Delivery management

Abstract

A method, an apparatus, and a computer program product for delivery management are provided. The method comprises the following steps: acquiring delivery data associated with a plurality of delivery time points in a first time window; the delivery data includes a resource cost associated with a delivery time point in the plurality of delivery time points and a contribution caused by the resource cost to the delivery purpose. A second time window is determined. A second time window is specified by the data provider to verify whether the contribution meets the delivery purpose. Then, based on the delivery data, the first time window and the second time window, a prediction model is obtained. The prediction model indicates an association relationship between the launched data and the predicted resource cost.

Description

Cross-references

[0001] This application claims the benefit of U.S. Patent Application 18 / 814,001, filed August 23, 2024, entitled "Deployment Management," the entire contents of which are incorporated herein by reference. Technical Field

[0002] This disclosure generally relates to delivery management, and more specifically to methods, apparatus and computer program products for managing data delivery. Background Technology

[0003] Data providers seek out potential consumers to promote their data with clear ROI requirements and limited budgets. In the data delivery field, Real-Time Bidding (RTB) allows data providers to bid on data impressions in real time. Here, data delivery is also known as delivery, content delivery, content placement, data placement, data publication, etc. To support data providers in achieving their goals, various identity-revealing bidding algorithms have been developed. Typically, these algorithms generate bids based on real-time feedback collected from sequences of user behavior events such as impressions, clicks, and conversions. However, the reliance on real-time and accurate user data has become increasingly controversial, leading to widespread concerns about data protection. At this point, it is desirable to ensure the effectiveness of delivery while simultaneously protecting user data. Summary of the Invention

[0004] In a first aspect of this disclosure, a method for delivery management is provided. In this method, delivery data associated with multiple delivery time points within a first time window is acquired. The delivery data includes resource costs associated with each delivery time point among the multiple delivery time points and the contribution of the resource costs to the delivery objective. A second time window is determined. The second time window is specified by a data provider to verify whether the contribution meets the delivery objective. A first length of the first time window is greater than a second length of the second time window. Based on the delivery data and the first and second time windows, a prediction model is acquired. The prediction model indicates the correlation between delivery data associated with multiple previous delivery time points within a third time window and predicted resource costs. The predicted resource costs indicate the total resource costs corresponding to a fourth time window following the third time window. The length of the third time window is less than the first length, and the length of the fourth time window is equal to the second length.

[0005] In a second aspect of this disclosure, an electronic device is provided. The electronic device includes a computer processor coupled to a computer-readable storage unit, the storage unit including instructions that, when executed by the computer processor, implement the method according to the first aspect of this disclosure.

[0006] In a third aspect of this disclosure, a computer program product is provided, comprising a computer-readable storage medium having program instructions embodied therein, which are executed by an electronic device to cause the electronic device to perform the method according to a first aspect of this disclosure.

[0007] The present invention is provided to present in a simplified form the selection of concepts further described below in the detailed description. The present invention is not intended to identify key or essential features of the claimed subject matter, nor is it intended to limit the scope of the claimed subject matter. Attached Figure Description

[0008] The above and other objects, features and advantages of this disclosure will become more apparent from the accompanying drawings, which describe some implementations of this disclosure in more detail, wherein the same reference numerals generally refer to the same parts in the implementations of this disclosure.

[0009] Figure 1 A diagram illustrating the delay between events and event reporting is provided.

[0010] Figure 2 An example diagram illustrating the management of data delivery according to an implementation of this disclosure is shown;

[0011] Figure 3 An example diagram of a deployment management framework according to an implementation of this disclosure is shown;

[0012] Figure 4 An example diagram demonstrating the THRESHOLD algorithm according to an implementation of this disclosure is shown;

[0013] Figure 5 An example diagram of an interpolation algorithm according to an implementation of this disclosure is shown;

[0014] Figure 6 An example diagram of an algorithm for multi-channel promotion according to an implementation of this disclosure is shown;

[0015] Figure 7 An example diagram of an industrial dataset according to an implementation of this disclosure is shown;

[0016] Figure 8 An example flowchart of a method for delivery management according to an implementation of this disclosure is shown; and

[0017] Figure 9 A block diagram of a computing device in which various implementations of the present disclosure may be implemented is shown. Detailed Implementation

[0018] The principles of this disclosure will now be described with reference to some implementations. It should be understood that these implementations are described for illustrative purposes only and to assist those skilled in the art in understanding and implementing this disclosure, and do not imply any limitation on the scope of this disclosure. The disclosure described herein can be implemented in various ways other than those described below.

[0019] In the following description and claims, unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains.

[0020] References to "an implementation," "implementation," "example implementation," etc., in this disclosure indicate that the described implementation may include specific features, structures, or characteristics, but not every implementation necessarily includes such features, structures, or characteristics. Furthermore, such phrases do not necessarily refer to the same implementation. Moreover, when a specific feature, structure, or characteristic is described in conjunction with an example implementation, it can be assumed that, whether explicitly described or not, the influence of such feature, structure, or characteristic on other implementations is within the knowledge of those skilled in the art.

[0021] It should be understood that although the terms “first” and “second” may be used in this document to describe various elements, these elements should not be limited by these terms. These terms are used only to distinguish one element from another. For example, a first element may be referred to as a second element, and similarly, a second element may be referred to as a first element, without departing from the scope of the example implementation. As used herein, the term “and / or” includes any and all combinations of one or more of the listed terms.

[0022] The terminology used herein is for the purpose of describing a particular implementation only and is not intended to limit the example implementations. As used herein, the singular forms “a,” “an,” and “the” are also intended to include the plural forms unless the context clearly indicates otherwise. It will be further understood that the terms “comprising,” “including,” “having,” “having,” “containing,” and / or “containing” as used herein specify the presence of the stated features, elements, and / or components, but do not preclude the presence or addition of one or more other features, elements, components, and / or combinations thereof.

[0023] The principles of this disclosure will now be described with reference to some implementations. It should be understood that these implementations are described for illustrative purposes only and to assist those skilled in the art in understanding and implementing this disclosure, and do not imply any limitation on the scope of this disclosure. The disclosure described herein can be implemented in various ways other than those described below. In the following description and claims, unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains.

[0024] It is understood that the data involved in this technical solution (including but not limited to the data itself, the acquisition or use of the data) should comply with the requirements of relevant laws, regulations and rules.

[0025] It is understood that before using the technical solutions disclosed in the various implementations of this disclosure, users should be notified in an appropriate manner, in accordance with relevant laws and regulations, of the types, scope of use, and usage scenarios of the personal information involved in this disclosure, and user authorization should be obtained.

[0026] For example, in response to receiving a user's active request, a prompt message is sent to the user to explicitly notify the user that the requested operation will require the acquisition and use of the user's personal information. Therefore, the user can independently choose, based on the prompt message, whether to provide personal information to the software or hardware such as electronic devices, applications, servers, or storage media that perform the operations of the technical solutions disclosed herein.

[0027] As an optional but not limited implementation, the method of sending a prompt to the user in response to a user's active request may include, for example, a pop-up window, in which the prompt message may be presented in text form. Furthermore, the pop-up window may also include a selection control allowing the user to choose "agree" or "disagree" to provide personal information to the electronic device.

[0028] It is understood that the above-described notification and user authorization process is merely exemplary and does not limit the implementation of this disclosure. Other methods that comply with applicable laws and regulations are also applicable to the implementation of this disclosure.

