A data product pricing method and device based on reinforcement learning and a medium

Abstract

The application discloses a data product pricing method and device based on reinforcement learning, and a medium. The method comprises the following steps: obtaining the cumulative total cost and the total number of calls of a target data product, and calculating a break-even bottom price of the target product; determining a value adjustment coefficient through a preset parameter mapping relationship according to multi-dimensional value evaluation parameters, and adjusting the break-even bottom price to obtain a benchmark price; calculating a price deviation value between the benchmark price and a market median price; when the price deviation value exceeds a preset threshold, generating an expected selling price based on the current operating state parameters of the market through a reinforcement learning agent; performing a gray-scale test on the expected selling price to determine a full-amount effective state; and based on the effective expected selling price, performing the verification and cancellation of the number of calls through a blockchain in response to the number of calls paid by a buyer. The cost and value are anchored, the reinforcement learning dynamic pricing and delivery feedback closed loop are linked, and the pricing can be adjusted according to market and quality changes.

Description

Technical Field

[0001] This application relates to the field of enterprise data product management technology, specifically to a data product pricing method, device, and medium based on reinforcement learning. Background Technology

[0002] With the rapid development of the digital economy, data has become an important factor of production, and its application scenarios in various industries are constantly expanding, leading to a growing market demand for data products. However, data differs from traditional tangible commodities, possessing unique characteristics such as non-excludability, ease of replication, uncertainty in quality, and time-sensitivity decay. These characteristics make it difficult for traditional commodity pricing and transaction models to directly adapt to data elements, creating numerous obstacles to the standardized and rational development of data transactions.

[0003] Currently, traditional data pricing methods typically employ cost-plus pricing, primarily using the direct costs incurred during data collection, storage, and initial processing as the calculation benchmark, setting a fixed percentage as a profit margin to determine the final transaction price of the data product. The core logic is that cost determines price, without any adaptation to the unique characteristics of data elements, simply equating data with ordinary tangible goods for pricing calculations. However, due to the ease of data replication, with replication costs near zero, the traditional cost-plus logic, which only calculates initial direct costs, fails to reflect the actual value of the data, leading to a severe disconnect between pricing and the data's use value. Furthermore, data's non-exclusivity, quality uncertainty, and time-degradation characteristics mean that this method lacks a mechanism linking pricing with delivery and feedback, making it difficult to track data quality. This easily leads to problems such as excessive data usage and multiple sales of the same data, and it's also difficult to adjust prices based on changes in data quality and market demand fluctuations, further distorting price signals. Summary of the Invention

[0004] To address the aforementioned issues, this application proposes a data product pricing method based on reinforcement learning, comprising: Obtain the cumulative total cost of the target data product within a preset historical period, as well as the total number of calls within the preset historical period; and calculate the break-even price of the target product based on the total cost and the total number of calls. Based on the multidimensional value assessment parameters of the target data product, a value adjustment coefficient is determined through a preset parameter mapping relationship. The break-even price is then adjusted based on the value adjustment coefficient to obtain the benchmark price of the target data product. Obtain the median market price of similar products in the market for the target data product, and calculate the price deviation between the benchmark price and the median market price; When the price deviation exceeds a preset threshold, the reinforcement learning agent determines the current price adjustment action from a preset set of price adjustment actions based on the current operating state parameters of the market, and generates the expected selling price of the target data product. The projected selling price is configured in a preset proportion of call traffic for canary testing, and the full effective status of the projected selling price is determined based on the test results; In response to the buyer's prepaid call count payment, the call count is reimbursed via blockchain based on the expected selling price after it takes effect, and buyer satisfaction data is collected after a preset call threshold is triggered, so as to adjust the value adjustment coefficient according to the collection results.

[0005] On the other hand, this application also proposes a data product pricing device based on reinforcement learning, comprising: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, which enables the at least one processor to perform a data product pricing method based on reinforcement learning as described in the above example.

