A training data asset detection pricing method and system

Abstract

The application provides a kind of training data asset detection pricing method and system, method includes: obtaining the training data asset to be processed and seller basic information;In the trusted detection environment, training data asset is detected, and data asset detection report is generated;Adopt data asset detection report and seller basic information to generate basic pricing input;Utilize the pricing model to constrain processing basic pricing input, obtain basic pricing result;Buyer demand information is obtained, and scene pricing result is generated based on basic pricing result and buyer demand information;Based on scene pricing result, controlled verification is executed;When controlled verification passes or both sides confirm, adopt scene pricing result, the verification result of controlled verification and the authorization condition contained in seller basic information to generate authorized delivery result, and record the transaction feedback information associated with authorized delivery result. Realize from data quality credible evaluation, privacy security controllable to pricing objective dynamic whole-process systematic optimization.

Description

Technical Field

[0001] This invention relates to the field of training data asset trading technology, specifically to a training data asset detection and pricing method and system. Background Technology

[0002] With the widespread application of artificial intelligence models in natural language processing, computer vision, multimodal understanding, video analytics, robot control, and industry intelligence, training data has evolved from ordinary data resources into a core foundational asset for model training, fine-tuning, evaluation, and scenario deployment. Unlike traditional structured data, the value of training data assets depends not only on data volume but also on data quality, annotation completeness, effective scale, modal pairing relationships, privacy risks, ownership and authorization boundaries, delivery methods, and verifiability. In actual transactions, training data often manifests in multimodal forms such as text, images, image-text pairings, videos, robot trajectories, and VLA (Vision-Language-Action) episodes, characterized by large volume, complex structure, and diverse ownership and privacy boundaries. Traditional transaction methods, primarily based on manual descriptions, static listings, and simple sample displays, are insufficient to meet the needs of credible verification and reasonable pricing for such assets.

[0003] Currently, existing data trading platforms typically offer basic functions such as data product publishing, querying, order placement, settlement, and access management, constituting the closest existing technology to training data asset trading. These platforms rely on textual descriptions, sample examples, or manual explanations provided by sellers to present the data scale, source, and application direction. Buyers complete initial screening and transaction matching through the platform, and the transaction price is mainly determined based on the seller's experienced quotation or manual negotiation between the buyer and seller.

[0004] However, the aforementioned existing technologies still have significant shortcomings in training data asset trading scenarios. First, data quality is difficult to reliably assess before a transaction, and buyers and platforms cannot evaluate data integrity, repetition rate, annotation reliability, multimodal pairing consistency, video readability, or trajectory synchronization without obtaining the complete original data. Second, privacy and security risks are difficult to control effectively. Training data may contain personal information, sensitive image content, or corporate secrets, and directly providing sample previews or original downloads can easily lead to data leaks. Third, the seller's base price lacks objective basis, and pricing relies heavily on experience or manual negotiation, lacking a systematic approach that combines data detection results, supplementary information from the seller, and pricing models. Furthermore, a single static listing price is difficult to adapt to different buyers' task types, budget ranges, authorization requirements, and verification methods, and cannot generate scenario-based dynamic pricing results. For unstructured or multimodal data such as images, videos, image-text pairings, and robot trajectories, traditional systems judge value solely based on the number of files or storage capacity, ignoring key indicators such as image-text pairing integrity, video frame rate and decoding quality, and robot action sequence and instruction matching, which can easily lead to high-quality data being underestimated or low-quality data being overestimated. Summary of the Invention

[0005] To address the aforementioned problems in the existing technology, this invention provides a method and system for pricing training data asset detection.

