first vector, max(0, λ+η). t ·V lower ), where λ is the exposure multiplier, η t V is the learning rate. lower This is the lower bound violation quantity.

[0059] Specifically, the quality thresholds of all services form a second vector. Using the initial value as the base value, the learning rate multiplied by the upper bound violation is used as the current increment to obtain the updated value of the second vector, max(0, λ+η). t ·V upper ), where λ is the exposure multiplier, η t V is the learning rate. upper This is the upper bound violation quantity.

[0060] In this way, by calculating the violation amount and learning rate based on real-time data, it can adapt to the instantaneous changes in traffic patterns, automatically correct parameter deviations, achieve the exposure target, reduce traffic efficiency loss, and improve traffic efficiency.

[0061] In some specific implementations, the violation rate and learning rate of each service are calculated based on real-time data, including: Obtain the real-time exposure of each service at the current moment and construct a real-time exposure vector; Construct a target exposure vector based on the exposure time-sharing targets for each business segment; Calculate the difference between the target exposure vector and the real-time exposure vector, and take the positive part as the lower bound violation quantity; Calculate the difference between the real-time exposure vector and the target exposure vector, and take the positive part as the upper bound violation.

[0062] The target exposure can be set by each business unit, while the actual exposure can be obtained through real-time statistics.

[0063] Specifically, based on the exposure time-sharing targets of all services, the target exposure vector L for the current moment is constructed; based on the actual exposure volume of all services, the actual exposure vector E for the current moment is constructed. Subtracting the two vectors from each other and then taking the larger value of the result with zero, a constraint vector for the violation is constructed.

[0064] Specifically, the lower bound violation can be: V lower =max(0,LE) Specifically, the upper bound violation can be: V upper =max(0,EL) In this way, by obtaining the exposure of each business in real time, we can promptly understand the gap between the current traffic allocation and the target.

[0065] In some specific implementations, the learning rate of each service is calculated based on real-time data, and the formula includes: ; in, V is the learning rate. lower V is the lower bound violation quantity. upper Let t be the upper bound violation, and t be the number of updates.

[0066] Here, the L2 norm, also known as the Euclidean norm, is a measure of the length or size of a vector in a vector space. The L2 norm is specifically calculated by taking the square root of the sum of the squares of the vector's elements.

[0067] In each cycle, the convergence of the violation constraint vector is calculated once.

[0068] Where t is the number of updates, that is, the number of steps the current parameter is updated. If it is updated once every minute, then t is the number of minutes of the current time of the day.

[0069] in, This is the learning rate decay term; as t increases, The learning rate decreases gradually with each iteration; the more iterations, the smaller the learning rate.

[0070] in, To enable adaptive adjustment, the more severe the constraint violation, the smaller the learning rate.

[0071] In one embodiment, the current target score tends to stabilize when the learning rate is detected to be equal to or close to 0.

[0072] This simplifies the burden of algorithm deployment and parameter tuning, while also being suitable for business scenarios with complex constraints and real-time data changes.

[0073] In some specific implementations, the product list is traversed, and based on the business corresponding to at least one product, a target score for at least one product is calculated using control rules, and corresponding traffic is distributed to at least one product, including: If the business corresponding to a product is zero, the product's initial score will be used as the target score. The control rules are as follows, based on the requirement that each product corresponds to at least one business transaction: ; Among them, adjusted_score i For the target score of item i, s i Let λ be the initial score for item i, and λ be the updated first vector. T m is the transpose of λ i ∈{0,1} n Let m be the business row vector of product i. i,b =1 indicates that the product belongs to business category b, c b The second variable threshold set for business b represents the minimum product quality standard allowed for exposure, max(0, c b -s i This indicates that a positive penalty is generated when the product score is lower than the quality threshold.

[0074] Where λ is the exposure multiplier vector, c b The quality threshold set for business b represents the minimum quality standard of the products that business is allowed to expose.

[0075] Here, the initial score can be either the ranking score or the score obtained by multiplying the predicted conversion rate by the product price. The ranking score refers to a comprehensive score calculated by the model for each candidate product during the ranking phase of the recommendation / search system. This score measures the product's relevance, attractiveness, and estimated conversion value to the current user in the current context.

[0076] Among them, the ReLU function is one of the most commonly used activation functions in deep learning. It introduces nonlinearity into neural networks, enabling models to learn and approximate complex nonlinear relationships.

[0077] Specifically, when a product does not belong to any business, its target score is equal to its initial score; when a product belongs to one or more businesses, the factors controlled by λ and c are calculated according to the configured control rules to obtain the product's target score.

[0078] In one embodiment, the first vector and the second vector may be assigned corresponding weights based on their importance.

[0079] In one embodiment, the control rules may further include: ; Where γ is the new business support coefficient (hyperparameter), II new (b) is an indicator function; if business b is a new business, then it is 1, ∅(λ) b =max(0,τ−λ) b ), providing additional incentives for low-λ services (τ is the threshold).