[0006] On the other hand, this application also proposes a non-volatile computer storage medium storing computer-executable instructions, wherein the computer-executable instructions are configured as: a data product pricing method based on reinforcement learning as described in the above example.

[0007] The data product pricing method based on reinforcement learning proposed in this application can bring the following benefits: By combining cost accounting with multi-dimensional value assessment to determine the benchmark price, we ensure that the pricing of data products is not lower than the actual investment cost, while also reflecting the intrinsic value of data quality and scarcity. Based on this, a reinforcement learning agent is introduced for dynamic price adjustment. When there is a significant deviation between the benchmark price and market conditions, the agent can automatically select an adjustment action based on real-time market conditions, and the effect of the price adjustment is verified through gray-scale testing. This mechanism allows pricing to respond quickly to market changes while avoiding the risk of user churn caused by blind price adjustments.

[0008] Furthermore, after the buyer prepays the fee, a call certificate is generated on the blockchain. Each call synchronously verifies the available number of times. During the call process, buyer satisfaction is collected periodically, and negative feedback results are fed back into the value adjustment coefficient, so that subsequent pricing can automatically reflect the real changes in data quality. Attached Figure Description

[0009] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a flowchart illustrating a data product pricing method based on reinforcement learning, as described in an embodiment of this application. Figure 2 This is a schematic diagram of a data product pricing device based on reinforcement learning, as described in an embodiment of this application. Detailed Implementation

[0010] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of them. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0011] The technical solutions provided by the various embodiments of this application are described in detail below with reference to the accompanying drawings.

[0012] like Figure 1 As shown in the figure, this application provides a data product pricing method based on reinforcement learning, including: S101. Obtain the cumulative total cost of the target data product within a preset historical period, and the total number of calls within the preset historical period. Calculate the break-even price of the target product based on the total cost and the total number of calls.

[0013] In practice, the cumulative total cost of the target data product within a preset historical period, as well as the total number of calls within that period, are obtained. The cumulative total cost includes labor costs, computing power costs, and storage costs. Specifically, labor hours data from the previous calendar month are aggregated through the financial system interface, and labor costs are calculated based on a preset uniform hourly wage; the uptime of graphics processing units (GPUs) or tensor processors (TPUs) is obtained through the cloud service platform billing interface, and computing power costs are calculated based on a preset computing power unit price; storage and traffic fees are obtained through the cloud storage service bill and are used as storage costs. The cumulative total cost is obtained by adding these three costs together.

[0014] The total number of calls is obtained by counting the number of successful calls to the target data product within a preset historical period through the Application Programming Interface (API) gateway logs, after excluding free trial or test call records. Based on the cumulative total cost and the total number of calls, the break-even price of the target data product is calculated, specifically by dividing the cumulative total cost by the total number of calls to obtain the break-even price per call.

[0015] In this embodiment of the application, the break-even price Retrieve three types of invoices automatically summarized daily at 00:00 for the previous calendar month: (Information on manual labor hours) × Standard Hourly Wage GPU / TPU boot hours × Hourly unit price Cloud billing storage traffic fees To obtain the total monthly cost Based on total number of calls in the past 30 days Calculate the base price .

[0016] Furthermore, based on the multidimensional value assessment parameters of the target data product, a value adjustment coefficient is determined through a pre-defined parameter mapping relationship. These multidimensional value assessment parameters include data quality parameters, market scarcity parameters, price acceptance parameters, and buyer structure parameters.

[0017] In practice, data quality parameters are obtained by statistically analyzing the field missing rate of the target data product within a preset statistical period. The data quality parameters are then compared with the preset missing rate thresholds corresponding to each interval to determine the missing rate interval and obtain the quality score value corresponding to the missing rate interval.

