Budget allocation method and device, electronic equipment and storage medium

Abstract

The invention provides a budget allocation method and apparatus, an electronic device and a storage medium. The method comprises the steps of obtaining historical bubbling data within a preset time period; the original feature vectors of the historical bubbling data are clustered, a plurality of clusters are obtained, and each cluster corresponds to an implicit scene; based on the historical bubbling data, learning a price elasticity curve of each implicit scene; and based on the learned price elasticity curve of each implicit scene, determining an optimal budget allocation strategy of each implicit scene under the total subsidy budget limit. According to the method and the device, the implicit scene division is obtained based on the representation clustering of the implicit low-dimensional space for the obtained original feature vector of the historical bubble data, and the price elasticity curve of each implicit scene is learned, so that the optimal budget allocation strategy of different implicit scenes is formulated, the labor cost is effectively saved, and the subsidy resource utilization rate is improved.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a budget allocation method, apparatus, electronic device, and storage medium. Background Technology

[0002] In marketing subsidy pricing, the total daily subsidy budget is usually fixed by B. How to reasonably allocate subsidies within the budget B and accurately distribute them to the corresponding business scenarios in order to maximize business goals is a problem that needs to be solved in marketing subsidy pricing.

[0003] In related technologies, subsidies are often allocated by manually dividing explicit business scenarios, which has problems such as high labor costs, insufficient allocation accuracy, and low utilization of subsidy resources, making it difficult to achieve the business goal of maximizing GMV or order volume under fixed budget constraints. Summary of the Invention

[0004] This application provides a budget allocation method, apparatus, electronic device, and storage medium to at least address the problems in related technologies where allocating subsidies by manually dividing explicit business scenarios results in high labor costs, insufficient allocation accuracy, and low utilization of subsidy resources. The technical solution of this application is as follows: According to a first aspect of the embodiments of this application, a budget allocation method is provided, comprising: Retrieve historical bubble data within a preset time period; The original feature vectors of the historical bubble data are clustered to obtain multiple clusters, each cluster corresponding to an implicit scenario; Based on the historical bubbling data, learn the price elasticity curve for each implicit scenario; Based on the price elasticity curve of each implicit scenario learned, the optimal budget allocation strategy for each implicit scenario is determined within the total subsidy budget.

[0005] Optionally, the step of clustering the original feature vectors of the historical bubble data to obtain multiple clusters, each cluster corresponding to an implicit scenario, includes: The original feature vectors of the historical bubble data are subjected to representation learning to obtain low-dimensional representation vectors. The low-dimensional representation vector is clustered to obtain multiple clusters, each cluster corresponding to an implicit scenario.

[0006] Optionally, the step of performing representation learning on the original feature vectors of the historical bubble data to obtain low-dimensional representation vectors includes: Extract the original feature vectors from the historical bubble data; The extracted original feature vectors are cleaned and feature-engineered to obtain the processed original feature vectors; The original feature vector after processing is subjected to representation learning using a deep neural network autoencoder model to obtain a low-dimensional representation vector.

[0007] Optionally, the step of using a deep neural network autoencoder model to perform representation learning on the processed original feature vector to obtain a low-dimensional representation vector includes: Unsupervised representation learning is performed on the preprocessed original feature vector using a deep neural network autoencoder model. The encoder in the deep neural network autoencoder model compresses and maps the high-dimensional original feature vector to an implicit low-dimensional space, outputting a low-dimensional representation vector with a dimension much smaller than that of the original feature vector.

[0008] Optionally, the clustering of the low-dimensional representation vector to obtain multiple clusters, each cluster corresponding to an implicit scenario, includes: The K-Means clustering algorithm is used to cluster the low-dimensional representation vector in the multidimensional representation vector space to obtain k clusters. Each cluster corresponds to an implicit scenario. The value of k is preset and is the same as the number of implicit scenarios required.

[0009] Optionally, learning the price elasticity curve for each implicit scenario based on the historical bubbling data includes: Based on the historical bubbling data, extract the historical business datasets corresponding to each implicit scenario; Based on the historical business dataset, the price elasticity curve for each implicit scenario is fitted and learned to obtain the order volume for each implicit scenario under different subsidy rates; the price elasticity curve.

[0010] Optionally, based on the price elasticity curve of each implicit scenario learned, and within the total subsidy budget, determining the optimal budget allocation strategy for each implicit scenario includes: Based on the order completion volume of each implicit scenario under different subsidy rates, an optimization model is constructed to maximize business objectives within the total subsidy budget. Based on the objective function of the optimization model that maximizes business objectives, the optimal subsidy rate for each implicit scenario is obtained.

[0011] According to a second aspect of the embodiments of this application, a budget allocation apparatus is provided, comprising: The acquisition module is used to acquire historical bubble data within a preset time period; The clustering module is used to cluster the original feature vectors of the historical bubble data to obtain multiple clusters, each cluster corresponding to an implicit scenario; The learning module is used to learn the price elasticity curve for each implicit scenario based on the historical bubbling data; The budget determination module is used to determine the optimal budget allocation strategy for each implicit scenario based on the learned price elasticity curve for each implicit scenario, within the total subsidy budget.

[0012] Optionally, the clustering module includes: The representation learning module is used to perform representation learning on the original feature vectors of the historical bubble data to obtain low-dimensional representation vectors. The representation clustering module is used to cluster the low-dimensional representation vector to obtain multiple clusters, each cluster corresponding to an implicit scenario.

[0013] Optionally, the representation learning module includes: The extraction module is used to extract the original feature vector of the historical bubble data; The cleaning module is used to clean and perform feature engineering on the extracted original feature vectors to obtain the processed original feature vectors. The low-dimensional representation vector determination module is used to perform representation learning on the processed original feature vector using a deep neural network autoencoder model to obtain a low-dimensional representation vector.

[0014] Optionally, the low-dimensional representation vector determination module includes: The unsupervised representation learning module is used to perform unsupervised representation learning on the preprocessed original feature vector using a deep neural network autoencoder model. The mapping module is used to compress and map the high-dimensional original feature vector to an implicit low-dimensional space through the encoder in the deep neural network autoencoder model, and output a low-dimensional representation vector with a dimension much smaller than that of the original feature vector.