[0080] This accelerates the cold start of new businesses.

[0081] In some specific implementations, the method further includes: Obtain user characteristics, product characteristics, and contextual characteristics; Input user features, product features, and contextual features into the ranking model to obtain the initial score for the product.

[0082] Here, user characteristics are used to describe a user's static attributes and dynamic behaviors, which may include gender, age, region, membership level, recently clicked product list, distribution of recently purchased product categories, average order value, etc.

[0083] User characteristics are obtained by querying user profiling systems or user feature databases.

[0084] Here, product features are used to describe the attributes of the product itself, which may include product ID (Identifier), category, brand, price, inventory, sales volume, rating, etc.

[0085] Among these methods, product features are obtained from product databases or product feature services.

[0086] Here, contextual features are used to describe the environmental information of the current request, which may include time, device, scene, network, etc.

[0087] Among them, contextual features are extracted from user requests.

[0088] Among them, fine-ranking models can include DNN (Deep Neural Network), DeepFM (Deep Factorization Machine), MMoE (Multi-gate Mixture-of-Experts), DIN (Deep Interest Network), etc.

[0089] Specifically, user features, product features, and contextual features are obtained, and these three types of features are merged into a single feature vector through feature concatenation and feature processing (such as normalization, embedding lookup, etc.). It should be understood that, although Figure 1 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 1 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0090] In one embodiment, such as Figure 2 As shown, a traffic control device is provided, comprising: a first processing module 201 for acquiring historical data and constraints of at least one product; a second processing module 202 for processing the constraints using the Lagrange relaxation method, introducing a first vector and a second vector, and constructing control rules; a third processing module 203 for obtaining the initial values ​​of the first vector and the second vector through iterative calculation based on historical data and the control rules; a fourth processing module 204 for collecting real-time data and updating the first vector and the second vector based on the initial values; and a fifth processing module 205 for traversing the product list, calculating the target score of at least one product based on the business corresponding to at least one product through the control rules, and distributing the corresponding traffic to at least one product, wherein the control rules include the first vector and the second vector.

[0091] In a preferred embodiment of this application, the third processing module 203 is specifically used to: initialize the first vector and the second vector; Based on historical data, the initial values ​​of the first and second vectors are initialized, and the control rules are combined to obtain the initial values ​​of the first and second vectors through iterative calculation.

[0092] As a preferred implementation, in this embodiment of the application, the fourth processing module 204 is specifically used to: calculate the violation amount and learning rate of each service based on real-time data; Update the first vector based on its initial value, lower bound violation, and learning rate; Update the second vector based on its initial value, upper bound violation, and learning rate.

[0093] As a preferred implementation, in this embodiment of the application, the fourth processing module 204 is further configured to: obtain the real-time exposure of each service at the current moment and construct a real-time exposure vector; Construct a target exposure vector based on the exposure time-sharing targets for each business segment; Calculate the difference between the target exposure vector and the real-time exposure vector, and take the positive part as the lower bound violation quantity; Calculate the difference between the real-time exposure vector and the target exposure vector, and take the positive part as the upper bound violation.

[0094] In a preferred embodiment of this application, the fourth processing module 204 is further configured to: ; in, V is the learning rate. lower V is the lower bound violation quantity. upper Let t be the upper bound violation, and t be the number of updates.

[0095] As a preferred implementation, in this embodiment of the application, the fifth processing module 205 is specifically used to: in response to the business corresponding to the product being zero, take the initial score of the product as the target score; The control rules are as follows, based on the requirement that each product corresponds to at least one business transaction: ; Among them, adjusted_score i For the target score of item i, s i Let λ be the initial score for item i, and λ be the updated first vector. T m is the transpose of λ i ∈{0,1} n Let m be the business row vector of product i. i,b =1 indicates that the product belongs to business category b, cb The second variable threshold set for business b represents the minimum product quality standard allowed for exposure, max(0, c b -s i This indicates that a positive penalty is generated when the product score is lower than the quality threshold.

[0096] As a preferred implementation, in this embodiment of the application, the fifth processing module 205 is further used to: acquire user features, product features, and context features; Input user features, product features, and contextual features into the ranking model to obtain the initial score for the product.

[0097] Specific limitations regarding the flow control device can be found in the limitations of the flow control method described above, and will not be repeated here. Each module in the aforementioned flow control device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0098] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram may be as follows: Figure 3 As shown, the computer device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements a flow control method. The display screen can be an LCD screen or an e-ink display screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad mounted on the computer device casing, or an external keyboard, touchpad, or mouse.

[0099] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0100] Embodiments of this application also provide a computer-readable storage medium storing a computer program, wherein the computer program is configured to execute the steps in any of the above-described traffic control method embodiments when it runs.