[0018] The market scarcity parameter is obtained by querying the data exchange's market data API to count the number of products of the same category as the target data product. This scarcity parameter is then compared with preset competitor quantity thresholds for each interval to determine the corresponding competitor quantity interval and obtain a scarcity score for that interval. The price acceptance parameter is obtained by calculating the ratio of the target data product's average transaction price over a preset statistical period to the median transaction price of similar products in the market. This price acceptance parameter is compared with preset acceptance thresholds for each interval to determine the corresponding acceptance interval and obtain a price acceptance score for that interval.

[0019] Buyer structure parameters are obtained by statistically analyzing the distribution of domain name suffixes of the target data product callers. These parameters are compared with preset structure thresholds for each interval to determine the current structure interval and obtain the corresponding buyer structure score. Based on the quality score, scarcity score, price acceptance score, and buyer structure score, a comprehensive value score is generated. A value adjustment coefficient is then determined based on the comprehensive value score and a preset value adjustment benchmark.

[0020] In this embodiment, the scoring rule is that the missing rate is calculated based on field-level null values ​​from the data service gateway logs over the past 30 days. Let the missing rate be... Then the missing rate score for: Let the number of competitors be... Then the scarcity score for: Let the ratio of the average transaction price of this data product over the past 30 days to the median market transaction price be . Then the price acceptance score for: Let the percentage of calls be based on specific industries (such as finance and healthcare). Then the buyer's structure score for: The overall value score (SS) is the sum of the scores from the four components: Value coefficient Calculated based on the overall value score and the preset benchmark value: Benchmark price To break even Multiply by the value coefficient: .

[0021] Furthermore, a time-sensitivity decay factor is introduced in the process of determining the value adjustment coefficient. Specifically, the time difference between the generation time of the target data product and the current time is obtained, and the time-sensitivity decay factor is calculated based on a preset half-life parameter. The comprehensive value score is then integrated with the time-sensitivity decay factor, combined with a preset value adjustment benchmark, to calculate the initial value adjustment coefficient. The value adjustment coefficient adjustment records of the target data product over historical periods are obtained, and the maximum allowable adjustment range within the current adjustment period is determined based on these records. The initial value adjustment coefficient is then constrained to the maximum adjustment range to obtain the final value adjustment coefficient. This mechanism ensures that the value adjustment coefficient does not fluctuate too drastically, avoiding significant price volatility.

[0022] The break-even price is adjusted based on a value adjustment factor to obtain the benchmark price of the target data product. Specifically, the break-even price is multiplied by the value adjustment factor to obtain the benchmark price of the target data product.

[0023] S102. Based on the multi-dimensional value assessment parameters of the target data product, a value adjustment coefficient is determined through a preset parameter mapping relationship. The break-even price is adjusted based on the value adjustment coefficient to obtain the benchmark price of the target data product.

[0024] In practice, based on the multidimensional value assessment parameters of the target data product, a value adjustment coefficient is determined through a pre-defined parameter mapping relationship. These multidimensional value assessment parameters include data quality parameters, market scarcity parameters, price acceptance parameters, and buyer structure parameters.

[0025] Specifically, data quality parameters are obtained by calculating the field missing rate of the target data product within a preset statistical period. The field missing rate equals the number of null values ​​divided by the total number of records. The data quality parameters are compared with preset missing rate thresholds for each interval to determine the missing rate interval and obtain the corresponding quality score. For example, if the missing rate is below the first threshold, it falls into the high-quality interval and receives a higher quality score; if the missing rate is above the second threshold, it falls into the low-quality interval and receives a lower quality score; if the missing rate is between the two, it falls into the medium-quality interval and receives a medium quality score.

[0026] The market scarcity parameter is obtained by querying the API of the data exchange to count the number of similar products to the target data product. Similar products refer to data products with the same field structure and update frequency. The market scarcity parameter is compared with preset competitor quantity thresholds for each interval to determine the current competitor quantity interval and obtain the scarcity score corresponding to that interval. The fewer the competitors, the scarcer the data product, and the higher the corresponding scarcity score.