[0015] Optionally, the clustering module is specifically used to cluster the low-dimensional representation vector in the multidimensional representation vector space using the K-Means clustering algorithm to obtain k clusters, each cluster corresponding to an implicit scenario, wherein the value of k is preset and is the same as the number of implicit scenarios required.

[0016] Optionally, the learning module includes: The business data extraction module is used to extract historical business datasets corresponding to each implicit scenario based on the historical bubbling data. The fitting learning module is used to fit and learn the price elasticity curve of each implicit scenario based on the historical business dataset, and obtain the order completion volume of each implicit scenario under different subsidy rates.

[0017] Optionally, based on the price elasticity curve of each implicit scenario learned, and within the total subsidy budget, determining the optimal budget allocation strategy for each implicit scenario includes: The module is used to build an optimization model that maximizes business objectives based on the order completion volume of each implicit scenario under different subsidy rates, within the total subsidy budget. The subsidy determination module is used to solve for the optimal subsidy rate for each implicit scenario based on the objective function of the optimization model that maximizes the business objective.

[0018] According to a third aspect of the embodiments of this application, an electronic device is provided, comprising: It includes a processor, a memory; and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the steps of the budget allocation method as described above.

[0019] According to a fourth aspect of the embodiments of this application, a readable storage medium is provided, on which a program or instructions are stored, which, when executed by a processor of an electronic device, implement the steps of the budget allocation method as described above.

[0020] According to a fifth aspect of the present application, a computer program product is provided, including a computer program or instructions that, when executed by a processor of an electronic device, implement the steps of the budget allocation method as described above.

[0021] The technical solutions provided by the embodiments of this application bring at least the following beneficial effects: In this embodiment, historical bubble data within a preset time period is acquired; the original feature vectors of the historical bubble data are clustered to obtain multiple clusters, each cluster corresponding to an implicit scenario; based on the historical bubble data, the price elasticity curve of each implicit scenario is learned; based on the learned price elasticity curve of each implicit scenario, the optimal budget allocation strategy for each implicit scenario is determined within the total subsidy budget. In other words, in this embodiment, the original feature vectors of a large amount of historical bubble data are clustered to obtain multiple clusters, each cluster corresponding to an implicit scenario. This means that implicit scenario segmentation is obtained based on implicit low-dimensional space representation clustering, and the price elasticity curve of each implicit scenario is learned, thereby formulating optimal budget allocation strategies for different implicit scenarios, effectively saving manpower costs and improving the utilization rate of subsidy resources.

[0022] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0023] The accompanying drawings, incorporated in and forming part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application. They do not constitute an undue limitation of this application. To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0024] Figure 1 This is a flowchart of a budget allocation method provided in an embodiment of this application.

[0025] Figure 2 This is a block diagram of a budget allocation device provided in an embodiment of this application.

[0026] Figure 3 This is a block diagram of a clustering module provided in an embodiment of this application.

[0027] Figure 4 This is a block diagram of a representation learning module provided in an embodiment of this application.

[0028] Figure 5 This is a block diagram of a low-dimensional representation vector determination module provided in an embodiment of this application.

[0029] Figure 6 This is a block diagram of a learning module provided in an embodiment of this application.

[0030] Figure 7 This is a block diagram of a budget determination module provided in an embodiment of this application.

[0031] Figure 8 This is a block diagram of an electronic device provided in an embodiment of this application.

[0032] Figure 9 This is a block diagram of a budget allocation device provided in an embodiment of this application. Detailed Implementation

[0033] To enable those skilled in the art to better understand the technical solutions of this application, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings.

[0034] It should be noted that the terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0035] Please see Figure 1 This is a flowchart of a budget allocation method provided in an embodiment of this application, such as... Figure 1 As shown, the method includes the following steps: Step 101: Obtain historical bubble data within a preset time period.

[0036] Step 102: Cluster the original feature vectors of the historical bubble data to obtain multiple clusters, each cluster corresponding to an implicit scenario.

[0037] Step 103: Based on the historical bubbling data, learn the price elasticity curve for each implicit scenario.

[0038] Step 104: Based on the learned price elasticity curve for each implicit scenario, determine the optimal budget allocation strategy for each implicit scenario within the total subsidy budget.

[0039] In this embodiment, the original feature vectors of the acquired historical bubble data are clustered to obtain different implicit scenarios, and the price elasticity curve of each implicit scenario is learned. This allows for the development of optimal budget allocation strategies for different implicit scenarios, applicable to business scenarios requiring dynamic adjustment of prices and budgets, such as advertising, e-commerce promotions, and marketing subsidy activities. Using this application can effectively save labor costs and improve the utilization rate of subsidy resources.

[0040] The budget allocation method described in this application can be applied to terminals, servers, etc., without limitation. The terminal implementation device can be an electronic device such as a smartphone, laptop, tablet, desktop computer, personal digital assistant (PDA), and wearable device. The server can be an independent server, a server cluster, or a server that provides cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks, or big data and artificial intelligence platforms, etc., without limitation.

[0041] The following is combined Figure 1 The specific implementation steps of a budget allocation method provided in the embodiments of this application will be described in detail.

[0042] In step 101, historical bubble data within a preset time period is obtained.

[0043] This step involves acquiring massive amounts of historical spoofing data from a ride-hailing platform within a preset time period. The preset time period can be 10 days, 1 month, 3 months, 6 months, etc., and this example does not impose any restrictions. The historical spoofing data can include full detailed data such as: city, time, mileage, base price, weather, driver cruising rate, call acceptance rate, spatiotemporal itinerary, historical orders, price subsidies, operational behavior, environmental scenarios, and data quality logs. It should be noted that all this data is authorized data from all parties involved.

[0044] The acquisition method can be remote or local; this embodiment does not impose any restrictions. The specific acquisition method is well-known to those skilled in the art and will not be elaborated upon here. In this embodiment, bubbling data typically refers to dynamic data such as user behavior, transaction conversion, and price feedback generated in real time in a business scenario, such as ad clicks, product purchases, and promotional response data. This data is the basic data source for subsequent analysis and requires preprocessing such as data cleaning, deduplication, and standardization.