[0006] The technical problem to be solved by this invention is achieved through the following technical solution: In a first aspect, the present invention provides a method for pricing training data asset detection, comprising: Obtain the training data assets to be processed and the seller's basic information; The training data assets are tested in a trusted testing environment to generate a data asset testing report; Basic pricing inputs are generated using data asset inspection reports and seller basic information; The basic pricing input is constrained using a pricing model to obtain the basic pricing result; the pricing model can be a rule model, a statistical model, or a machine learning model. Obtain buyer demand information and generate scenario-based pricing results based on the basic pricing results and buyer demand information; Controlled verification is performed based on scenario-based pricing results; Once the controlled verification is passed or confirmed by both the buyer and seller, the authorized delivery result is generated using the scenario pricing result, the verification result of the controlled verification, and the authorization conditions contained in the seller's basic information. The transaction feedback information associated with the authorized delivery result is also recorded.

[0007] Secondly, the present invention provides a training data asset detection and pricing system, which includes: an acquisition unit, a detection unit, a basic pricing unit, a constraint processing unit, a scenario pricing unit, a controlled verification unit, and an output unit. The acquisition unit is used to acquire the training data assets to be processed and the seller's basic information; The detection unit is used to detect training data assets in a trusted detection environment and generate a data asset detection report. The basic pricing unit is used to generate basic pricing inputs using data asset inspection reports and seller basic information; The constraint processing unit is used to apply constraints to the basic pricing input using the pricing model to obtain the basic pricing result; wherein the pricing model adopts a rule model, a statistical model, or a machine learning model; The scenario pricing unit is used to obtain buyer demand information and generate scenario pricing results based on the basic pricing results and buyer demand information; A controlled verification unit is used to perform controlled verification based on scenario pricing results; The output unit is used to generate an authorized delivery result by using the scenario pricing result, the verification result of the controlled verification, and the authorization conditions contained in the seller's basic information after the controlled verification is passed or the buyer and seller confirm. It also records the transaction feedback information associated with the authorized delivery result.

[0008] This invention provides a method and system for pricing training data assets. The method includes: acquiring training data assets to be processed and seller basic information; testing the training data assets in a trusted testing environment to generate a data asset testing report; generating basic pricing input using the data asset testing report and seller basic information; constraining the basic pricing input using a pricing model to obtain a basic pricing result; wherein the pricing model employs a rule-based model, a statistical model, or a machine learning model; acquiring buyer demand information and generating a scenario pricing result based on the basic pricing result and buyer demand information; performing controlled verification based on the scenario pricing result; and generating an authorized delivery result using the scenario pricing result, the verification result of the controlled verification, and the authorization conditions contained in the seller basic information, and recording transaction feedback information associated with the authorized delivery result. In this invention, by comprehensively testing the training data assets in a trusted testing environment and generating a data asset testing report, the problem of unreliable data quality assessment before a transaction is solved, allowing buyers and platforms to evaluate multi-dimensional indicators such as data integrity, duplication rate, and annotation reliability without obtaining complete original data. Then, by introducing a controlled verification mechanism, authorized delivery results are generated only after verification is passed or confirmation is made by both the buyer and seller, and transaction feedback information is linked to this mechanism. This solves the problem of effectively controlling privacy and security risks and avoids data leakage caused by directly opening sample previews or downloading the original data. Finally, by combining data asset testing reports and seller basic information to generate basic pricing inputs, and using rule models, statistical models, or machine learning models for constraint processing to obtain basic pricing results, and then generating scenario-based dynamic pricing results based on buyer demand information, this solves the problems of seller basic prices lacking objective basis, single static listing prices being unable to adapt to different buyer needs, and inaccurate value judgments of unstructured or multimodal data. In summary, this invention achieves a systematic optimization of the entire process from reliable data quality assessment and controllable privacy and security to objective and dynamic pricing, overcoming many shortcomings of existing technologies in training data asset transactions.

[0009] The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0010] Figure 1 A flowchart illustrating a training data asset detection pricing method provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of a training data asset detection and pricing system provided in an embodiment of the present invention. Detailed Implementation

[0011] The present invention will be further described in detail below with reference to specific embodiments, but the implementation of the present invention is not limited thereto.