[0029] As briefly mentioned above, identity-revealing bidding algorithms rely on user data, triggering widespread concerns about data security, such as privacy. In response to the rising controversy, regulators and companies have developed a range of strategies to protect user data by making it untraceable. Therefore, under data security regulations, revealing data is far less efficient, and non-data-driven performance optimization is difficult.

[0030] A more sensitive equilibrium is expected in the data delivery field, and some companies have already proposed corresponding strategies. For example, strategies have been developed to help data providers measure the success of data delivery while maintaining user data security. However, these strategies still impact data delivery in two main ways. First, event aggregation makes it impossible to attribute conversion events to a single user click; they can only be attributed to a group of users based on certain aggregation rules. Therefore, bidding algorithms may no longer utilize fine-grained and real-time ROI data. Second, reporting delays involve intentionally delaying conversion reports by 24 to 48 hours. This delay can cause bidding strategies to respond slowly. Furthermore, random factors are often introduced during this delay to mitigate the harm to data security from timed attacks.

[0031] Existing research typically focuses on improving traditional methods of identity disclosure, which only address part of the problem and are greatly affected by data security constraints.

[0032] For the challenge of reporting latency, most solutions have been proposed in the field of conversion rate of change (CVR) prediction models. Some related works propose nonparametric latency feedback models to estimate time latency. Some related works directly quantify conversions into multiple windows as multi-head models. However, considering the large and stochastic latency of conversions caused by data security strategies, it is difficult to ensure the stability of the estimation.

[0033] Because it heavily relies on real-time conversion feedback, the widely used Proportional-Integral-Derivative (PID) algorithm has very limited application in data security scenarios. Another classic approach in the online advertising industry is Model Predictive Control (MPC), which uses fine-grained auction replay data to model the relationship between bidding, spending, and conversions to predict the optimal bid. However, it suffers from conversion delays and inaccuracies caused by strict data security regulations.

[0034] Reinforcement learning (RL)-based approaches attempt to optimize delivery performance by using bidding or spending controls through learned policies or agents. However, these methods are largely ineffective in data security scenarios due to the lack of real-time and fine-grained interactive feedback with the environment.

[0035] The following is in conjunction with the appendix Figure 1 Describe the issue of report delay. Figure 1 This diagram illustrates the delay between an event and its report. Figure 1As shown, data 112 is delivered at time T0 on day 110. In response to the data delivery, event 114 (e.g., a click event, a download event, etc.) occurs at time T1 on day 110, and event 116 (e.g., a download event) occurs at time TL on day 110. However, report 124 corresponding to event 114 occurs at a time point between T0 and T1 on day 120, and report 126 corresponding to event 116 occurs at a time point between T1 and TL on day 120. The time interval between day 110 and day 120 can be several days. Therefore, the delay between events and event reports can be several days. Consequently, feedback from users is not accurate, and data may not be delivered to the appropriate users who are interested in the data.

[0036] In view of the above, this disclosure combines Figure 2 A delivery management scheme is proposed, and an example diagram of delivery management according to an embodiment of this application is shown. Figure 2 As shown, delivery data 230 is acquired that is associated with multiple delivery time points (time point 232 and time point 234) within a first time window 210. Delivery data 230 includes resource cost 236 associated with delivery time point 232 and the contribution 238 of resource cost 236 to the delivery objective. A second time window 220 is determined. The second time window 220 is specified by the data provider of the data to be delivered to verify whether the contribution 238 meets the delivery objective, and a first length (e.g., 7 days) of the first time window 210 is greater than a second length (e.g., 1 day) of the second time window 220.

[0037] Then, based on the delivery data 230 and the first and second time windows, a prediction model 240 is obtained. Prediction model 240 indicates the correlation between delivery data associated with multiple previous delivery time points in the third time window and the predicted resource cost. The predicted resource cost indicates the total resource cost corresponding to the fourth time window following the third time window. The length of the third time window is less than the length of the first time window, and the length of the fourth time window is equal to the length of the second time window. Using these implementations of the present disclosure, a prediction model can be obtained based on long-term aggregated data, thereby protecting user data security and ensuring the effectiveness of data delivery.

[0038] Taking into account the ROI requirements of data providers and the platform ecosystem, this disclosure aims to address the problem of maximizing Total Merchandise Value (GMV) using ROI and expenditure constraints (i.e., the purpose of the campaign). Some relevant notations are defined in Table 1 below, and a mathematical representation of the problem is derived using a method similar to the online random knapsack problem. Table 1: Symbol Description

[0039] For a given data delivery 'a' from a data provider, assume there are N bidding opportunities within a preset spending period (e.g., one day). These opportunities are represented as 'auction' according to the order in which they are generated. i Based on the definitions stated above, the GMV and cumulative expenditure for N auctions within the expenditure period are respectively identified as P. G and P S This corresponds to Where v a This represents the value derived from the transformation event from the data provider. Therefore, the expected ROI result (hereinafter abbreviated as R) can be obtained as follows. res ):

[0040] Regarding data delivery a, the objective of this disclosure is to [do something] in S. cap and R target Maximize P under constraints G It is formalized as: set

[0041] The optimal bidding formula from a single auction i Angle is defined as:

[0042] In equation (5), r i It can be regarded as an auction i The ROI is [not specified]. It's worth noting that the proposed problem involves a variation of the knapsack problem (KP), specifically an online randomized variation. Each bidding opportunity can be considered as having a value [not specified]. and weight Items. Assume there are items with a capacity of S. cap The goal of this knapsack problem is to load the knapsack as full as possible to maximize its cumulative value, while ensuring the expected ROI and not exceeding its capacity. However, since bidding, winning, clicks, and conversions occur sequentially, accurate value and weight information, i.e., c i and wp i This data may not be available in advance. Therefore, in bidding strategies, historical conversion rates or model estimates can be used instead of c. i and wp i And a decision must be made immediately whether to package the item.

[0043] As mentioned above, due to subsequent reasons, the problem represented by equation (2) becomes challenging under the constraints imposed by data security regulations. First, in the auction...i c i The values ​​are not one-to-one, which complicates the optimization process of fine-grained estimation methods such as model-based methods. Secondly, c i The reporting delay exceeded 24 hours, while the frequently used spending cycle was one day. This difference means that the regulation of real-time bidding strategies was being carried out without truth labels. Then, c i The value of c is not fixed, but rather the feedback becomes noisier after aggregation, making it almost impossible to estimate. i The distribution of .

[0044] These data security-related challenges in c i The estimated deviation d(t) causes a considerable and irregular change. Given the effect of d(t), equation (5) transforms into:

[0045] Based on equation (6), the applicability of the three existing technical methods in data security scenarios is analyzed.

[0046] Real-time feedback control c i It is the observed conversion event, while the online bidding process typically requires calculating the bid price b. i And it returns within tens of milliseconds. This concept contradicts the lag characteristic of data security scenarios.

[0047] Model predictive control (MPC) uses the click-through rate (CTR) estimated by the model. i and conversion rate i That is, c i =ctr i *cur i Replace the true value. This reliance on model generalization can partially address the lag problem. However, since historical patterns cannot be adapted to the future under stochastic constraints, and transformations cannot be accurately applied back to each fine-grained individual sample under coarse-grained constraints, the accuracy and stability of model estimation cannot be guaranteed. Therefore, MPC-based methods have low applicability in data security scenarios.

[0048] Reinforcement learning (RL)-based methods require real-time feedback to adapt to typically model-based bidding strategies and actions. Therefore, they encounter similar obstacles to MPC methods in data security scenarios.