[0027] The price acceptance parameter is obtained by calculating the ratio of the average transaction price of the target data product within a preset statistical period to the median transaction price of similar products in the market. The price acceptance parameter is compared with preset acceptance thresholds for each interval to determine the acceptance interval and obtain the price acceptance score corresponding to that interval. The higher the ratio, the higher the market acceptance of the target data product's current price, and the higher the corresponding price acceptance score.

[0028] The buyer structure parameters are obtained by statistically analyzing the distribution of domain name suffixes among the users of the target data product. The buyer structure parameters are compared with preset structure thresholds for each interval to determine the current structure interval and obtain the corresponding buyer structure score. When users are mainly concentrated in a specific industry, it indicates that the target data product has industry specificity, and the corresponding buyer structure score is higher; when users are more dispersed, it indicates that the target data product has general applicability, and the corresponding buyer structure score is lower.

[0029] A comprehensive value score is generated based on the quality score, scarcity score, price acceptance score, and buyer structure score. Specifically, the four scores are summed to obtain the comprehensive value score. A value adjustment coefficient is then determined based on the comprehensive value score and a preset value adjustment benchmark.

[0030] Furthermore, a time-related decay factor is introduced in the process of determining the value adjustment coefficient. This factor is calculated based on the time difference between the generation time of the target data product and the current time, as well as a preset half-life parameter. It should be noted that the half-life parameter represents the time required for the data value to decay to half of its initial value. Different types of data products can be configured with different half-life parameters, and the time-related decay factor decreases as the data age increases.

[0031] The initial value adjustment coefficient is calculated based on the comprehensive value score, the time-sensitivity decay factor, and the preset value benchmark. Specifically, the comprehensive value score and the time-sensitivity decay factor are combined, and then the preset value benchmark is used to calculate the initial value adjustment coefficient. The value adjustment coefficient adjustment records of the target data product within historical periods are obtained, and the maximum allowable adjustment range within the current adjustment period is determined based on these records. The initial value adjustment coefficient is then constrained to the maximum adjustment range to obtain the final value adjustment coefficient. The break-even price is adjusted based on a value adjustment factor to obtain the benchmark price of the target data product. Specifically, the break-even price is multiplied by the value adjustment factor to obtain the benchmark price of the target data product.

[0032] S103. Obtain the median market price of similar products in the market for the target data product, and calculate the price deviation between the benchmark price and the median market price.

[0033] In practice, the median market price of similar products to the target data product is obtained. This median market price is obtained by querying transaction details from a data exchange. Specifically, the market data API provided by the data exchange is called to retrieve all transaction records of the same category as the target data product within a preset statistical period. After removing the highest and lowest extreme values, the median of the remaining transaction records is calculated as the median market price. The "same category" refers to data products with the same field structure and the same update frequency.

[0034] Calculate the price deviation between the benchmark price and the market median price. Specifically, divide the difference between the benchmark price and the market median price by the market median price to obtain the absolute value of the price deviation. This price deviation value measures the degree of deviation between the benchmark price of the target data product and the current mainstream market transaction price.

[0035] The price deviation is compared with a preset threshold. When the price deviation does not exceed the preset threshold, it indicates that the benchmark price is basically in line with the market situation, and the benchmark price can be directly used as the basis for subsequent pricing. When the price deviation exceeds the preset threshold, it indicates that there is a significant deviation between the benchmark price and the market situation, and the reinforcement learning agent needs to be activated to dynamically adjust the price.

[0036] In this embodiment, when the price deviation exceeds a preset threshold, the system prompts the operator with a one-click follow option to linearly smooth the benchmark price to the market median price. If the operator confirms within a preset time, the market median price is directly used as the pricing basis; if no confirmation is made within the time limit, the system automatically enters the price adjustment channel of the reinforcement learning agent.

[0037] S104. When the price deviation value exceeds a preset threshold, the reinforcement learning agent determines the current price adjustment action from the preset price adjustment action set based on the current operating state parameters of the market, and generates the expected selling price of the target data product.