[0012] In order to achieve systematic optimization of the entire process from reliable data quality assessment and controllable privacy and security to objective and dynamic pricing, and to overcome the shortcomings of existing technologies in training data asset trading, this invention provides a training data asset detection and pricing method. Figure 1 This is a flowchart illustrating a training data asset detection and pricing method provided in an embodiment of the present invention, as shown below. Figure 1 As shown, it includes: S101. Obtain the training data assets to be processed and the seller's basic information.

[0013] In this step, the seller can access the training data assets by uploading data packets, mounting data directories, providing access paths, or submitting local detection results. Training data assets include, but are not limited to, text data, image data, image-text pairing data, video data, trajectory data, VLA episode (Vision-Language-Action episode) data, or other multimodal data.

[0014] The system obtains basic information about the seller, which includes at least: the source of the training data assets, ownership / license conditions, cost, delivery method, and whether verification is supported. This information is used in the subsequent basic pricing stage to determine the transaction boundaries of the training data assets and the sufficiency of price evidence.

[0015] In one implementation, when accessing training data assets, the system can automatically infer the type of training data assets based on file structure, annotation files, directory hierarchy, or metadata descriptions, and generate a preliminary data list. If the system cannot automatically identify the asset type, it outputs a prompt for supplementary information, allowing the seller to supplement the data type, annotation method, data source, or delivery boundary. This process enables subsequent trusted detection processes to select the corresponding detection strategy based on the asset type, rather than using a single detection method for all training data assets.

[0016] It should be noted that this invention does not limit the specific data format of the training data assets. The method of this invention can be applied to text data, image data, image-text matching data, video data, trajectory data, robot operation data, or other multimodal training data, as long as they can generate a data asset detection report through a trusted detection environment and be further used for basic pricing and scenario transaction processing.

[0017] S102. Detect the training data assets in a trusted detection environment and generate a data asset detection report.

[0018] Optionally, the trusted testing environment can be a trusted execution environment, a trusted sandbox, a trusted cloud node, a seller-side trusted probe, a data exchange-designated testing node, or a controlled execution environment with isolated access control capabilities.

[0019] Optionally, S102 includes: Identify the asset types of the training data assets; asset types include: text data, image data, image-text pairing data, video data, general trajectory data, VLA episode data, and multimodal combination data; In a trusted detection environment, the training data assets are subjected to corresponding basic structure quality detection based on their asset type to obtain structural quality results. ; In a trusted detection environment, redundancy deduplication is performed on the training data assets according to their asset types to obtain the effective scale result. ; In a trusted detection environment, privacy and security risk detection is performed on the training data assets according to their asset types to obtain the risk detection results. ; In a trusted testing environment, the delivery and verification methods supported by the training data assets are determined based on the asset type, thus obtaining the verifiability assessment result. ; Integrating structural quality results Effective Scale Results Risk detection results and verifiability judgment results Generate a data asset inspection report.

[0020] Optionally, privacy and security risk detection can be performed on the training data assets according to their asset types to obtain the risk detection results. ,include: The training data assets were subjected to personal information risk detection, sensitive personal information risk detection, business sensitivity risk detection, and metadata risk detection respectively, and each sub-risk level was obtained. The risk detection results are determined based on a comprehensive assessment of each sub-risk level. ; Risk detection results This includes handling recommendations for each sub-risk level; among which, the handling recommendations include: allowing continued basic pricing, restricting sample previews, restricting original downloads, restricting formal listing, and providing recommendations for de-identification or manual review.

[0021] In this embodiment, the detection content in S102 includes structure and scale detection, quality detection, redundancy detection, privacy and security risk detection, and verifiability detection.

[0022] Specifically, in S102, after acquiring the training data assets to be detected, the system creates a trusted detection task and configures the detection environment and detection strategy. The trusted detection task can be bound to the detection program version, detection strategy version, and output range to restrict the detection program to only output detection reports, scoring results, risk results, and summary information.