[0049] As analyzed above, in data security scenarios, the issues under review can be identified as online random key points (KPs). In online auctions, c i and wp i The incompleteness indicates that conventional optimization methods (e.g., dynamic programming) are not applicable. It has been asserted that when the weight of the items is significantly less than the knapsack capacity, the greedy algorithm serves as an approximate optimal algorithm, i.e., wp.i ≤(1-λ)S cap , 0≤λ≤1, where λ represents the similarity to the optimal solution. As an example, in data delivery platforms, λ is typically quite large, thus validating that greedy algorithms are suitable for approximate problem solving.

[0050] A typical example of a greedy algorithm is THRESHOLD (threshold), which is used only when efficiency is low. i / weight i (in this case, c) i ·v a / ·wp i =r i ) equal to or exceeding a predefined threshold R thr When R is full, pack item i until the backpack is full (expenditure reaches the limit) or there are no remaining items (auction opportunity). thr Once determined, it can be done using R. thr Replace r in equation (5) i To infer the optimal bid. Identifying specific thresholds through feedback mechanisms or model learning is inherently limited in data security scenarios due to aggregation, lag, and stochastic properties. In contrast, the SPB algorithm (SPB Two-Stage Decomposition) has proven resilient to these obstacles, thus facilitating valuable results in data security situations. The SPB framework in Algorithm 1 will be introduced below, followed by a detailed description. Figure 3 Example Figure 300 illustrates an algorithm 1 for data delivery management according to an implementation of this disclosure. (As shown...) Figure 3 As shown, the input to Algorithm 1 is the target ROI R set by the data provider. target and expenditure limit S cap The output of Algorithm 1 is for the next time control interval t. l+1 Optimal b l+1 Specifically, the first step is to input the target ROI R. target and expenditure limit S cap To calculate the total optimal expenditure S (opt) The second step is to obtain the result for the next time interval t. l+1 Optimal expenditure The third step is to use point pairs Construct an interpolation queue and obtain b using interpolation or extrapolation methods. l+1 .

[0051] The SPB algorithm offers several distinct advantages. First, it provides an innovative two-stage decomposition framework for online bidding. This approach effectively mitigates the impact of model estimation errors. Second, the algorithm is suitable for data security scenarios. The application of long-term accumulated data allows for the effective management of the three challenges involved in data security scenarios: coarse-grainedness, lag, and stochastic properties. This level of resolution cannot be achieved using other non-decomposition methods. Third, it reduces complexity. (The last sentence appears to be incomplete and possibly refers to a different algorithm or approach.) i ,...,x N The initial solution space starts from 2. N Reduced to a single dimension, it only needs to determine S (opt) .

[0052] The SPB algorithm consists of two stages, macro and micro. The reason for dividing the algorithm in this way will be explained, and further details about each stage will be provided. First, we establish a theorem based on an ideal scenario, without considering d(t). Then, it is extended to more common scenarios where d(t) is taken into account. For short time intervals, the ROI of a single bid can be defined as... In this context, the THRESHOLD greedy algorithm in an ideal scenario works as follows. First, it determines based on r... i All bidding opportunities are sorted in descending order. Then, it selects bidding opportunities from top to bottom, up to r. i Satisfying the optimal threshold And by using Replace r in equation (5) i To obtain the corresponding bid.

[0053] Figure 4 Example Figure 400 illustrates a demonstration of the THRESHOLD algorithm according to an implementation of this disclosure. Figure 4 As shown, when r i Greater than or equal to hour( Figure 4 The shaded area S1 in the middle wins all bidding opportunities to achieve the goal of constraining R. target Maximization under constraints. Referring to equation (6), ignoring d(t) implies c i The estimated value equals the true value. For each bidding opportunity, it can be accurately placed... Figure 4 The corresponding position in the text. Then, the theorem, Theorem 1, is provided.

[0054] In Theorem 1, when R thr The optimal solution is achieved when all bidding opportunities remain consistent. The theorem is proved by contradiction. Consider two distinct R... thr1 and R thr2 In the case of R thr1 >R thr2When, assuming the population P G If it is the largest, then we will use R. thr1 Move down by a tiny amount and R thr2 Move up by a tiny amount in Since all bidding opportunities are sorted in descending order of ROI, there exists Given Δ It naturally follows That is, R thr2 A portion of the expenditure in R is moved to R thr1 The overall P′ can be obtained G , P′ G >P G This is related to P G This contradicts the biggest initial assumption.

[0055] However, in data security scenarios, the value of d(t) cannot be ignored because it hinders the achievement of real-time c in short time intervals. i Value. As mentioned, data security policies typically impose a specific delay (e.g., SKAN not exceeding 48 hours). By aggregating data over multiple days, it is possible to approach c. i The actual value of d(t) is not highly volatile over very short time intervals. Therefore, regardless, if for all bidding opportunities, r i Maintaining the order regardless of the value of d(t) makes the THRESHOLD algorithm a viable solution even in short timeframes. This insight provides the motivation to divide the online bidding process into two phases, macro and micro. The macro phase designs the optimal expenditure S for a given expenditure cycle based on long-term archived data. (opt) Then, according to the budget allocation curve, S is allocated for short time intervals. (opt) To obtain Subsequently, micro-stages are based on Generate real-time bidding prices. It's worth noting that budget distribution curve studies (focusing on optimizing budget allocation) ensure that once the allocated budget is fully utilized, it guarantees a consistent Ri. thr Although this study utilizes existing work to potentially optimize budget allocation efficiency.

[0056] Regarding macro-spending planning, as mentioned earlier, the macro phase designs the long-run optimal expenditure S. (opt) This ensures that the expenditure allocation aligns with the long-run optimum in equation (2). The problem is modeled by exploring the relationship between optimal GMV and optimal ROI, and this will be illustrated with examples. Initially, without considering d(t), the following theorem is proposed under ideal conditions. Theorem 2 is proposed. For different R... target and Scap Constrained data delivery, optimal ROI Compared to Monotonically increasing. To more clearly and intuitively verify the proof of Theorem 2, in Figure 4 The diagram is shown below. Assuming different R... target and S cap Under constraints, exist and Moving between areas will provide more bidding opportunities in the S2 zone. Because r i Sort in descending order, so any r in region S1 i Any r greater than S2 j Therefore, the resulting ROI also follows the same inequality relationship. Consider equation (1), and we obtain the following equation.

[0057] The optimal ROI for the new winning region S1+S2 is: Then compare:

[0058] Equation (8) indicates along with It decreases and thus decreases, and therefore proves to be complete.

[0059] Theorem 3 is provided as follows. For different R... target and S cap Constrained data delivery, optimal GMV and expenditure about Monotonically decreasing. Then, the proof of Theorem 3 is provided. Similar to the proof of Theorem 2, it is assumed that the R values ​​are different. target and S cap Under constraints, exist and Move between them and win. and The additional bidding opportunities are definitely greater than 0. Then and That is, the gmv and expenditure of the new winning region S1+S2 are greater than those of the original winning region S1, and at the same time Less than Note that in Theorem 2, It has been proven to be about Monotonically increasing, thus obtaining and about Since it is monotonically decreasing, the proof of Theorem 3 is complete.