[0101] In one exemplary embodiment, the aforementioned computer-readable storage medium may include, but is not limited to, various media capable of storing computer programs, such as a USB flash drive, read-only memory (ROM), random access memory (RAM), portable hard disk, magnetic disk, or optical disk.

[0102] Embodiments of this application also provide a computer program product, which includes a computer program that, when executed by a processor, implements the steps in any of the above-described traffic control method embodiments.

[0103] Embodiments of this application also provide another computer program product, including a non-volatile computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps in any of the above-described flow control method embodiments.

[0104] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0105] The above provides a detailed description of the flow control method, apparatus, table parsing system, and computer equipment provided in this application. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the embodiments above are only intended to help understand the method and core ideas of this application. It should be noted that those skilled in the art can make various improvements and modifications to this application without departing from its principles, and these improvements and modifications also fall within the protection scope of the claims of this application.

Claims

1. A method of flow regulation, characterized by, The method comprises: acquiring historical data and constraint conditions of at least one commodity; processing the constraint conditions by a Lagrange relaxation method, introducing a first vector and a second vector, and constructing a regulation rule; obtaining initial values of the first vector and the second vector by iterative calculation according to the historical data and in combination with the regulation rule; updating the first vector and the second vector in combination with the initial values by collecting real-time data; traversing a commodity list, calculating target scores of the at least one commodity by the regulation rule according to businesses corresponding to the at least one commodity, and distributing corresponding traffic to the at least one commodity, wherein the regulation rule comprises the first vector and the second vector.

2. The flow regulation method of claim 1, wherein, The obtaining of the initial values of the first vector and the second vector by iterative calculation according to the historical data and in combination with the regulation rule comprises: initializing the first vector and the second vector; obtaining the initial values of the first vector and the second vector by iterative calculation according to the historical data, the initialized first vector and the initialized second vector, and in combination with the regulation rule.

3. The flow regulation method of claim 1, wherein, The updating of the first vector and the second vector in combination with the initial values by collecting real-time data comprises: calculating violation amounts and learning rates of each business according to the real-time data; updating the first vector according to the initial value of the first vector, a lower bound violation amount and the learning rate; updating the second vector according to the initial value of the second vector, an upper bound violation amount and the learning rate.

4. The flow regulation method of claim 3, wherein, The calculation of the violation amounts and the learning rates of each business according to the real-time data comprises: obtaining real-time exposure amounts of each business at a current time to construct a real-time exposure vector; constructing a target exposure vector according to exposure time targets of the businesses; calculating a difference between the target exposure vector and the real-time exposure vector, and taking a positive part as a lower bound violation amount; calculating a difference between the real-time exposure vector and the target exposure vector, and taking a positive part as an upper bound violation amount.

5. The flow regulation method of claim 4, wherein, The formula for calculating the learning rates of each business according to the real-time data comprises: ； wherein, is the learning rate, V lower is the lower bound violation, V upper is the upper bound violation, t is the number of updates.

6. The flow regulation method of claim 1, wherein, The traversing of the commodity list, the calculation of the target scores of the at least one commodity by the regulation rule according to the businesses corresponding to the at least one commodity, and the distribution of the corresponding traffic to the at least one commodity comprise: in response to a business corresponding to a commodity being zero, taking an initial score of the commodity as a target score; in response to the business corresponding to the commodity being at least one, the regulation rule is: ； Among them, adjusted_score i For the target score of item i, s i Let λ be the initial score for item i, and λ be the updated first vector. T m is the transpose of λ i ∈{0,1} n Let m be the business row vector of product i. i,b =1 indicates that the product belongs to business b, c b The second variable threshold set for business b represents the minimum product quality standard allowed for exposure, max(0, c b -s i This indicates that a positive penalty is generated when the product score is lower than the quality threshold.

7. The flow regulation method of claim 6, wherein, The method further comprises: acquiring user features, commodity features and context features; inputting the user features, the commodity features and the context features into a fine arrangement model to obtain initial scores of commodities.

8. A flow regulating device, characterized by The apparatus comprises: a first processing module configured to acquire historical data and constraint conditions of at least one commodity; a second processing module configured to process the constraint conditions by a Lagrange relaxation method, introduce a first vector and a second vector, and construct a regulation rule; a third processing module configured to obtain initial values of the first vector and the second vector by iterative calculation according to the historical data and in combination with the regulation rule; A fourth processing module is configured to collect real-time data and update the first vector and the second vector in combination with initial values; A fifth processing module is configured to traverse a commodity list, calculate a target score of at least one commodity according to a business corresponding to the at least one commodity through the regulation and control rule, and distribute corresponding traffic to the at least one commodity, wherein the regulation and control rule comprises the first vector and the second vector.

9. A computer device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, The computer program is executed by the processor to implement the steps of the method of any one of claims 1 to 7.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by the processor to implement the steps of the method of any one of claims 1 to 7.

Images