[0023] The system first identifies the types of training data assets and selects the appropriate detection process based on the asset type. For text data, the detection content includes field integrity, text length, label coverage, duplicate text, and text privacy risks; for image data, the detection content includes image readability, resolution, blurriness, duplicate images, and metadata risks; for image-text pairing data, the detection content includes image readability, descriptive text integrity, and image-text pairing integrity; for video data, the detection content includes video decodeability, duration, frame rate, resolution, black frames, and still frames; for VLA episode data, the detection content includes observation, action, state, instruction, and time synchronization.

[0024] In some implementations, for text data, field integrity can be achieved by checking the proportion of missing values, text length can be measured by counting characters or words, tag coverage can be assessed by checking tag distribution and category coverage, duplicate text can be deduplicated using hash or edit distance algorithms, and text privacy risks can be addressed by scanning personal information using regular expressions, named entity recognition, or a pre-built sensitive word library.

[0025] For image data, image readability is determined by decoding success rate, resolution is directly read by width and height pixels, ambiguity is quantified by Laplacian variance or frequency domain analysis, duplicate images are compared by perceptual hashing or feature matching, and metadata risks are identified by extracting sensitive fields such as geographic location and device model from EXIF ​​information.

[0026] For image-text pairing data, image readability also relies on decoding detection, text integrity is described by checking text length, grammatical coherence and keyword coverage, and image-text pairing integrity is calculated by using cross-modal similarity models (such as CLIP) to calculate image-text semantic matching scores.

[0027] For video data, decodeability is confirmed by attempting to open the video stream normally. Duration, frame rate, and resolution are obtained directly from the encoded metadata. Black frames and still frames are determined by the variance of the frame-by-frame luminance histogram or the threshold of motion vector change.

[0028] For VLA episode data, observation can be verified through sensor data format validation and missing frame detection; action can be checked by examining the numerical range and continuity; state can be verified by comparing the validity of the expected state space; instruction can be verified by parsing natural language to confirm its logical consistency with the action sequence; and time synchronization can be verified by comparing whether the timestamp deviations of each modality are within the allowable error range.

[0029] The trusted testing process sequentially yields basic quality results, effective size results, and risk detection results, and further determines whether the training data asset supports sample preview, interface verification, controlled verification, or other delivery methods. Finally, a data asset testing report is generated.

[0030] The data asset inspection report is not a copy of the original training data, nor is it used to directly present the complete sample content to the buyer. Instead, it expresses the objective state of the training data asset that can be verified before the transaction. The inspection report may include data identity results, structural quality results, effective size results, risk detection results, verifiability judgment results, and delivery constraint results. For different asset types, the above results may be generated by different inspection processes, but they are ultimately converted into a unified inspection report for use in subsequent basic pricing steps.

[0031] In privacy and security risk detection, the system can separately assess personal information risk, sensitive personal information risk, business-sensitive risk, and metadata risk. For low-risk data, basic pricing can continue; for data containing location metadata, sensitive fields, or high-risk content, sample previews, original downloads, or official listing can be restricted, and suggestions for de-identification or manual review can be output.

[0032] In one implementation, when the training data asset to be detected is The detection strategy is Trusted testing environment is Then the data asset inspection report It can be represented as: ; In the formula, This represents a trusted detection processing function. Represents training data assets, Indicates the detection strategy, Indicates a trusted testing environment. This refers to a data asset inspection report consisting of structural quality, effective size, risk detection, and verifiability assessment.

[0033] Furthermore, if the structural quality result is denoted as... The results of redundancy and effective size are denoted as The privacy and security risk results are recorded as follows: The verifiability judgment result is denoted as Then the data asset inspection report This can be represented by the following combinational relationship: ; In the formula, Used to characterize the readability, completeness, and basic quality of training data assets. Used to characterize the effective size of the deduplicated data. Used to characterize privacy, security, and compliance risks. Used to characterize whether training data assets support sample preview, interface verification, controlled verification, or other delivery methods.