[0038] In practice, when the price deviation exceeds a preset threshold, a reinforcement learning agent is activated to dynamically adjust the price. This reinforcement learning agent employs a proximal policy optimization algorithm to learn the optimal price adjustment strategy in a continuous market environment.

[0039] The system collects current operational status parameters of the target data product in the market. These parameters include the current price, recent sales volume, median market price, user ratings, and industry popularity. The current price is the actual price currently being executed; recent sales volume is the number of calls made within a preset period; the median market price is the median transaction price of similar products obtained through a data exchange's market data interface; user ratings are obtained by collecting buyer feedback; and industry popularity is obtained by querying an industry index platform. These parameters are used as input to the reinforcement learning agent to represent the current pricing environment.

[0040] The current operating state parameters are input into the reinforcement learning agent, which then calculates the expected revenue for each price adjustment action in a pre-defined set of actions under the current operating state parameters. The pre-defined set of actions includes five discrete options: a 5% price reduction, a 2% price reduction, no price change, a 2% price increase, and a 5% price increase. The expected revenue is calculated based on a revenue mapping relationship pre-learned by the reinforcement learning agent, which reflects the potential revenue changes associated with executing each price adjustment action under different state parameters.

[0041] Specifically, during the training phase of the reinforcement learning agent, price adjustment records of the target data product within historical periods are acquired. Historical operational state parameters and corresponding historical revenue results are extracted from these records before and after each historical price adjustment action. The historical revenue results include changes in revenue and user churn within a preset period after the price adjustment. Training samples are constructed based on the historical operational state parameters, historical price adjustment actions, and historical revenue results to train the reinforcement learning agent, enabling it to learn the revenue mapping relationship between the current operational state parameters and the price adjustment actions. Offline learning is employed during training, utilizing historically accumulated price adjustment experience data to avoid the impact of online training on actual transactions.

[0042] During the inference phase, the current operating state parameters and each price adjustment action are input into the trained reinforcement learning agent. The expected revenue corresponding to each price adjustment action is output through the revenue mapping relationship. The price adjustment action with the highest expected revenue is determined as the current price adjustment action. The benchmark price is adjusted according to the current price adjustment action to generate the expected selling price of the target data product.

[0043] For example, if the current operating status parameters show that the current price is lower than the market median price, sales are stable, and user ratings are high, the reinforcement learning agent may choose to raise the price, setting the expected selling price at a certain percentage above the benchmark price; conversely, if the current price is higher than the market median price and sales are declining, the reinforcement learning agent may choose to lower the price, setting the expected selling price at a certain percentage below the benchmark price.

[0044] S105. The expected selling price is configured in a preset proportion of call traffic for gray-scale testing, and the full-scale effective status of the expected selling price is determined based on the test results.

[0045] In practice, the projected selling price is allocated to a preset proportion of the call traffic for canary testing. The target call traffic for the target data product, representing a preset proportion, is allocated the projected selling price to form a canary group, while the original price is allocated to the remaining call traffic to form a control group.

[0046] Within the preset testing period, the changes in gray-scale revenue and churn for the gray-scale group, as well as the changes in control revenue and churn for the control group, were collected. It should be noted that the preset testing period was set to a complete business cycle to fully observe the impact of price adjustments on user behavior and revenue; revenue changes were calculated by comparing the revenue difference between the testing period and the previous period; churn changes were calculated by comparing the number of users churned between the testing period and the previous period.

[0047] The reward value is calculated based on the difference between the change in gray-scale revenue and the change in control revenue, and the difference between the change in gray-scale churn and the change in control churn. Specifically, the reward value equals the absolute value of the difference in revenue divided by the absolute value of the change in control revenue, minus the preset weighting coefficient multiplied by the difference in churn divided by the absolute value of the change in control churn.

[0048] When the reward value is greater than zero and the churn increase corresponding to the churn change difference is lower than the preset churn threshold, the expected selling price will be used as the price for full implementation. For example, if the revenue of the gray group increases compared to the control group and the churn rate does not increase significantly, the new price is deemed valid and will be implemented for all users.