[0060] Considering d(t), in the THRESHOLD algorithm The value of can vary. However, the final optimal GMV and ROI still satisfy Theorem 3. Therefore, we can construct the optimal... and The relationship functions between them. As set by the data provider, R... target This is considered the optimal ROI (i.e., equation (4) is equivalent), thus allowing the calculation of the optimal GMV. And thus calculate the optimal expenditure. The function parameters are calculated based on posterior data of GMV and ROI aggregated over a long period (such as n days). Considering the small influence of d(t) on the long-term posterior data, c... i Full recovery is possible within 48 hours, and only accumulated data is used, thus avoiding coarse-grained and lag challenges. Furthermore, considering the small impact of d(t) on long-term posterior data, the impact of random challenges is significantly smaller in the short term than that of small samples due to the larger sample size of long-term outcomes. This allows for an approximate optimal functional relationship. For example, as... Figure 4 As shown, R thr Instructions for R for each additional bidding opportunity res The reduction of this value precisely represents the concept of marginal ROI. Function Various forms can be adopted depending on the situation. For clarity, an example is provided, inferring a linear relationship between the optimal marginal ROI and the optimal GMV based on Theorem 2 and Theorem 3.

[0061] In equation (9), a and b are hyperparameters, and the following equation can be obtained.

[0062] Based on equation (1), the following equation can be further obtained.

[0063] To calculate parameters a and b, the cumulative GMV and ROI over a defined expenditure period (such as one day) can be synthesized to generate a single sample point. Data points from multiple days can then be aggregated and derived through multi-point fitting. Many existing methods can be referenced for model parameter determination to achieve the minimum MSE across multiple sample points. Once parameters a and b are determined, the data provider's target ROI can be input into equation (12) to obtain the optimal expenditure for the defined expenditure duration. Given the expenditure constraint equation (3), S (opt) Assessed as In found Then, by distributing it over shorter periods using a budget allocation curve, we obtain the results for short time intervals t. l of This can guide the micro-stage t l The optimal bid within.

[0064] Regarding micro-bid optimization, as mentioned earlier, the overall optimal expenditure has already been determined at the macro stage. Then, a budget allocation method is used to obtain the optimal allocation for short time intervals. The purpose of microprocesses is to... l Precise expenditure Ideally, the THRESHOLD algorithm can be directly used to calculate R via equation (9). thr Then, the optimal bid is calculated using equation (5). However, in data security scenarios where d(t) needs to be considered, short-term GMV and ROI are affected by the aforementioned challenges, inhibiting the acquisition of suitable a and b parameters for equation (9). Therefore, the THRESHOLD algorithm needs to be improved to find a specific R without relying on GMV and ROI. thr Values, thereby ensuring satisfaction expenditure.

[0065] It is worth noting that in data security scenarios, although c i It is affected by d(t), but its expenditure is unaffected. Once the data is displayed via a widely accepted OCPM mechanism (or any other pricing method that charges per display or click), expenditure can be collected immediately, indicating that expenditure can be collected in real time, unaffected by lag and random challenges.

[0066] Therefore, the aim is to build a system for R thr and expenditure P S The model. According to Theorem 2 and Theorem 3, in the ideal case, R thr and P S Monotonically decreasing. Assume d(t) maintains r for all bidding opportunities. i Theorem 2 and Theorem 3 continue to hold in the data security scenario, characterized by an optimal R that differs from the ideal case. thr As shown in equation (6), the following algorithm is proposed to construct a system for R. thr and P S The linear interpolation model is called the interpolation-based MPC method (IMPC). By utilizing only expenditure data from a short time slice tl, it is possible to calculate the interpolation for subsequent time intervals t. l+1 R thr and b l+1 The specific algorithm is described in Algorithm 2. Figure 5 Example Figure 500 shows an implementation of Algorithm 2 according to this disclosure.

[0067] In summary, such as Figure 5 The IMPC described in Algorithm 2 offers several key advantages. First, there is no cumulative error. Unlike feedback control methods, for each t... l The optimal bid depends only on the pre-determined price. Furthermore, it remains unaffected by prior control effects. Secondly, it requires no prior function distribution. Essentially a nonparametric regression model, it ensures high accuracy. Thirdly, IMPC offers robustness and portability. Relying solely on real-time settlement data, it has proven stable and efficient even in high-frequency calculations.

[0068] Regarding multi-channel promotion, in practice, data providers often distribute data on more than one channel, regardless of data security constraints such as shared expenditures, and each channel has its own R... target Constraints. Here, channels can include, but are not limited to, applications, websites, or other channels that can display data. Therefore, the problem to be solved extends to: R res ≥R target (17)

[0069] Due to the estimation bias d between different channels in hybrid data security and non-data security scenarios. l and target ROIR target Due to differences, the single-channel SPB scheme proposed above cannot be directly applied. SPB is further extended to multiple channels, with the macro part solved jointly and the micro part solved separately. In the macro part, it is necessary to simultaneously generate the expected optimal expenditure for each channel. Before proceeding, Theorem 4 and its proof are provided below.

[0070] Regarding Theorem 4, when the population P G At its maximum value, R for each channel thr They must be equal. The proof is derived by contradiction. Considering the cases of two channels ch1 and ch2, assume the total population P... G when Time is the greatest, then Move down by a tiny amount And And move up by a tiny amount in Obviously, there is and Then It can be obtained, that is, moving a portion of the expenditure in ch2 to cg1 can obtain the total P′.G , P′ G >P G This is consistent with our initial hypothesis P. G This is the biggest contradiction.

[0071] According to Theorem 4, the optimal expenditure for each channel can be determined using the binary search method detailed in Algorithm 3. Figure 6 Example Figure 600 shows an implementation of Algorithm 3 according to this disclosure. (As shown...) Figure 6 As shown, when determining R thr Later, due to different channels d l The changes in cannot be directly bid using equation (5). Therefore, a macro-stage-like process should be calculated first. A functional correlation between GMV and ROI is constructed for each channel. Then, the IMPC algorithm is applied independently to each channel. j To derive the optimal bid

[0072] The performance of the SPB method can be validated through online and offline experiments. In one example, an online experiment can be conducted on an industrial dataset collected from a data delivery platform to compare the performance of SPB with other advanced methods in real-world industrial application environments. Figure 7 An example diagram of an industrial dataset according to an implementation of this disclosure is shown. (e.g.) Figure 7 As shown, three datasets were randomly selected for the experiment: one from activities using SKAN attribution, one from activities using PCM attribution, and another from activities using non-data-safety constraints. The experimental results show that the SPB method improves GMV and revenue compared to conventional methods.

[0073] The delivery management in the SPB method has been briefly described; more details will follow. Data to be delivered can include multimedia data such as messages, videos, and advertisements provided by data providers. Data providers can submit data delivery requests with bidding, and if they win the bid, they can deliver the data specified in the request. In the implementation of this disclosure, the return value is... Figure 2 It is possible to obtain delivery data 230 associated with multiple delivery time points in a first time window 210. The first time window 210 is a long-term window, for example, it could be a 7-day (or another length) time window. In one example, multiple delivery data requests may exist at each delivery time point, and data specified by only one delivery data request that meets the requirements (e.g., winning the bid) can be delivered at each delivery time point.

[0074] The campaign data includes resource costs associated with multiple campaign time points and the contribution of those resource costs to the campaign objectives. Resource costs can include time costs, labor costs, operational costs, etc. Contributions can refer to conversions, ROI, CPA (cost per action), etc. For example, if data is delivered at time point t0 and there are click events related to that data, then the click event is considered a conversion.

[0075] In the implementation of this application, a second time window 220 is determined. The second time window 220 is specified by the data provider of the data to be delivered to verify whether the contribution meets the delivery objective. The second time window 220 is a short time window, for example, it can be a one-day (or another length) time window. The data provider may have an expected delivery objective (e.g., 10 conversions per day, etc.), and the data provider can verify whether the contribution meets the delivery objective within the second time window 220.