[0034] S103. Generate basic pricing input using data asset inspection reports and seller basic information.

[0035] The basic pricing input is used to describe the quality, risk, ownership, authorization, cost, and delivery boundaries of the training data asset.

[0036] The system extracts data quality, effective scale, privacy and security risks, verifiability, and delivery constraints from the data asset inspection report, and obtains information such as ownership, authorization, cost, and delivery boundaries from the seller's basic information. These two types of information together constitute the basis for pricing.

[0037] The basis for pricing is not simply reading the quote submitted by the seller, but rather unifying the credible testing results with the seller's commercial boundaries. For the same training data asset, even if the testing quality is high, a high-confidence official listing price should not be directly generated if its source is public data, lacks exclusive authorization, has insufficient cost explanation, or poses high security risks. Conversely, if it is real business data with clear ownership, a well-defined scope of authorization, and supports verifiable delivery, the credibility of the basic pricing result can be improved.

[0038] Assume the seller's basic information is as follows: The data asset inspection report is The basic pricing input is The basic pricing input can then be represented as: ; In the formula, This represents the information fusion function. This indicates a data asset inspection report. Indicates the seller's basic information. This represents the basic pricing input. Through this processing, the system unifies the data ontology detection results with the seller's business boundary into inputs that can be used for pricing model processing.

[0039] In one implementation, the basic pricing input This can be further expressed as: ; In the formula, Indicates ownership and data source information. Indicates the scope of authorization information. Indicates cost information, This indicates the delivery method information. By distinguishing the above information, the system can process the data ontology value, authorization value, cost floor, and delivery service value separately during the pricing stage.

[0040] S104. Use the pricing model to constrain the basic pricing input to obtain the basic pricing result.

[0041] The pricing model can be a rule-based model, a statistical model, or a machine learning model.

[0042] Optionally, S104 includes: Input the basic pricing data into the pricing model to generate a preliminary price; By imposing price constraints on the initial price, the basic pricing result is obtained; The price constraints include at least one of the following: source and exclusivity constraints, authorization boundary constraints, cost constraints, security risk constraints, and market transaction case constraints.

[0043] Pricing models can also be transaction case-tuned models, or combinations of the pricing models mentioned above. The initial price output by the pricing model still needs to undergo price constraint processing to prevent publicly available non-exclusive data from being priced as exclusive high-value assets, data with unclear ownership from being generated with formal listing prices, or high-risk data from being directly made available for raw download. For example, when the training data asset belongs to a public dataset and does not have exclusive authorization, the system restricts it from forming a reference price as a service-oriented data asset; when ownership certificates, cost descriptions, or transaction cases are missing, the system lowers the price confidence level or restricts the generation of formal listing prices; when the privacy and security risks are high, the system restricts sample previews and raw downloads.

[0044] Price constraints may include, but are not limited to, the following: triggering a service pricing cap for publicly available and non-exclusive data assets; generating only a reference price when ownership or authorization materials are insufficient; reducing the confidence level of the minimum price when cost information is missing; restricting formal listing and original downloads when security risks reach a preset level; and lowering the price evidence level when real transaction cases are lacking. These constraints are used to prevent pricing models from outputting excessively high or non-compliant prices based solely on data size or quality scores.

[0045] The basic pricing result can be divided into a reference price and a formal listing suggested price. When the completeness of materials, ownership authorization, and security risks all meet the preset conditions, the system can output a formal listing suggested price; when the above conditions are insufficient but the data is still detectable, the system outputs a reference price and prompts the seller to supplement the missing materials.

[0046] In one implementation, the basic pricing result It can be represented as: ; In the formula, Represents the pricing model. Indicates the basic pricing input. Indicates the parameters of the pricing model. Indicates price constraints. This indicates constraint processing operations. Basic pricing result. It should include at least the basic price range, suggested listing price, minimum price, room for negotiation, and pricing explanation.