[0049] When the reward value is not greater than zero or the churn increase is not less than the preset churn threshold, the expected selling price will be rolled back to the original price, and the price adjustment action in this gray-scale test will be marked as a negative sample. For example, if the gray-scale group's revenue decreases or the churn rate increases significantly, the new price will be deemed invalid and rolled back immediately.

[0050] S106. In response to the buyer's prepaid call count amount, the call count is reimbursed via blockchain based on the expected selling price after it takes effect, and buyer satisfaction data is collected after a preset call threshold is triggered, so as to adjust the value adjustment coefficient according to the collection results.

[0051] In practice, in response to the buyer's prepayment for the number of calls, the number of calls is reimbursed via the blockchain based on the expected selling price after the effective date. The smart contract on the blockchain is invoked to calculate the available number of calls corresponding to the prepaid amount, generating a call credential. The call credential includes the identifier of the target data product, the available number of calls, the expected selling price after the effective date, and the validity period.

[0052] Upon receiving a request to access the target data product, the smart contract executes an on-chain atomic decrement operation to reduce the available call count by one, until the available call count reaches zero. Specifically, for each call request processed, the gateway synchronously triggers the smart contract to decrement the available call count recorded in the call credential by one. This operation is tied to the data return logic; if the decrement fails, no data is returned; if the decrement succeeds, data is returned. When the available call count reaches zero, the interface returns an access-forbidden status code.

[0053] During the call count verification process, when the cumulative number of available calls for the target data product reaches the preset call threshold, a satisfaction collection request is sent to the buyer, and the buyer's satisfaction score is received.

[0054] The negative review rate is calculated by counting the number of satisfaction ratings collected within a preset statistical period that are below a preset rating threshold. The preset statistical period can be set to thirty days, and the preset rating threshold can be set to three stars or lower. The negative review rate equals the number of negative reviews divided by the total number of questionnaires. When the negative review rate exceeds the preset negative review threshold, the value adjustment coefficient is reduced by a preset step size. For example, if the negative review rate exceeds 10%, the value adjustment coefficient is reduced by 0.05, causing the benchmark price for the next period to decrease accordingly.

[0055] When the negative review rate is lower than the preset high-quality threshold for several consecutive statistical periods, in response to the seller's upward adjustment instruction, the value adjustment coefficient is increased by a preset step size, and the increased value adjustment coefficient does not exceed the preset upper limit value.

[0056] This application determines the benchmark price by combining cost accounting and multi-dimensional value assessment. This ensures that the pricing of data products is not lower than the actual investment cost, while also reflecting the intrinsic value of data quality and scarcity. Based on this, a reinforcement learning agent is introduced for dynamic price adjustment. When there is a significant deviation between the benchmark price and market conditions, the agent can automatically select an adjustment action based on real-time market conditions, and verify the effect of the price adjustment through gray-scale testing. This mechanism allows pricing to respond quickly to market changes while avoiding the risk of user churn caused by blind price adjustments.

[0057] Furthermore, after the buyer prepays the fee, a call certificate is generated on the blockchain. Each call synchronously verifies the available number of times. During the call process, buyer satisfaction is collected periodically, and negative feedback results are fed back into the value adjustment coefficient, so that subsequent pricing can automatically reflect the real changes in data quality.

[0058] like Figure 2 As shown in the embodiments of this application, a data product pricing device based on reinforcement learning is also proposed, comprising: At least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform a data product pricing method based on reinforcement learning as described in any of the above embodiments.

[0059] This application also provides a non-volatile computer storage medium storing computer-executable instructions, wherein the computer-executable instructions are configured as: a data product pricing method based on reinforcement learning as described in any of the above embodiments.

[0060] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the device and medium embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the description of the method embodiments.