[0045] The historical bubbling data refers to the full historical detailed data that is reported by terminal behavior and then cleaned and archived. It can be archived on a "daily / weekly / monthly" basis, covering the complete business chain data of the past several weeks / months / years.

[0046] In step 102, the original feature vectors of the historical bubble data are clustered to obtain multiple clusters, each cluster corresponding to an implicit scenario.

[0047] This step includes: 1) performing representation learning on the original feature vectors of the historical bubble data to obtain low-dimensional representation vectors.

[0048] In this step, the original feature vector of the historical bubble data is first extracted; then the extracted original feature vector is cleaned and feature-engineered to obtain the processed original feature vector; finally, the processed original feature vector is learned by using a deep neural network autoencoder model to obtain a low-dimensional representation vector.

[0049] In this step, the extracted original feature vectors are cleaned and feature-engineered. This involves cleaning operations such as filtering invalid data, removing outliers, and filling missing values ​​in the original feature vectors, as well as feature engineering operations such as feature encoding transformation, continuous feature normalization, and core feature selection. The resulting processed feature vectors are then used for subsequent unsupervised representation learning.

[0050] In this step, representation learning uses a deep neural network autoencoder model to learn the representations of the original feature vectors in the bubbling pattern, obtaining an embedding vector representation with a bubbling dimension (this representation vector dimension is much smaller than the original feature vector dimension). In other words, the original feature vectors are transformed into an implicit low-dimensional vector space. This means the model automatically learns effective features from the data, rather than directly using the original data. Good representation learning can capture the underlying patterns in the data and improve its effectiveness. Common representation learning methods include PCA (Principal Component Analysis), word2vec, and BERT (Autoencoder).

[0051] It should be noted that through representation learning, high-dimensional sparse features (features with hundreds of dimensions) can be transformed into low-dimensional dense vectors (the dimension can be customized to 8, 16, or 32 dimensions, etc. The higher the dimension, the more complex the computation and the higher the performance requirements).

[0052] Of course, in this embodiment, the original feature vector is subjected to representation learning to obtain the implicit representation vector (i.e., the low-dimensional space representation vector). The representation learning can also adopt deep learning (such as autoencoder, CNN, Transformer, etc.) or traditional machine learning (such as PCA, LDA) representation learning algorithms to map high-dimensional, heterogeneous historical bubbling data to a low-dimensional implicit space and extract the core features of the data, such as user preference features, scene features, price sensitivity features, etc., in order to reduce the subsequent computational complexity.

[0053] The process of using a deep neural network autoencoder model to perform representation learning on the processed original feature vector to obtain a low-dimensional representation vector includes: using a deep neural network autoencoder model to perform unsupervised representation learning on the preprocessed original feature vector; and using the encoder in the deep neural network autoencoder model to compress and map the high-dimensional original feature vector to an implicit low-dimensional space, outputting a low-dimensional representation vector with a dimension much smaller than that of the original feature vector.

[0054] In this step, unsupervised representation learning refers to the process of automatically mining core latent features from the original feature vectors of high-dimensional, unlabeled historical bubbling data without the need for manual scene labeling, and mapping them to a low-dimensional latent space to generate low-dimensional representation vectors.

[0055] In this embodiment, a deep neural network autoencoder (DAE) can be used as the core model for unsupervised representation learning (i.e., called the deep neural network autoencoder model). This model consists of an encoder and a decoder. Through unsupervised training using a "compression-reconstruction" method, the encoder learns the core correlation information of the original features. Specifically, the encoder can compress the high-dimensional original feature vector (e.g., 200-dimensional) into a low-dimensional representation vector (e.g., 8-dimensional), thereby extracting the core features. The decoder reconstructs the low-dimensional representation vector into a vector with the same dimensions as the original features, ensuring that the low-dimensional vector retains the core information of the original data by minimizing the reconstruction error.

[0056] In this embodiment, the deep neural network autoencoder model needs to be trained. The training objective is to unsupervised optimize the reconstruction error (mean squared error, MSE), without requiring manual labeling of scenes, subsidy sensitivity, or other labels. The specific training process may include: 11) Input data preparation: Input the original feature vectors of historical bubble data after cleaning, encoding and normalization into the model for training. The purpose of cleaning, encoding and normalization is to ensure that the data is free of anomalies and has consistent dimensions.

[0057] 12) Model training: Train the autoencoder in an unsupervised manner, setting the Adam optimizer (learning rate of 0.001 as an example), batch size of 100, and number of iterations of 200 rounds. Use a Dropout layer (dropout rate of 0.2) to prevent overfitting until the reconstruction error is ≤5%, which completes the model training.

[0058] 13) Representation extraction: After training, discard the decoder and keep only the encoder. Input the original feature vector into the encoder and output a low-dimensional vector, which is the low-dimensional representation vector obtained by unsupervised learning.

[0059] In this embodiment, feature extraction and dimensionality reduction of historical bubbling data are achieved through unsupervised representation learning: with a deep neural network autoencoder as the core model, the encoder compresses the high-dimensional original feature vector into a low-dimensional representation vector without manual annotation. The decoder optimizes the model training by reconstructing the error, and finally obtains a low-dimensional representation vector that retains the core association of the original data, providing high-quality input for subsequent implicit scene clustering.

[0060] 2) Cluster the low-dimensional representation vectors to obtain multiple clusters, each cluster corresponding to an implicit scenario.

[0061] In this step, the K-Means clustering algorithm can be used to cluster the low-dimensional representation vectors in the multi-dimensional representation vector space, resulting in k clusters. Each cluster corresponds to an implicit scenario, where the value of k is preset and is the same as the number of implicit scenarios required. In other words, the value of k is set manually; that is, the value of k is set according to the number of implicit scenarios needed. It should be noted that the implicit scenarios in this embodiment do not have explicit physical meaning, but only implicit, intrinsic connections that are not understood by humans.

[0062] In other words, this step utilizes the K-Means clustering algorithm to cluster low-dimensional representation vectors in a multi-dimensional representation vector space. The aim is to efficiently group and mine implicit scenario relationships within the data, generating k implicit scenarios to provide scenario-dimensional segmentation for subsequent price elasticity learning and budget allocation. Specifically, this includes: 21) Clustering data preparation.