[0047] Among them, price constraints This can be further expressed as: ; In the formula, Indicates source and exclusivity constraints. Indicates the authorization boundary constraints, Indicates cost constraints. This indicates security risk constraints. This represents market or transaction case constraints. Using this set of constraints, the system can validate and correct the pricing model output.

[0048] Furthermore, the pricing model of this invention can be obtained or generated in the following ways: First, the detection results of historical training data assets (such as structural quality, effective size, risk level, etc.), corresponding seller basic information (source, cost, authorization conditions, etc.), and actual transaction prices or market reference prices are collected in advance to construct a labeled training sample set. If a rule-based model is used, domain experts formulate explicit pricing formulas and constraint mapping tables based on factors such as data quality grading, effective size weight, exclusivity coefficient, and cost markup. If a statistical model (such as multiple regression or quantile regression) is used, the contribution weights of each input feature to the transaction price are fitted based on the sample set. If a machine learning model (such as gradient boosting trees, neural networks, or transaction case optimization models) is used, supervised training is performed using the sample set, and cross-validation and calibration are performed using historical transaction cases or market benchmarks. Finally, the trained model parameters are solidified into a callable pricing model, and the model output is verified and corrected by price constraints consisting of source and exclusivity, authorization boundaries, cost, security risks, and market transaction cases, thereby obtaining a compliant and reliable basic pricing result.

[0049] S105. Obtain buyer demand information and generate scenario pricing results based on the basic pricing results and buyer demand information.

[0050] Optionally, buyer requirements information includes: mission purpose, budget range, authorization requirements, verification preferences, and delivery preferences.

[0051] Buyers can submit their requirements via natural language or structured forms. The system extracts scenario conditions such as task purpose, budget range, authorization requirements, verification preferences, and delivery preferences from the buyer's requirements. Scenario pricing is not a reassessment of the basic value of the training data asset, but rather a price adjustment based on the buyer's scenario conditions and the base pricing result.

[0052] In one implementation, buyer demand information can be obtained through natural language parsing or structured input. Instead of directly using buyer demand to modify the underlying value of the training data assets themselves, the system uses it to determine the authorization depth, delivery services, verification requirements, and budget constraints under a specific transaction scenario, thereby obtaining a scenario-based pricing result tailored to that buyer's scenario.

[0053] Let the buyer's scenario conditions be... The basic pricing result is The scenario pricing result is The scenario pricing result can then be expressed as: ; In the formula, Represents the scenario pricing function. This indicates the basic pricing result. Indicates the buyer's scenario conditions. This indicates the scenario pricing result. The scenario pricing result can include the scenario price range, the scenario suggested price, validation recommendations, and delivery recommendations.

[0054] If we consider the buyer's scenario conditions Further breakdown into usage conditions Budget conditions Authorization conditions Verification conditions and delivery conditions The scenario pricing result can then be expressed as: ; In the formula, This indicates the buyer's intended use or the model training task. Indicates budget constraints. Indicates the scope of authorization. Indicates verification preference, This indicates delivery preferences. Through this processing, the base price can be transformed into different scenario prices under different transaction scenarios.

[0055] S106. Perform controlled verification based on scenario-based pricing results.

[0056] Controlled verification can include sample previews, API trials, partial verification, sandbox verification, model service verification, or other verification methods that do not directly expose the complete original training data. Through controlled verification, buyers can verify the value of training data assets, and sellers can avoid directly releasing the complete original data before the transaction is completed.

[0057] The results of controlled verification can include statuses such as verification passed, partially passed, failed, or requiring supplementary materials. If the verification results indicate a significant discrepancy between the training data asset and the buyer's needs, the system can output suggestions for requoting, supplementing verification, or terminating the transaction; if the verification results meet expectations, the process proceeds to authorized delivery.