[0076] In this implementation, a prediction model 240 is obtained based on the delivery data and a first time window and a second time window. Delivery data occurring within the first time window 210 can be used to train the prediction model 240, and the prediction model 240 effectively predicts within the duration of the first time window 210. For example, if the first time window 210 is a 7-day time window, then given data from the previous 6 days, the prediction model 240 can be used to predict the data for the seventh day. In one example, the prediction model 240 can be represented as follows: F(cost) = convert (18)

[0077] In equation (18), cost represents the resource cost associated with the deployment time point, and convert represents the contribution.

[0078] Predictive model 240 indicates the correlation between delivery data associated with multiple previous delivery points within a third time window and the predicted resource cost, which indicates the total resource cost corresponding to a fourth time window following the third time window. For example, if predictive model 240 is effective for a 7-day time window, the third time window is the first 6 days of the 7-day time window, and the fourth time window is the 7th day of the 7-day time window. In other words, predictive model 240 can predict the resource cost of the fourth time window (e.g., the seventh day) based on delivery data from the third time window (e.g., the first 6 days). Using these implementations, predictive model 240 is trained using data from a long window, the delay between events and reports is covered by the long window, and therefore the predictive model can be trained accurately.

[0079] In the implementation of this disclosure, for each delivery time point within the first time window 210, an intermediate model can be determined based on an inverse proportional function, which describes the degree of influence of resource costs on contribution. According to the law of diminishing marginal utility, as conversion (as an example of contribution) increases, marginal utility gradually decreases. Therefore, there exists an inverse proportional function describing the degree of influence of resource costs on contribution. The inverse proportional function can be a linear function, an exponential function, a logarithmic function, etc. Using these implementations of this disclosure, the relationship between conversion and marginal utility can be accurately determined, and thus an appropriate intermediate model can be determined.

[0080] In the implementation of this disclosure, the inverse proportional function can be a linear inverse proportional function represented by a set of linear parameters. Since marginal utility (expressed as F′(cost) according to equation (18)) decreases with increasing conversion, the inverse proportional function can be expressed as follows:

[0081] In equation (18), F′(cost) represents marginal utility, F(cost) represents transformation, and a and b represent a set of linear parameters. Utilizing these implementations of the present disclosure, the linear inverse proportional function is easy to calculate and analyze due to its simplicity and intuitiveness, thereby improving the effectiveness of data delivery.

[0082] After determining the intermediate model, a predictive model can be determined based on the intermediate model and the deployed data.

[0083] In the implementation of this disclosure, a set of candidate values ​​for a set of linear parameters can be determined based on a linear inverse proportional function and deployment data. The resource costs and contributions contained in the deployment data can be inserted into equation (19) to obtain the candidate values.

[0084] In the implementation of this disclosure, for each delivery time point within a first time window, a set of candidate values ​​can be determined by updating a linear inverse proportional function using the resource cost associated with that delivery time point and the contribution of that resource cost to the delivery objective. Each delivery time point corresponds to a resource cost and a contribution. For example, if there are 1000 delivery time points, there are 1000 pairs of resource costs and contributions. By updating the linear inverse proportional function with these 1000 pairs of resource costs and contributions, a set of candidate values ​​can be determined.

[0085] Then, the prediction model 240 can be represented by a set of candidate values. The prediction model 240 represented by a set of candidate values ​​can be expressed as follows:

[0086] In the implementation of this disclosure, previous delivery data associated with multiple previous delivery points within a previous time window can be obtained. This previous delivery data may include previous resource costs associated with previous delivery points among the multiple previous delivery points, and previous contributions made by these previous resource costs to the delivery objective. Then, based on the prediction model and the previous delivery data, the predicted resource cost associated with the subsequent time window can be determined. Taking prediction model 240 for a 7-day time window as an example, the previous delivery data may include delivery data from the previous 6 days. Prediction model 240 can output the predicted resource cost associated with the 7th day (i.e., the subsequent time window) based on the previous 6 days. Using these implementations of this disclosure, stable, lightweight, and efficient prediction models can be proposed.

[0087] In this implementation, the unit cost and resource threshold specified by the data provider can be obtained. In one example, the unit cost can be obtained by dividing the total resource cost by the number of conversions, where the total resource cost is less than the resource threshold.

[0088] After obtaining the unit cost and resource threshold, and under the constraints of these factors, the predicted resource cost associated with subsequent time windows can be determined based on the prediction model and previous deployment data. The process of determining the predicted resource cost can be represented as follows:

[0089] In equation (21), adv-cpa represents the unit cost (also known as the performance of data delivery, or delivery performance), and adv_budget represents the resource threshold. By applying the two constraints in equation (21), the following expression can be obtained.

[0090] Equation (22) can be transformed into:

[0091] By using equation (23), the predicted resource costs associated with subsequent time windows can be determined.

[0092] In this implementation, multiple candidate resource costs can be determined based on a predictive model and previous deployment data. The predictive model can generate multiple candidate resource costs based on previous deployment data. Then, a candidate resource cost that satisfies the unit cost constraint is selected as the predicted resource cost from the multiple candidate resource costs. The unit cost can be represented by the candidate resource cost and the predicted contribution corresponding to the candidate resource cost. In one example, the candidate resource cost not only satisfies the unit cost constraint but also maximizes the value specified by the data provider (denoted as F(cost)*adv-cpa).

[0093] In the implementation of this disclosure, multiple candidate resource costs can be determined under resource threshold constraints, and the costs of multiple candidate resources are lower than the resource threshold.

[0094] In the implementation of this disclosure, the first length of the first time window can be greater than the delay between the time point when the contribution is received and that time point. Due to data security strategies, there may be a time delay between the time point when the contribution is received and that time point. The first length can cover the time delay to incorporate more data in the training of prediction 240. Using these implementations of this disclosure, the training data is accurate data reflecting the transformations collected within the long time window. Therefore, reporting delays caused by data protection strategies can be mitigated, and thus the predictive model can be trained in a more accurate manner.

[0095] It should be understood that equations (18)-(23) above provide only one implementation for managing resource allocation. Another implementation may include more or fewer steps; for example, in the SPB framework above, macro-expenditure planning can be implemented based on the proposed long-term and short-term window schemes. Specifically, equations (7)-(12) can be used to determine resource costs. Here, R... target and S cap Corresponding to unit cost and resource threshold constraints, Corresponding to contribution, Corresponding to resource costs, the predictive model can therefore be constructed in an accurate manner.

[0096] The preceding paragraphs have described the details of campaign management. Based on the implementation of this disclosure, a method for campaign management is provided. Further details regarding this method will be provided in [reference needed]. Figure 8 ,in Figure 8 An example flowchart of a method 800 for delivery management according to an implementation of this disclosure is shown. In block 810, delivery data associated with multiple delivery points in a first time window is obtained. The delivery data includes resource costs associated with each delivery point among the multiple delivery points and the contribution of the resource costs to the delivery objective. In block 820, a second time window is determined. The second time window is specified by the data provider to verify whether the contribution meets the delivery objective. A first length of the first time window is greater than a second length of the second time window. In block 830, a prediction model is obtained based on the delivery data and the first and second time windows. The prediction model indicates the association between delivery data associated with multiple previous delivery points in a third time window and predicted resource costs. The predicted resource costs indicate the total resource costs corresponding to a fourth time window following the third time window. The length of the third time window is less than the first length, and the length of the fourth time window is equal to the second length.