[0063] The low-dimensional representation vector set obtained from unsupervised representation learning is used as the clustering input. The vector set has been normalized and has no outliers, ensuring that the distribution pattern of all samples in the multi-dimensional representation vector space can be mined. At the same time, the number of clusters k is preset according to the business granularity requirements of subsidy budget allocation (k may be tens, hundreds or thousands, depending on the complexity of the travel platform scenario).

[0064] 22) K-Means clustering parameter configuration.

[0065] 221) Cluster center initialization: The K-Means++ algorithm is used to select the initial cluster centers to avoid local optima in the clustering results caused by random initialization and improve the stability of clustering.

[0066] 222) Distance measurement method: Euclidean distance is used to calculate the similarity between the sample and the cluster center. The specific calculation process is well known to those skilled in the art and will not be described in detail here.

[0067] 223) The clustering iteration termination condition can be set to a maximum of 300 iterations (other values ​​can also be set, which are not limited in this embodiment), or when the change in the position of the cluster center is ≤10. -6 The iteration is terminated when the time is right, balancing clustering efficiency and result convergence.

[0068] 23) Clustering execution and cluster partitioning.

[0069] The low-dimensional representation vector is input into the configured K-Means model, and the Euclidean distance from each sample to each cluster center is calculated iteratively. The sample is then assigned to the cluster to which the nearest cluster center belongs. After the iteration is completed, k independent clusters (i.e. k implicit scenarios) are obtained. The low-dimensional representation features of samples within each cluster are highly similar, while the features of samples between clusters are significantly different.

[0070] In step 103, the price elasticity curve for each implicit scenario is learned based on the historical bubbling data.

[0071] In this step, based on the historical bubble data, historical business datasets corresponding to each implicit scenario are extracted. These datasets may include: different subsidy amounts under each implicit scenario and full historical details of the corresponding business metrics. Based on these historical business datasets, a price elasticity curve for each implicit scenario is fitted and learned to obtain the order completion volume for each implicit scenario under different subsidy rates. The price elasticity curve represents the quantitative correlation between changes in subsidy amounts and changes in business metrics within the corresponding implicit scenario.

[0072] Optionally, the price elasticity curve uses the rate of change in subsidies as the independent variable and the rate of change in business indicators as the dependent variable. The price elasticity coefficient corresponding to the curve is calculated using historical data on changes in subsidies and business indicators, and is used to reflect the sensitivity of each implicit scenario to changes in subsidies.

[0073] In this step, after dividing all historical bubble data into different implicit scenarios, for each implicit scenario, historical bubble data such as weather, humidity, temperature, whether it is snowing, number of empty vehicles, minimum number of completed orders, 25th percentile of completed orders, median number of completed orders, 75th percentile of completed orders, maximum number of completed orders, and various percentiles of the number of hands raised can be used to fit the relationship curve between subsidy rate and number of completed orders, i.e., the price elasticity curve (the change in sales volume under different prices), through logistic regression, decision tree regression, or XBG.

[0074] In step 104, based on the price elasticity curve of each implicit scenario learned, the optimal budget allocation strategy for each implicit scenario is determined within the total subsidy budget.

[0075] In this step, based on the order completion volume of each implicit scenario under different subsidy rates, an optimization model that maximizes the business objective is constructed within the total subsidy budget. Based on the objective function of the constructed optimization model that maximizes the business objective, the optimal subsidy rate for each implicit scenario is obtained.

[0076] The optimization model is constructed by "using the subsidy rate of each implicit scenario as the decision variable, the total subsidy budget as the hard constraint, and maximizing the core business objectives as the ultimate goal". Based on the correspondence between "subsidy rate and number of completed orders" in each implicit scenario (derived from the price elasticity curve), a mathematical optimization model is built to fully align with the actual business situation of the travel platform's subsidy budget allocation.

[0077] In other words, in this embodiment, a mathematical model (i.e., optimization model) is constructed to maximize business objectives (GMV, order volume) under the constraint of the total subsidy budget. An operations research algorithm is then used to solve this system of equations, and the solver yields the optimal budget allocation strategy (or budget allocation scheme). The budget allocation strategy may include the proportion of subsidy amounts for each implicit scenario and the corresponding subsidy adjustment range. Finally, the operations research problem can be solved using the Lagrange method, a well-known technique that will not be elaborated upon here.

[0078] The objective function of the optimization model that maximizes business objectives is shown below: F

[0079] st

[0080]

[0081] In this formula, si is the subsidy rate of implicit scenario i; gi(si) is the order volume of implicit scenario i under the subsidy rate si, which can be obtained through the price elasticity curve; and pi is the average price per order of implicit scenario i.

[0082] Where GMV stands for "Total Transaction Value," and for a single implicit scenario, GMV = number of completed orders × average price per order. Total GMV is the sum of GMV across all scenarios. The objective function is: Fmax = .

[0083] The core of solving the objective function is to find a set of optimal subsidy rates {s_1,s_2,...,s_k} that maximizes the total GMV of all implicit scenarios, aligning with the platform's core demand for "increasing revenue".

[0084] The above operations research optimization algorithm formula can be used to solve for the optimal subsidy rate in each implicit scenario, maximizing GMV (business objective) while satisfying the total subsidy budget constraint. Operations research optimization algorithms are common in operations research. Given an objective: maximizing revenue, constraints are listed after st (the initial cost). Here, there are two constraints: st

[0085] ).

[0086] In this embodiment, the optimization model can use the subsidy allocation amount for each implicit scenario as the decision variable, the total subsidy budget limit as the constraint, and the maximization of the sum of business indicators for each implicit scenario as the optimization objective. The optimization model can be solved using linear programming or convex optimization algorithms through a solver. However, in specific applications, it is not limited to this.

[0087] The solver can be a Gurobi or CPLEX solver. The optimal subsidy budget allocation strategy obtained by the solution may include the proportion of subsidy amount for each implicit scenario and the corresponding subsidy adjustment range, but it is not limited to this. It can also be adaptively adjusted according to actual needs. This embodiment does not impose any restrictions.