[0058] S107. Once the controlled verification is passed or confirmed by both the buyer and seller, the authorized delivery result is generated using the scenario pricing result, the verification result of the controlled verification, and the authorization conditions contained in the seller's basic information. The transaction feedback information associated with the authorized delivery result is also recorded.

[0059] The system determines the final transaction plan based on the buyer's verification feedback and the negotiation results between the buyer and seller, and generates an authorized delivery result. The authorized delivery result may include access permissions, API call permissions, download credentials, model service permissions, or other authorized delivery credentials. The authorized delivery result is associated with the training data asset version, authorization scope, delivery method, and verification results.

[0060] Let the verification result be The authorization conditions are as follows The scenario pricing result is Then the authorized delivery result It can be represented as: ; In the formula, This indicates the authorized delivery processing function. Indicates the pricing result for the scenario. This indicates the verification result. Indicates the authorization conditions, This indicates the authorized delivery result. The system can also record the transaction price, reason for non-transaction, verification result, scope of authorization, and feedback information, which can be used for subsequent pricing model adjustment and system optimization.

[0061] In addition, the transaction feedback information in this invention includes the transaction price, verification results, scope of authorization, and delivery method.

[0062] Optionally, following S107, it also includes: The transaction feedback information is linked and stored with the basic pricing input, basic pricing result, and scenario pricing result to form linked information; The pricing model is updated using the associated information to obtain the updated pricing model.

[0063] In this invention, transaction feedback information can be used to update the pricing model or the transaction case library. In one embodiment, the system will use transaction feedback information... Compared with the original basic pricing input Basic pricing results and scenario pricing results By establishing a correlation, the pricing of subsequent similar training data assets can be based on historical transaction data, non-transaction data, and validation feedback.

[0064] ; In the formula, This represents the updated pricing model parameters. This indicates transaction feedback information. This indicates model tuning or parameter update processing. This formula is only used to illustrate the relationship between feedback information and the pricing model, and does not limit the specific model training algorithm.

[0065] This invention provides a method for pricing training data assets. First, by comprehensively testing the training data assets in a trusted testing environment and generating a data asset testing report, it solves the problem of unreliable data quality assessment before a transaction, allowing buyers and platforms to evaluate multi-dimensional indicators such as data integrity, duplication rate, and labeling reliability without obtaining the complete original data. Second, by introducing a controlled verification mechanism, authorized delivery results are generated only after verification is passed or confirmation is made by both the buyer and seller, and this is linked to transaction feedback information, solving the problem of difficulty in effectively controlling privacy and security risks and avoiding data leakage caused by directly opening sample previews or downloading the original data. Finally, by combining the data asset testing report and seller basic information to generate basic pricing input, and using rule models, statistical models, or machine learning models for constraint processing to obtain the basic pricing result, a scenario-based dynamic pricing result is generated based on buyer demand information. This solves the problems of lack of objective basis for seller basic prices, the inability of a single static listing price to adapt to different buyer needs, and inaccurate value judgment of unstructured or multimodal data. In summary, this invention achieves a systematic optimization of the entire process from reliable data quality assessment and controllable privacy and security to objective and dynamic pricing, overcoming many shortcomings of existing technologies in training data asset transactions.

[0066] The method provided in this embodiment of the invention can be applied to electronic devices. Specifically, the electronic device can be a desktop computer, a portable computer, a smart mobile terminal, a server, etc., and this embodiment of the invention does not limit the application to such devices.