[0097] In the implementation of this disclosure, determining the prediction model includes: for each of the multiple delivery time points in the first time window, determining an intermediate model based on an inverse proportional function, where the inverse proportional function describes the degree of influence of resource cost on contribution; and determining the prediction model based on the intermediate model and delivery data.

[0098] In the implementation of this application, the inverse proportional function is a linear inverse proportional function represented by a set of linear parameters.

[0099] In the implementation of this disclosure, determining the prediction model based on the intermediate model and the delivery data includes: determining a set of candidate values ​​for a set of linear parameters based on the linear inverse proportional function and the delivery data; and representing the prediction model using the set of candidate values.

[0100] In the implementation of this disclosure, determining a set of candidate values ​​includes: for a delivery time point among multiple delivery time points in a first time window, updating a linear inverse proportional function with the resource cost associated with the delivery time point and the contribution of the resource cost to the delivery purpose to determine a set of candidate values.

[0101] In the implementation of this disclosure, method 800 further includes: obtaining previous delivery data associated with multiple previous delivery time points in a previous time window, the previous delivery data including: previous resource costs associated with previous delivery time points among the multiple previous delivery time points and previous contributions caused by previous resource costs to the delivery purpose; and determining the predicted resource cost associated with subsequent time windows based on the prediction model and the previous delivery data.

[0102] In the implementation of this disclosure, determining the predicted resource cost associated with subsequent time windows includes: obtaining the unit cost and resource threshold specified by the data provider; and, under the constraints of the unit cost and resource threshold, determining the predicted resource cost associated with subsequent time windows based on the prediction model and previously deployed data.

[0103] In the implementation of this disclosure, determining the predicted resource cost under the constraint of unit cost includes: determining multiple candidate resource costs based on the prediction model and previously deployed data; and selecting a candidate resource cost that satisfies the constraint of unit cost from the multiple candidate resource costs as the predicted resource cost, wherein the unit cost is represented by the candidate resource cost and the predicted contribution corresponding to the candidate resource cost.

[0104] In the implementation of this disclosure, determining the predicted resource cost under the constraint of a resource threshold includes: determining multiple candidate resource costs under the constraint of a resource threshold, wherein the multiple candidate resource costs are lower than the resource threshold.

[0105] In the implementation disclosed herein, the first length of the first time window is greater than the delay between the time points when the contribution is received.

[0106] According to an implementation of this disclosure, an apparatus for delivery management is provided. The apparatus includes: a delivery data acquisition module configured to acquire delivery data associated with multiple delivery time points within a first time window, the delivery data including: resource costs associated with delivery time points among the multiple delivery time points and the contribution of resource costs to the delivery objective; a determination module configured to determine a second time window, the second time window being specified by a data provider to verify whether the contribution meets the delivery objective, a first length of the first time window being greater than a second length of the second time window; and a prediction model acquisition module configured to acquire a prediction model based on the delivery data and the first and second time windows, the prediction model indicating the correlation between delivery data associated with multiple previous delivery time points within a third time window and predicted resource costs, the predicted resource cost indicating the total resource cost corresponding to a fourth time window following the third time window, the length of the third time window being less than the first length, and the length of the fourth time window being equal to the second length.

[0107] According to an implementation of this disclosure, an electronic device for implementing method 800 is provided. The electronic device includes a computer processor coupled to a computer-readable storage unit, the storage unit including instructions that, when executed by the computer processor, implement a method for delivery management. The method includes: acquiring delivery data associated with a plurality of delivery time points in a first time window, the delivery data including: resource costs associated with delivery time points among the plurality of delivery time points and contributions of the resource costs to a delivery objective; determining a second time window, specified by a data provider to verify whether the contribution meets the delivery objective, a first length of the first time window being greater than a second length of the second time window; and acquiring a prediction model based on the delivery data and the first and second time windows, the prediction model indicating the association between delivery data associated with a plurality of previous delivery time points in a third time window and predicted resource costs, the predicted resource costs indicating the total resource costs corresponding to a fourth time window following the third time window, the length of the third time window being less than the first length, and the length of the fourth time window being equal to the second length.

[0108] In the implementation of this disclosure, determining the prediction model includes: for each of the multiple deployment time points in the first time window, determining an intermediate model based on an inverse proportional function, where the inverse proportional function describes the degree of influence of resource cost on contribution; and determining the prediction model based on the intermediate model and deployment data.

[0109] In the implementation of this application, the inverse proportional function is a linear inverse proportional function represented by a set of linear parameters.

[0110] In the implementation of this disclosure, determining the prediction model based on the intermediate model and the delivery data includes: determining a set of candidate values ​​for a set of linear parameters based on the linear inverse proportional function and the delivery data; and representing the prediction model using the set of candidate values.

[0111] In the implementation of this disclosure, determining a set of candidate values ​​includes: for a delivery time point among multiple delivery time points in a first time window, updating a linear inverse proportional function with the resource cost associated with the delivery time point and the contribution of the resource cost to the delivery purpose, thereby determining a set of candidate values.

[0112] In the implementation of this disclosure, method 800 further includes: obtaining previous delivery data associated with multiple previous delivery time points in a previous time window, the previous delivery data including: previous resource costs associated with previous delivery time points among the multiple previous delivery time points and previous contributions caused by previous resource costs to the delivery purpose; and determining the predicted resource cost associated with subsequent time windows based on the prediction model and the previous delivery data.

[0113] In the implementation of this disclosure, determining the predicted resource cost associated with subsequent time windows includes: obtaining the unit cost and resource threshold specified by the data provider; and, under the constraints of the unit cost and resource threshold, determining the predicted resource cost associated with subsequent time windows based on the prediction model and previous deployment data.

[0114] In the implementation of this disclosure, determining the predicted resource cost under the constraint of unit cost includes: determining multiple candidate resource costs based on the prediction model and previously deployed data; and selecting a candidate resource cost that satisfies the constraint of unit cost from the multiple candidate resource costs as the predicted resource cost, wherein the unit cost is represented by the candidate resource cost and the predicted contribution corresponding to the candidate resource cost.

[0115] In the implementation of this disclosure, determining the predicted resource cost under the constraint of a resource threshold includes: determining multiple candidate resource costs under the constraint of a resource threshold, wherein the multiple candidate resource costs are lower than the resource threshold.

[0116] In this embodiment of the application, the first length of the first time window is greater than the delay between the time points when the contribution is received.

[0117] According to an implementation of this disclosure, a computer program product is provided, the computer program product including a computer-readable storage medium having program instructions embodied therein, the program instructions being executed by an electronic device to cause the electronic device to perform method 800.

[0118] Figure 9 A block diagram of a computing device 900 in which various implementations of the present disclosure may be implemented is shown. It should be understood that... Figure 9 The computing device 900 shown is for illustrative purposes only and does not imply any limitation on the functionality and scope of this disclosure. The computing device 900 can be used to implement the method 1000 described in the implementation of this disclosure. Figure 9 As shown, the computing device 900 can be a general-purpose computing device. The computing device 900 may include at least one or more processors or processing units 910, memory 920, storage unit 930, one or more communication units 940, one or more input devices 950, and one or more output devices 960.

[0119] Processing unit 910 can be a physical or virtual processor and can implement various processes based on program 925 stored in memory 920. In a multiprocessor system, multiple processing units execute computer-executable instructions in parallel to improve the parallel processing capability of computing device 900. Processing unit 910 may also be referred to as a central processing unit (CPU), microprocessor, controller, or microcontroller.