[0088] Yes, the method may also include: re-collecting the latest business data at a preset period, repeating steps 101 to 104, and updating the implicit scenario boundary, price elasticity curve and optimal subsidy budget allocation scheme to adapt to dynamic changes in business.

[0089] In this embodiment, historical bubble data within a preset time period is acquired; the original feature vectors of the historical bubble data are clustered to obtain multiple clusters, each cluster corresponding to an implicit scenario; based on the historical bubble data, the price elasticity curve of each implicit scenario is learned; based on the learned price elasticity curve of each implicit scenario, the optimal budget allocation strategy for each implicit scenario is determined within the total subsidy budget. In other words, in this embodiment, the original feature vectors of a large amount of historical bubble data are clustered to obtain multiple clusters, each cluster corresponding to an implicit scenario. This means that implicit scenario segmentation is obtained based on implicit low-dimensional space representation clustering, and the price elasticity curve of each implicit scenario is learned, thereby formulating optimal budget allocation strategies for different implicit scenarios, effectively saving manpower costs and improving the utilization rate of subsidy resources.

[0090] For better understanding, please refer to... Figure 1 See the following application examples.

[0091] First application example: 1. Collect historical bubble data and extract the original features (i.e., the original feature vector, including but not limited to the features described below).

[0092] 11) Basic characteristics.

[0093] City: The city where the event took place (such as Beijing, Shanghai, etc.).

[0094] Time: The time when the event occurred (e.g., 08:00 on October 1, 2023).

[0095] Mileage: The total distance traveled for the order (e.g., 5.3 kilometers).

[0096] Base price: The original price of the order (e.g., 10 yuan).

[0097] Weather: The weather conditions at the time of the event (e.g., sunny, rainy, snowy, etc.).

[0098] Temperature: The temperature at that time (e.g., 25°C).

[0099] Humidity: The ambient air humidity at that time (e.g., 60%).

[0100] Is it peak period: Is it peak period (1 indicates peak, 0 indicates off-peak).

[0101] 12) Behavioral characteristics (including but not limited to the following characteristics).

[0102] Driver cruising rate: The percentage of idle drivers within a specific grid in a city during the current time period (e.g., 0.75%).

[0103] Call acceptance rate: The percentage of calls accepted by drivers within a specific grid in a city during the current time period (e.g., 0.85).

[0104] Order cancellation rate: The percentage of orders canceled within a specific grid in a city during the current time period (e.g., 0.1%).

[0105] 13) Spatiotemporal characteristics (including but not limited to the following characteristics).

[0106] Latitude and longitude: The latitude and longitude of the order's origin (e.g., [116.4074, 39.9042]).

[0107] Region division: The grid to which the geographical location belongs (e.g., changing to the h3 grid).

[0108] Time window: the hour in which the event occurs (e.g., Monday 08:00~09:00 corresponds to 8).

[0109] Road congestion status: degree of road congestion (e.g., 0: smooth traffic, 1: congested, 2: severe congestion).

[0110] 14) User characteristics (including but not limited to the following characteristics).

[0111] User type: Whether the user is a new user (1 indicates a new user, 0 indicates an old user). Historical order frequency: The number of orders placed by a user in the last 7 days (e.g., 15 orders).

[0112] 15) Contextual features (including but not limited to the following features).

[0113] Holiday marker: Whether it is a holiday (1 indicates a holiday, 0 indicates a non-holiday).

[0114] City Tier: The current city tier (e.g., first-tier city, second-tier city, etc.).

[0115] 2. Representation learning.

[0116] In this embodiment, the goal of representation learning is to map high-dimensional feature vectors to a low-dimensional implicit space while preserving as much of the main information of the original features as possible. This embodiment uses an autoencoder in a deep neural network as an example for representation learning. An autoencoder is an unsupervised learning model. Deep neural networks mainly consist of an encoder and a decoder; this embodiment only uses the encoder. The encoder compresses the high-dimensional input features into a low-dimensional implicit space. The decoder reconstructs the original high-dimensional input from the low-dimensional implicit space.

[0117] In this embodiment, an unsupervised learning model is used to learn the representation of the feature vector of the bubble data, which includes the above-mentioned features. Specifically, this includes: First, one-hot encoding is applied to the non-numerical feature vectors. The non-numerical feature vectors are converted into numerical feature vectors (e.g., weather - sunny day is converted to [1, 0,0,0,0,...,0]), and then encoded. The encoded feature vector is as follows: a = [1,0,0,0, 8, 5.3, 15.0, 1,0,0, 0.75, 0.85, 1, ..., 0]. Assuming the original feature vector dimension d = 200, then the length of a is 200.

[0118] Secondly, an autoencoder model is constructed, and the 200-dimensional original vector is compressed into an implicit space of n = 8 dimensions using the autoencoder model. The autoencoder model will output an 8-dimensional embedding vector, such as [0.5, -0.2, 0.8, -0.3, 0.7, -0.2, 0.8, -2].

[0119] 3. Characterization clustering.

[0120] Suppose our current business requires dividing data into 1000 implicit scenarios. Taking the K-Means clustering algorithm (i.e., clustering model) as an example, we set k=1000, using the embedding vectors learned from representations as input to K-Means. The output is which class each vector belongs to; each class represents an implicit scenario. Within the same implicit scenario, the price sensitivity of each bubble data point is almost identical; the same implicit scenario can be said to share the same price sensitivity curve. Different implicit scenarios correspond to different price sensitivity curves.

[0121] 4. Study the price sensitivity curve F (i.e., the price elasticity curve), the operations section.

[0122] Suppose that 1000 implicit scenarios are defined above. In this embodiment, OR-tools can be used to determine the appropriate subsidy rate for each implicit scenario to maximize GMV. This yields the subsidy rate for each of the 1000 implicit scenarios, for example, [15%, 10%, 0%, 2%, ..., 20%]. This subsidy rate maximizes GMV. Finally, the subsidy rate for each implicit scenario can be applied online. The process is as follows: The platform receives a bubbling data set, obtains an embedding vector through representation learning, and then inputs it into a clustering model to obtain the corresponding implicit scenario ID. Assuming the target subsidy rate for this implicit scenario ID is 15%, then the target subsidy rate for all orders under this implicit scenario throughout the day is 15%. The specific amount to be distributed per order is further processed using the appropriate model, which will not be elaborated here.