[0061] The devices and media provided in this application are one-to-one with the methods. Therefore, the devices and media also have similar beneficial technical effects as their corresponding methods. Since the beneficial technical effects of the methods have been described in detail above, the beneficial technical effects of the devices and media will not be repeated here.

[0062] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0063] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0064] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0065] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0066] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0067] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0068] Computer-readable media include both permanent and non-permanent, removable and non-removable media that can store information by any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic tape, disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0069] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0070] The above description is merely an embodiment of this application and is not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A data product pricing method based on reinforcement learning, characterized in that, include: Obtain the cumulative total cost of the target data product within a preset historical period, as well as the total number of calls within the preset historical period; and calculate the break-even price of the target product based on the total cost and the total number of calls. Based on the multidimensional value assessment parameters of the target data product, a value adjustment coefficient is determined through a preset parameter mapping relationship. The break-even price is then adjusted based on the value adjustment coefficient to obtain the benchmark price of the target data product. Obtain the median market price of similar products in the market for the target data product, and calculate the price deviation between the benchmark price and the median market price; When the price deviation exceeds a preset threshold, the reinforcement learning agent determines the current price adjustment action from a preset set of price adjustment actions based on the current operating state parameters of the market, and generates the expected selling price of the target data product. The projected selling price is configured in a preset proportion of call traffic for canary testing, and the full effective status of the projected selling price is determined based on the test results; In response to the buyer's prepaid call count payment, the call count is reimbursed via blockchain based on the expected selling price after it takes effect, and buyer satisfaction data is collected after a preset call threshold is triggered, so as to adjust the value adjustment coefficient according to the collection results.

2. The data product pricing method based on reinforcement learning according to claim 1, characterized in that, The multidimensional value assessment parameters include data quality parameters, market scarcity parameters, price acceptance parameters, and buyer structure parameters. The step of determining the value adjustment coefficient based on the multidimensional value assessment parameters of the target data product through a preset parameter mapping relationship specifically includes: The data quality parameters are compared with the preset missing rate thresholds corresponding to each interval to determine the missing rate interval and obtain the quality score value corresponding to the missing rate interval. The market scarcity parameter is compared with the preset competitor quantity threshold corresponding to each interval to determine the competitor quantity interval and obtain the scarcity score value corresponding to the competitor quantity interval. The price acceptance parameter is compared with the preset acceptance threshold corresponding to each interval to determine the acceptance interval and obtain the price acceptance score corresponding to the acceptance interval. The buyer structure parameters are compared with the preset structure thresholds corresponding to each interval to determine the structure interval in which the buyer is located, and the buyer structure score value corresponding to the structure interval is obtained. A comprehensive value score is generated based on the quality score, the scarcity score, the price acceptance score, and the buyer structure score. A value adjustment coefficient is then determined based on the comprehensive value score and a preset value adjustment benchmark.

3. The data product pricing method based on reinforcement learning according to claim 2, characterized in that, The determination of the value adjustment coefficient based on the comprehensive value score and the preset value benchmark specifically includes: Based on the time difference between the generation time of the target data product and the current time, and the preset half-life parameter, calculate the timeliness decay factor of the target data product; Calculate the initial value adjustment coefficient based on the comprehensive value score, the timeliness decay factor, and the preset value benchmark. Obtain the value adjustment coefficient adjustment record of the target data product in the historical period, determine the maximum allowable adjustment range in the current adjustment period based on the adjustment record, and constrain the initial value adjustment coefficient to the range of the maximum adjustment range to obtain the value adjustment coefficient.

4. The data product pricing method based on reinforcement learning according to claim 1, characterized in that, The step of using a reinforcement learning agent to determine the current price adjustment action from a preset set of price adjustment actions based on the current operating state parameters of the market, and generating the expected selling price of the target data product, specifically includes: Collect the current operating status parameters of the target data product in the market; the current operating status parameters include the current price, the sales volume of the most recent period, the median price in the market, user ratings, and industry popularity; The current operating state parameters are input into the reinforcement learning agent, and the reinforcement learning agent calculates the expected revenue of each price adjustment action in the preset price adjustment action set under the current operating state parameters. The price adjustment action with the highest expected return is identified as the current price adjustment action. The benchmark price is then adjusted according to the current price adjustment action to generate the expected selling price of the target data product.