[0120] Computing device 900 typically includes various computer storage media. Such media can be any media accessible to computing device 900, including but not limited to volatile and non-volatile media, or removable and non-removable media. Memory 920 can be volatile memory (e.g., registers, cache, random access memory (RAM)), non-volatile memory (such as read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), or flash memory) or any combination thereof. Storage cell 930 can be any removable or non-removable media and may include machine-readable media such as memory, flash drives, disks, or other media that can be used to store information and / or data and can be accessed within computing device 900.

[0121] The computing device 900 may also include additional removable / non-removable volatile / non-volatile memory media. Although in Figure 9 Not shown, but may provide disk drives for reading from and / or writing to removable non-volatile disks, and optical disk drives for reading from and / or writing to removable non-volatile optical disks. In this case, each drive may be connected to a bus (not shown) via one or more data media interfaces.

[0122] The communication unit 940 communicates with another computing device via a communication medium. Furthermore, the functionality of the components in the computing device 900 can be implemented by a single computing cluster or multiple computing machines that can communicate via communication connections. Therefore, the computing device 900 can operate in a networked environment using logical connections to one or more other servers, networked personal computers (PCs), or other general-purpose network nodes.

[0123] Input device 950 can be one or more of various input devices, such as a mouse, keyboard, trackball, voice input device, etc. Output device 960 can be one or more of various output devices, such as a monitor, speaker, printer, etc. With the aid of communication unit 940, computing device 900 can also communicate with one or more external devices (not shown), such as storage devices and display devices, where one or more devices enable a user to interact with computing device 900 or any device (such as a network card, modem, etc.), enabling computing device 900 to communicate with one or more other computing devices (if needed). This communication can be performed via input / output (I / O) interface (not shown).

[0124] In some implementations, some or all components of computing device 900 may be deployed within a cloud computing architecture, rather than integrated into a single device. In a cloud computing architecture, components can be remotely provided and work together to achieve the functionality described in this disclosure. In some implementations, cloud computing provides computing, software, data access, and storage services without requiring end users to know the physical location or configuration of the systems or hardware providing these services. In various implementations, cloud computing provides services via a wide area network (WAN), such as the Internet, using appropriate protocols. For example, a cloud computing provider offers applications via a WAN that can be accessed through a web browser or any other computing component. The software or components of the cloud computing architecture, along with the corresponding data, may be stored on servers at remote locations. Computing resources in a cloud computing environment may be consolidated or distributed across locations in remote data centers. Cloud computing infrastructure can provide services through shared data centers, although they act as a single access point for users. Therefore, cloud computing architectures can be used to provide the components and functionality described herein from service providers at remote locations. Alternatively, they may be provided from conventional servers or directly installed or otherwise installed on client devices.

[0125] The functions described herein can be performed at least in part by one or more hardware logic components. For example, but not limited to, illustrative types of hardware logic components that can be used include field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CP LDs), etc.

[0126] Program code used to perform the methods described herein can be written in any combination of one or more programming languages. The program code can be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing device, such that when executed by the processor or controller, the program code enables the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code can be executed entirely or partially on a machine, partially as a standalone software package on a machine, partially on a remote machine, or entirely on a remote machine or server.

[0127] In the context of this disclosure, a machine-readable medium can be any tangible medium that may contain or store a program used by or in connection with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. More specific examples of machine-readable storage media will include electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable optical disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0128] Furthermore, although operations are shown in a specific order, this should not be construed as requiring that such operations be performed in the specific order shown or sequentially, or that all shown operations be performed to achieve the desired result. In some cases, multitasking and parallel processing may be advantageous. Similarly, while several specific implementation details are included in the above discussion, these should not be interpreted as limiting the scope of the topics described herein, but rather as descriptions of features that may be specific to a particular implementation. Certain features described in the context of a single implementation may also be implemented in combination within a single implementation. Conversely, various features described in a single implementation may also be implemented individually or in any suitable sub-combination in multiple implementations.

[0129] Although the subject matter has been described in language specific to structural features and / or methodological actions, it should be understood that the subject matter specified in the appended claims is not necessarily limited to the specific features or actions described above. Rather, the specific features and actions described above are disclosed as exemplary forms for implementing the claims.

[0130] Based on the foregoing, it should be understood that this document has described specific implementations of the currently disclosed technology for illustrative purposes, but various modifications can be made without departing from the scope of this disclosure. Therefore, the technology disclosed herein is not limited except for the appended claims.

[0131] The subject matter and functional operations described in this disclosure can be implemented in various systems, digital electronic circuits, or computer software, firmware, or hardware (including the structures disclosed in this specification and their structural equivalents), or combinations thereof. Implementations of the subject matter described in this specification can be implemented as one or more computer program products, i.e., one or more modules of computer program instructions encoded on a tangible and non-transitory computer-readable medium for execution by or control of the operation of a data processing apparatus. The computer-readable medium can be a machine-readable storage device, a machine-readable storage substrate, a memory device, a combination of substances influencing machine-readable propagation signals, or combinations thereof. The terms "data processing unit" or "data processing apparatus" encompass all means, devices, and machines for processing data, including, for example, a programmable processor, a computer, or multiple processors or computers. In addition to hardware, the apparatus may include code that creates an execution environment for the computer program in question, such as code constituting processor firmware, a protocol stack, a database management system, an operating system, or combinations thereof.

[0132] A computer program (also called a program, software, software application, script, or code) can be written in any programming language, including compiled or interpreted languages, and can be deployed in any form, including as a standalone program or as a module, component, subroutine, or other unit suitable for a computing environment. A computer program does not necessarily correspond to a file in a file system. A program can be stored as part of a file that holds other programs or data (e.g., one or more scripts stored in a markup language document), as a single file dedicated to the program in question, or as multiple coordinating files (e.g., a file storing one or more modules, subroutines, or code sections). A computer program can be deployed to execute on one or more computers located at a single site or distributed across multiple sites and interconnected via a communication network.

[0133] Processors suitable for executing computer programs include, for example, general-purpose and special-purpose microprocessors, and any one or more processors of any type of digital computer. Typically, a processor receives instructions and data from read-only memory or random access memory, or both. The basic components of a computer are a processor for executing instructions and one or more memory devices for storing instructions and data. Typically, a computer will also include, or be operatively coupled to, receiving data from or transferring data to, or both to, one or more mass storage devices (e.g., magnetic disks, magneto-optical disks, or optical disks) for storing data. However, a computer does not need to have such devices. Computer-readable media suitable for storing computer program instructions and data include all forms of non-volatile memory, media, and memory devices, including, for example, semiconductor memory devices such as EPROM, EEPROM, and flash memory devices. The processor and memory may be supplemented by or incorporated into special-purpose logic circuitry.

[0134] The specification together with the accompanying drawings is intended to be considered exemplary only, where exemplary means example. As used herein, the use of "or" is intended to include "and / or" unless the context clearly indicates otherwise.

[0135] While this disclosure contains numerous details, these details should not be construed as limiting the scope of any disclosure or claimable content, but rather as descriptions of features specific to particular implementations of a particular disclosure. Certain features described in this disclosure in the context of a single implementation may also be implemented in combination within a single implementation. Conversely, various features described in the context of a single implementation may also be implemented individually or in any suitable sub-combination in multiple implementations. Furthermore, although features may be described above as functioning in certain combinations and even initially claimed in this way, in some cases, one or more features from the claimed combination may be removed from the combination, and the claimed combination may involve sub-combinations or variations thereof.