[0123] Second application example: This application example uses a ride-hailing platform operating in 300 cities nationwide, generating over 100 million historical data points daily (including order information, driver activity, time and location data, and weather information). The platform's original subsidy method involved manually categorizing subsidies into explicit scenarios such as "morning rush hour in first-tier cities" and "rainy days in third-tier cities," resulting in low subsidy accuracy and insufficient budget utilization. After adopting the restructured subsidy allocation scheme provided in this application's embodiment, the subsidy ROI increased by 28%, and GMV increased by 22%. The specific process includes: 1. Obtain historical bubble data.

[0124] The platform can extract the full historical bubble data for the past month (using this as an example, but not limited to this in actual applications), including the following examples, but not limited to these in actual applications: 11) Basic attributes: city code, driver ID, order mileage, basic pricing.

[0125] 12) Operational behavior: Driver cruising rate (empty time / online time), call answering rate (number of accepted orders / total number of calls).

[0126] 13) Spatiotemporal environment: GeoHash code of the boarding point, travel time (morning peak / off-peak / evening peak), weather type (sunny / rainy / snowy).

[0127] 14) Transaction data: subsidy amount, number of completed orders, and average price per order.

[0128] 2. Representation learning, implicit low-dimensional space representation.

[0129] A deep neural network autoencoder is used to map the original 200-dimensional feature vector (containing all the features mentioned above) to an 8-dimensional implicit low-dimensional space, generating a low-dimensional representation vector. For example: 21) Original feature vector: [First-tier city, morning rush hour, rain, cruise rate 0.6, pick-up rate 0.7, mileage 8km, base price 15 yuan, ...].

[0130] 22) Low-dimensional representation vector: [0.23, 0.78, 0.11, 0.65, 0.32, 0.89, 0.45, 0.19] (no explicit physical meaning, only retaining the core feature association).

[0131] 3. Representation clustering yields implicit scenarios.

[0132] The K-Means clustering algorithm is used to cluster the 8-dimensional low-dimensional representation vector into k=1000 implicit scenarios (the number of clusters is preset manually). Each cluster corresponds to an implicit scenario without a clear physical name.

[0133] It should be noted that these implicit scenarios in this embodiment do not have predefined physical names. They are automatically divided by the algorithm through data feature association. They are difficult to define accurately by humans, but this application can truly reflect the core laws of platform operation.

[0134] 4. Learn the price elasticity curve for each scenario.

[0135] For each implicit scenario, extract the correlation between "subsidy rate - completed order volume" from historical data and fit a price elasticity curve: 41) Scenario 1: Price elasticity coefficient E=-3.2 (for every 1% increase in subsidy rate, the number of completed orders increases by 3.2%), which is a highly sensitive scenario (the subsidy has a significant effect on boosting the number of completed orders).

[0136] 42) Scenario 2: Price elasticity coefficient E=-0.8 (for every 1% increase in the subsidy rate, the number of completed orders increases by 0.8%), which is a low-sensitivity scenario (even if the subsidy is increased, the increase in the number of completed orders is limited because the core bottleneck is insufficient driver supply).

[0137] 43) Scenario 3: Price elasticity coefficient E=-1.5 (for every 1% increase in the subsidy rate, the order volume increases by 1.5%), which is a medium-sensitive scenario.

[0138] 5. Operations research to find the optimal budget allocation scheme.

[0139] If the platform's total subsidy budget B = 10 million yuan, an optimization model is constructed with the goal of maximizing total GMV to solve for the optimal subsidy rate in each scenario. This model may also include the budget allocation ratio. The implicit scenario number is the optimal subsidy rate, budget allocation amount, budget ratio, and expected increase in completed orders. Scenario 1: 20% 4 million yuan 40% 32%.

[0140] Scenario 2: 5% for 1 million yuan, 10% for 4%.

[0141] Scenario 3 12% 2.5 million yuan 25% 18%.

[0142] The other three scenarios are 8%-15%, 2.5 million yuan, 25%, and 8%-12%.

[0143] In other words, allocating 40% of the budget to the highly sensitive scenario 1 and only 10% of the budget to the less sensitive scenario 2 not only avoids wasting subsidy resources, but also achieves a 22% increase in total GMV through precise allocation, which is far better than the effect of manual allocation.

[0144] This application's embodiments obtain implicit scene segmentation based on implicit low-dimensional space representation clustering. This scheme is automatic and efficient, and better matches the implicit relationships behind massive amounts of data. Furthermore, based on the automatic implicit scene segmentation, an automatic subsidy budget allocation mechanism is introduced between implicit scenes. Using operations research optimization algorithms, the optimal solution can be obtained, thereby achieving the optimal subsidy rate for each implicit scene.

[0145] It should be noted that, for the sake of simplicity, the method embodiments are all described as a series of actions. However, those skilled in the art should understand that this disclosure is not limited to the described order of actions, because according to this application, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also understand that the embodiments described in the specification are all preferred embodiments, and the actions involved are not necessarily essential to this application.

[0146] Please also see Figure 2 This is a block diagram of a budget allocation device provided in an embodiment of this application. The device includes: an acquisition module 201, a clustering module 202, a learning module 203, and a budget determination module 204, wherein... The acquisition module 201 is used to acquire historical bubble data within a preset time period; Clustering module 202 is used to cluster the original feature vectors of the historical bubble data to obtain multiple clusters, each cluster corresponding to an implicit scenario; Learning module 203 is used to learn the price elasticity curve for each implicit scenario based on the historical bubbling data; Budget determination module 204 is used to determine the optimal budget allocation strategy for each implicit scenario based on the learned price elasticity curve for each implicit scenario, within the total subsidy budget.

[0147] Optionally, the clustering module 202 includes a representation learning module 301 and a representation clustering module 302, the structural block diagram of which is shown below. Figure 3 As shown, where, The representation learning module 301 is used to perform representation learning on the original feature vector of the historical bubble data to obtain a low-dimensional representation vector. The characterization clustering module 302 is used to cluster the low-dimensional characterization vector to obtain multiple clusters, each cluster corresponding to an implicit scenario.