5. A data product pricing method based on reinforcement learning according to claim 4, characterized in that, The step of calculating the expected revenue of each price adjustment action in the preset set of price adjustment actions under the current operating state parameters through the reinforcement learning agent specifically includes: Obtain the price adjustment records of the target data product in the historical period, and extract the historical operating status parameters before and after each historical price adjustment action and the corresponding historical revenue results from the price adjustment records; Training samples are constructed based on the historical operating state parameters, the historical price adjustment actions, and the historical profit results to train the reinforcement learning agent, so that the reinforcement learning agent learns the profit mapping relationship between the current operating state parameters and the price adjustment actions. The current operating state parameters and each price adjustment action are input into the trained reinforcement learning agent, and the expected revenue corresponding to each price adjustment action is output through the revenue mapping relationship.

6. The data product pricing method based on reinforcement learning according to claim 1, characterized in that, The step of configuring the expected selling price in a preset proportion of call traffic for canary testing, and determining the full-scale effectiveness of the expected selling price based on the test results, specifically includes: The target call traffic of the target data product is called in a preset proportion, and the expected selling price is configured in the target call traffic to form a gray group, and the original price is configured in the remaining call traffic to form a control group; Within a preset testing period, the changes in grayscale revenue and grayscale loss of the grayscale group, as well as the changes in control revenue and control loss of the control group, were collected respectively. The reward value is calculated based on the difference between the change in grayscale revenue and the change in control revenue, and the difference between the change in grayscale churn and the change in control churn. When the reward value is greater than zero and the churn increase corresponding to the churn change difference is lower than the preset churn threshold, the expected selling price will be used as the full effective price. When the reward value is not greater than zero or the churn increase is not less than the preset churn threshold, the expected selling price is rolled back to the original price, and the price adjustment action in this gray-scale test is marked as a negative sample.

7. A data product pricing method based on reinforcement learning according to claim 6, characterized in that, The response to the buyer's prepaid call count payment, based on the expected selling price after the transaction takes effect, involves the verification of the call count via blockchain, specifically including: In response to the buyer's prepayment for the number of calls, a smart contract on the blockchain is invoked to calculate the available number of calls corresponding to the prepayment based on the expected selling price after the effective date, and a call voucher is generated; the call voucher includes the identifier of the target data product, the available number of calls, the expected selling price after the effective date, and the validity period; When a call request for the target data product is received, an on-chain atomic decrement operation is performed through the smart contract to reduce the number of available calls by one, until the number of available calls is zero.

8. A data product pricing method based on reinforcement learning according to claim 7, characterized in that, The step of collecting buyer satisfaction data after triggering a preset call threshold, and adjusting the value adjustment coefficient based on the collection results, specifically includes: When the cumulative number of available calls to the target data product reaches a preset call threshold, a satisfaction collection request is sent to the buyer, and the buyer's satisfaction score is received. The number of satisfaction ratings collected within a preset statistical period that are below a preset rating threshold is counted, and the negative rating rate is calculated. When the negative review rate exceeds the preset negative review threshold, the value adjustment coefficient will be reduced by a preset step size. When the negative review rate is lower than the preset quality threshold for multiple consecutive statistical periods, the value adjustment coefficient is increased by a preset step size in response to the seller's upward adjustment instruction.

9. A data product pricing device based on reinforcement learning, characterized in that, include: At least one processor; as well as, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform a data product pricing method based on reinforcement learning as described in any one of claims 1 to 8.

10. A non-volatile computer storage medium storing computer-executable instructions, characterized in that, The computer-executable instructions are configured to execute a data product pricing method based on reinforcement learning as described in any one of claims 1 to 8.

Images