[0136] Similarly, although operations are shown in a specific order in the accompanying drawings, this should not be construed as requiring the operations to be performed in the specific order shown or sequentially, or to perform all shown operations to achieve the desired result. Furthermore, the separation of various system components in the implementations described in this disclosure should not be construed as requiring such separation in all implementations. Only some implementations and examples have been described, and other implementations, enhancements, and variations can be made based on what is described and shown in this disclosure.

Claims

1. A delivery management method, comprising: Acquire delivery data associated with multiple delivery time points in a first time window, the delivery data including: resource costs associated with delivery time points among the multiple delivery time points and the contribution of the resource costs to the delivery objective; A second time window is determined, specified by the data provider, to verify whether the contribution meets the deployment objective; the first length of the first time window is greater than the second length of the second time window; and Based on the delivery data and the first and second time windows, a prediction model is obtained. The prediction model indicates the correlation between delivery data associated with multiple previous delivery time points in the third time window and the predicted resource cost. The predicted resource cost indicates the total resource cost corresponding to the fourth time window after the third time window. The length of the third time window is less than the first length, and the length of the fourth time window is equal to the second length.

2. The method according to claim 1, wherein determining the prediction model comprises: For each of the multiple deployment time points within the first time window, an intermediate model is determined based on an inverse proportional function, whereby the inverse proportional function describes the degree of influence of the resource cost on the contribution; and The prediction model is determined based on the intermediate model and the delivery data.

3. The method according to claim 2, wherein the inverse proportional function is a linear inverse proportional function represented by a set of linear parameters.

4. The method according to claim 3, wherein determining the prediction model based on the intermediate model and the delivery data comprises: Based on the linear inverse proportional function and the delivery data, a set of candidate values ​​for the set of linear parameters is determined; as well as The prediction model is represented by the set of candidate values.

5. The method of claim 4, wherein determining the set of candidate values ​​comprises: For each of the plurality of delivery time points in the first time window, the linear inverse proportional function is updated using the resource cost associated with the delivery time point and the contribution of the resource cost to the delivery objective to determine the set of candidate values.

6. The method according to claim 1, further comprising: Obtain previous delivery data associated with multiple previous delivery time points within a previous time window, the previous delivery data including: previous resource costs associated with previous delivery time points among the multiple previous delivery time points and previous contributions of the previous resource costs to the delivery objective; and Based on the prediction model and the previously deployed data, the predicted resource cost associated with subsequent time windows is determined.

7. The method of claim 6, wherein determining the predicted resource cost associated with the subsequent time window comprises: Obtain the unit cost and resource threshold specified by the data provider; as well as Under the constraints of the unit cost and the resource threshold, the predicted resource cost associated with the subsequent time window is determined based on the prediction model and the previously deployed data.

8. The method of claim 7, wherein determining the predicted resource cost under the constraint of the unit cost comprises: Based on the prediction model and the previously deployed data, the costs of multiple candidate resources are determined; as well as The candidate resource cost that satisfies the constraint of the unit cost is selected from the plurality of candidate resource costs to be used as the predicted resource cost, wherein the unit cost is represented by the candidate resource cost and the predicted contribution corresponding to the candidate resource cost.

9. The method of claim 7, wherein determining the predicted resource cost under the constraint of the resource threshold comprises: The costs of the plurality of candidate resources are determined under the constraint of the resource threshold, wherein the costs of the plurality of candidate resources are lower than the resource threshold.

10. The method of claim 1, wherein the first length of the first time window is greater than the time delay between the time point at which the contribution is received and the time point.

11. An electronic device comprising a computer processor coupled to a computer-readable storage unit, the storage unit including instructions that, when executed by the computer processor, implement a method for delivery management, the method comprising: Acquire delivery data associated with multiple delivery time points in a first time window, the delivery data including: resource costs associated with delivery time points among the multiple delivery time points and the contribution of the resource costs to the delivery objective; A second time window is determined, specified by the data provider, to verify whether the contribution meets the deployment objective; the first length of the first time window is greater than the second length of the second time window; and Based on the delivery data and the first and second time windows, a prediction model is obtained. The prediction model indicates the correlation between delivery data associated with multiple previous delivery time points in the third time window and the predicted resource cost. The predicted resource cost indicates the total resource cost corresponding to the fourth time window after the third time window. The length of the third time window is less than the first length, and the length of the fourth time window is equal to the second length.

12. The electronic device of claim 11, wherein determining the prediction model comprises: For each of the multiple deployment time points within the first time window, an intermediate model is determined based on an inverse proportional function, whereby the inverse proportional function describes the degree of influence of the resource cost on the contribution; and The prediction model is determined based on the intermediate model and the delivery data.

13. The electronic device according to claim 12, wherein the inverse proportional function is a linear inverse proportional function represented by a set of linear parameters.

14. The electronic device of claim 13, wherein determining the prediction model based on the intermediate model and the delivery data comprises: Based on the linear inverse proportional function and the delivery data, a set of candidate values ​​for the set of linear parameters is determined; as well as The prediction model is represented by the set of candidate values.

15. The electronic device of claim 14, wherein determining a set of candidate values ​​comprises: For each of the plurality of delivery time points in the first time window, the linear inverse proportional function is updated using the resource cost associated with the delivery time point and the contribution of the resource cost to the delivery objective to determine the set of candidate values.

16. The electronic device of claim 11, wherein the method further comprises: Obtain previous delivery data associated with multiple previous delivery time points within a previous time window, the previous delivery data including: previous resource costs associated with previous delivery time points among the multiple previous delivery time points and previous contributions of the previous resource costs to the delivery objective; and Based on the prediction model and the previously deployed data, the predicted resource cost associated with subsequent time windows is determined.

17. The electronic device of claim 16, wherein determining the predicted resource cost associated with the subsequent time window comprises: Obtain the unit cost and resource threshold specified by the data provider; as well as Under the constraints of the unit cost and the resource threshold, the predicted resource cost associated with the subsequent time window is determined based on the prediction model and the previously deployed data.

18. The electronic device of claim 17, wherein determining the predicted resource cost under the constraint of the unit cost comprises: Based on the prediction model and the previously deployed data, the costs of multiple candidate resources are determined; as well as The candidate resource cost that satisfies the constraint of the unit cost is selected from the plurality of candidate resource costs to be used as the predicted resource cost, wherein the unit cost is represented by the candidate resource cost and the predicted contribution corresponding to the candidate resource cost.

19. The electronic device of claim 17, wherein determining the predicted resource cost under the constraint of the resource threshold comprises: The costs of the plurality of candidate resources are determined under the constraint of the resource threshold, wherein the costs of the plurality of candidate resources are lower than the resource threshold.

20. A non-transitory computer program product, the computer program product comprising a computer-readable storage medium having program instructions embodied therein, the program instructions being executed by an electronic device to cause the electronic device to perform a method for delivery management, the method comprising: Acquire delivery data associated with multiple delivery time points in a first time window, the delivery data including: resource costs associated with delivery time points among the multiple delivery time points and the contribution of the resource costs to the delivery objective; A second time window is determined, specified by the data provider, to verify whether the contribution meets the deployment objective; the first length of the first time window is greater than the second length of the second time window; and Based on the delivery data and the first and second time windows, a prediction model is obtained. The prediction model indicates the correlation between delivery data associated with multiple previous delivery time points in the third time window and the predicted resource cost. The predicted resource cost indicates the total resource cost corresponding to the fourth time window after the third time window. The length of the third time window is less than the first length, and the length of the fourth time window is equal to the second length.

Images