[0148] Optionally, the representation learning module 301 includes: an extraction module 401, a cleaning processing module 402, and a low-dimensional representation vector determination module 403, the block diagram of which is shown below. Figure 4 As shown, where, Extraction module 401 is used to extract the original feature vector of the historical bubble data; The cleaning and processing module 402 is used to clean and perform feature engineering on the extracted original feature vector to obtain the processed original feature vector. The low-dimensional representation vector determination module 403 is used to perform representation learning on the processed original feature vector using a deep neural network autoencoder model to obtain a low-dimensional representation vector.

[0149] Optionally, the low-dimensional representation vector determination module 403 includes: an unsupervised representation learning module 501 and a mapping module 502, the structure of which is as follows: Figure 5 As shown, Unsupervised representation learning module 501 is used to perform unsupervised representation learning on the preprocessed original feature vector using a deep neural network autoencoder model. The mapping module 502 is used to compress and map the high-dimensional original feature vector to an implicit low-dimensional space through the encoder in the deep neural network autoencoder model, and output a low-dimensional representation vector with a dimension much smaller than that of the original feature vector.

[0150] Optionally, the clustering module is specifically used to cluster the low-dimensional representation vector in the multidimensional representation vector space using the K-Means clustering algorithm to obtain k clusters, each cluster corresponding to an implicit scenario, wherein the value of k is preset and is the same as the number of implicit scenarios required.

[0151] Optionally, the learning module 203 includes: a business data extraction module 601 and a fitting learning module 602, the structural block diagram of which is shown below. Figure 6 As shown, where, The business data extraction module 601 is used to extract historical business datasets corresponding to each implicit scenario based on the historical bubble data; wherein, the datasets may include: full historical details of different subsidy amounts and corresponding business indicators under each scenario; The fitting learning module 602 is used to fit and learn the price elasticity curve of each implicit scenario based on the historical business dataset, and obtain the order completion volume of each implicit scenario under different subsidy rates.

[0152] Optionally, the budget determination module 204 includes a construction module 701 and a subsidy determination module 702, the structural block diagram of which is shown below. Figure 7 As shown, where, Module 701 is used to build an optimization model that maximizes business objectives based on the order completion volume of each implicit scenario under different subsidy rates, within the total subsidy budget. The subsidy determination module 702 is used to solve for the optimal subsidy rate for each implicit scenario based on the objective function of the constructed optimization model that maximizes the business objective.

[0153] Optionally, embodiments of this application also provide an electronic device, including: It includes a processor, a memory; and a program or instructions stored in the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the steps of the budget allocation method as described above.

[0154] Optionally, embodiments of this application also provide a readable storage medium storing a program or instructions that, when executed by a processor of an electronic device, implement the steps of the budget allocation method described above.

[0155] Optionally, embodiments of this application also provide a computer program product, including a computer program or instructions, which, when executed by a processor of an electronic device, implement the steps of the budget allocation method as described above.

[0156] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0157] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0158] Figure 8 This is a block diagram of an electronic device 800 provided in an embodiment of this application. For example, the electronic device 800 can be a mobile terminal or a server; in this embodiment, a mobile terminal is used as an example for explanation. For example, the electronic device 800 can be a mobile phone, computer, digital broadcasting terminal, messaging device, game console, tablet device, medical device, fitness equipment, personal digital assistant, etc.

[0159] Reference Figure 8 The electronic device 800 may include one or more of the following components: a processing component 802, a memory 804, a power component 806, a multimedia component 808, an audio component 810, an input / output (I / O) interface 812, a sensor component 814, and a communication component 816.

[0160] Processing component 802 typically controls the overall operation of electronic device 800, such as operations associated with display, telephone calls, data communication, camera operation, and recording operations. Processing component 802 may include one or more processors 820 to execute instructions to complete all or part of the steps of the methods described above. Furthermore, processing component 802 may include one or more modules to facilitate interaction between processing component 802 and other components. For example, processing component 802 may include a multimedia module to facilitate interaction between multimedia component 808 and processing component 802.

[0161] Memory 804 is configured to store various types of data to support the operation of device 800. Examples of this data include instructions for any application or method operating on electronic device 800, contact data, phonebook data, messages, pictures, videos, etc. Memory 804 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0162] Power supply component 806 provides power to various components of electronic device 800. Power supply component 806 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to electronic device 800.

[0163] Multimedia component 808 includes a screen that provides an output interface between the electronic device 800 and the user. In some embodiments, the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touchscreen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touches, swipes, and gestures on the touch panel. The touch sensors may sense not only the boundaries of the touch or swipe action but also the duration and pressure associated with the touch or swipe operation. In some embodiments, multimedia component 808 includes a front-facing camera and / or a rear-facing camera. When the device 800 is in an operating mode, such as a shooting mode or a video mode, the front-facing camera and / or the rear-facing camera may receive external multimedia data. Each front-facing camera and rear-facing camera may be a fixed optical lens system or have focal length and optical zoom capabilities.

[0164] Audio component 810 is configured to output and / or input audio signals. For example, audio component 810 includes a microphone (MIC) configured to receive external audio signals when electronic device 800 is in an operating mode, such as call mode, recording mode, and voice recognition mode. The received audio signals may be further stored in memory 804 or transmitted via communication component 816. In some embodiments, audio component 810 also includes a speaker for outputting audio signals.

[0165] I / O interface 812 provides an interface between processing component 802 and peripheral interface modules, such as keyboards, click wheels, buttons, etc. These buttons may include, but are not limited to, home buttons, volume buttons, power buttons, and lock buttons.

[0166] Sensor assembly 814 includes one or more sensors for providing state assessments of various aspects of electronic device 800. For example, sensor assembly 814 may detect the on / off state of device 800, the relative positioning of components such as the display and keypad of electronic device 800, changes in position of electronic device 800 or a component of electronic device 800, the presence or absence of user contact with electronic device 800, orientation or acceleration / deceleration of electronic device 800, and temperature changes of electronic device 800. Sensor assembly 814 may include a proximity sensor configured to detect the presence of nearby objects without any physical contact. Sensor assembly 814 may also include a light sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, sensor assembly 814 may also include an accelerometer, gyroscope, magnetometer, pressure sensor, or temperature sensor.

[0167] Communication component 816 is configured to facilitate wired or wireless communication between electronic device 800 and other devices. Electronic device 800 can access wireless networks based on communication standards, such as WiFi, carrier networks (such as 2G, 3G, 4G, or 5G), or combinations thereof. In one exemplary embodiment, communication component 816 receives broadcast signals or broadcast-related information from an external broadcast management system via a broadcast channel. In one exemplary embodiment, communication component 816 also includes a near-field communication (NFC) module to facilitate short-range communication. For example, the NFC module may be implemented based on radio frequency identification (RFID) technology, Infrared Data Association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology, and other technologies.

[0168] In the embodiments, the electronic device 800 may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the various processes of the budget allocation method embodiments shown above and achieve the same technical effects. To avoid repetition, it will not be described again here.

[0169] In one embodiment, a readable storage medium is also provided, on which a program or instructions are stored. When executed by a processor of a processing electronic device, the program or instructions implement the steps of the budget allocation method described above. The readable storage medium includes computer-readable storage media such as ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage devices.

[0170] In this embodiment, a computer program product is also provided, including a computer program or instructions. When the computer program or instructions are executed by the processor 820 of the electronic device 800, the electronic device 800 performs the various processes of the above-described budget allocation method embodiment and achieves the same technical effect. To avoid repetition, these will not be described again here.

[0171] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid state disk (SSD)).

[0172] Figure 9 This is a block diagram of an apparatus 900 for budget allocation provided in an embodiment of this application. For example, apparatus 900 can be provided as a server. See also... Figure 9 The apparatus 900 includes a processing component 922, which further includes one or more processors, and memory resources represented by memory 932 for storing instructions, such as application programs, that can be executed by the processing component 922. The application programs stored in memory 932 may include one or more modules, each corresponding to a set of instructions. Furthermore, the processing component 922 is configured to execute instructions to perform the methods described above.

[0173] Device 900 may also include a power supply component 926 configured to perform power management of device 900, a wired or wireless network interface 950 configured to connect device 900 to a network, and an input / output (I / O) interface 958. Device 900 may operate on an operating system stored in memory 932, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, or similar.

[0174] All historical bubble data, historical business datasets, and other related data involved in this application are data authorized by the respective units or parties.

[0175] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), and includes several instructions to cause a terminal (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0176] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A budget allocation method, characterized in that, include: Retrieve historical bubble data within a preset time period; The original feature vectors of the historical bubble data are clustered to obtain multiple clusters, each cluster corresponding to an implicit scenario; Based on the historical bubbling data, learn the price elasticity curve for each implicit scenario; Based on the price elasticity curve of each implicit scenario learned, the optimal budget allocation strategy for each implicit scenario is determined within the total subsidy budget.

2. The budget allocation method according to claim 1, characterized in that, The original feature vectors of the historical bubble data are clustered to obtain multiple clusters, each cluster corresponding to an implicit scenario, including: The original feature vectors of the historical bubble data are subjected to representation learning to obtain low-dimensional representation vectors. The low-dimensional representation vector is clustered to obtain multiple clusters, each cluster corresponding to an implicit scenario.

3. The budget allocation method according to claim 2, characterized in that, The process of learning the representation of the original feature vectors of the historical bubble data to obtain low-dimensional representation vectors includes: Extract the original feature vectors from the historical bubble data; The extracted original feature vectors are cleaned and feature-engineered to obtain the processed original feature vectors; The original feature vector after processing is subjected to representation learning using a deep neural network autoencoder model to obtain a low-dimensional representation vector.

4. The budget allocation method according to claim 3, characterized in that, The process of using a deep neural network autoencoder model to learn the representation of the processed original feature vector to obtain a low-dimensional representation vector includes: Unsupervised representation learning is performed on the preprocessed original feature vector using a deep neural network autoencoder model. The encoder in the deep neural network autoencoder model compresses and maps the high-dimensional original feature vector to an implicit low-dimensional space, outputting a low-dimensional representation vector with a dimension much smaller than that of the original feature vector.

5. The budget allocation method according to claim 2, characterized in that, The clustering of the low-dimensional representation vectors yields multiple clusters, each cluster corresponding to an implicit scenario, including: The K-Means clustering algorithm is used to cluster the low-dimensional representation vector in the multidimensional representation vector space to obtain k clusters. Each cluster corresponds to an implicit scenario. The value of k is preset and is the same as the number of implicit scenarios required.

6. The budget allocation method according to claim 1, characterized in that, The process of learning the price elasticity curve for each implicit scenario based on the historical bubbling data includes: Based on the historical bubbling data, extract the historical business datasets corresponding to each implicit scenario; Based on the historical business dataset, the price elasticity curve for each implicit scenario is fitted and learned to obtain the order volume for each implicit scenario under different subsidy rates; the price elasticity curve.

7. The budget allocation method according to claim 1, characterized in that, The price elasticity curve for each implicit scenario based on learning, under the total subsidy budget, determines the optimal budget allocation strategy for each implicit scenario, including: Based on the order completion volume of each implicit scenario under different subsidy rates, an optimization model is constructed to maximize business objectives within the total subsidy budget. Based on the objective function of the optimization model that maximizes business objectives, the optimal subsidy rate for each implicit scenario is obtained.

8. A budget allocation device, characterized in that, include: The acquisition module is used to acquire historical bubble data within a preset time period; The clustering module is used to cluster the original feature vectors of the historical bubble data to obtain multiple clusters, each cluster corresponding to an implicit scenario; The learning module is used to learn the price elasticity curve for each implicit scenario based on the historical bubbling data; The budget determination module is used to determine the optimal budget allocation strategy for each implicit scenario based on the learned price elasticity curve for each implicit scenario, within the total subsidy budget.

9. An electronic device, characterized in that, include: Including processor and memory; And a program or instructions stored on the memory and executable on the processor, wherein the program or instructions, when executed by the processor, implement the steps of the budget allocation method as described in any one of claims 1 to 7.

10. A readable storage medium, characterized in that, The readable storage medium stores a program or instructions that, when executed by a processor of an electronic device, implement the steps of the budget allocation method as described in any one of claims 1 to 7.

Images