[0067] Based on the same inventive concept, embodiments of the present invention also provide a training data asset detection and pricing system. Figure 2 This is a schematic diagram of the structure of a training data asset detection and pricing system provided in an embodiment of the present invention, as shown below. Figure 2 As shown, it includes: an acquisition unit 201, a detection unit 202, a basic pricing unit 203, a constraint processing unit 204, a scenario pricing unit 205, a controlled verification unit 206, and an output unit 207. Acquisition unit 201 is used to acquire the training data assets to be processed and the seller's basic information; Detection unit 202 is used to detect training data assets in a trusted detection environment and generate a data asset detection report; Basic pricing unit 203 is used to generate basic pricing inputs using data asset inspection reports and seller basic information; The constraint processing unit 204 is used to process the basic pricing input using the pricing model to obtain the basic pricing result; wherein the pricing model adopts a rule model, a statistical model or a machine learning model. The scenario pricing unit 205 is used to obtain buyer demand information and generate scenario pricing results based on the basic pricing results and buyer demand information; Controlled verification unit 206 is used to perform controlled verification based on scenario pricing results; Output unit 207 is used to generate an authorized delivery result by using the scenario pricing result, the verification result of the controlled verification, and the authorization conditions contained in the seller's basic information after the controlled verification is passed or the buyer and seller confirm it, and to record the transaction feedback information associated with the authorized delivery result.

[0068] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Furthermore, those skilled in the art can combine and integrate the different embodiments or examples described in this specification.

[0069] Although the invention has been described herein in conjunction with various embodiments, those skilled in the art, by reviewing the accompanying drawings and the disclosure, will understand and implement other variations of the disclosed embodiments in carrying out the claimed invention. In this description, the word "comprising" does not exclude other components or steps, "a" or "an" does not exclude a plurality, and "a plurality" means two or more, unless otherwise explicitly specified. Furthermore, while different embodiments may describe certain measures, this does not mean that these measures cannot be combined to produce good results.

[0070] The above description, in conjunction with specific preferred embodiments, provides a further detailed explanation of the present invention. It should not be construed that the specific implementation of the present invention is limited to these descriptions. For those skilled in the art, various simple deductions or substitutions can be made without departing from the inventive concept, and all such modifications and substitutions should be considered within the scope of protection of the present invention.

Claims

1. A training data asset detection pricing method, characterized in that, include: Obtain the training data assets to be processed and the seller's basic information; The training data assets are tested in a trusted testing environment to generate a data asset testing report; The basic pricing input is generated using the data asset inspection report and the seller's basic information; The basic pricing input is constrained using a pricing model to obtain the basic pricing result; wherein the pricing model is a rule model, a statistical model, or a machine learning model. Obtain buyer demand information, and generate scenario pricing results based on the basic pricing results and the buyer demand information; Perform controlled verification based on the pricing results of the described scenario; Once the controlled verification is passed or confirmed by both the buyer and seller, an authorized delivery result is generated using the scenario pricing result, the verification result of the controlled verification, and the authorization conditions contained in the seller's basic information, and transaction feedback information associated with the authorized delivery result is recorded.

2. The training data asset detection pricing method of claim 1, wherein, The seller's basic information includes at least: the source, ownership / authorization conditions, cost, delivery method, and whether verification is supported for the training data assets.

3. The training data asset detection pricing method of claim 1, wherein, The trusted detection environment can be a trusted execution environment, a trusted sandbox, a trusted cloud node, a seller-side trusted probe, a data exchange-designated detection node, or a controlled execution environment with isolated access control capabilities.

4. The method for pricing training data assets according to claim 1, characterized in that, The step of detecting the training data assets in a trusted detection environment and generating a data asset detection report includes: Identify the asset types of the training data assets; the asset types include: text data, image data, image-text pairing data, video data, general trajectory data, VLA episode data, and multimodal combination data; In the trusted detection environment, the training data assets are subjected to corresponding infrastructure quality detection according to the asset type to obtain the structural quality results. ; In the trusted detection environment, redundancy deduplication calculations are performed on the training data assets according to the asset type to obtain the effective scale result. ; In the trusted detection environment, privacy and security risk detection is performed on the training data assets according to the asset type to obtain the risk detection results. ; In the trusted detection environment, the delivery verification method supported by the training data asset is determined based on the asset type, thereby obtaining a verifiability judgment result. ; Integrating the structural quality results The effective scale results The risk detection results and the verifiability judgment result Generate the data asset detection report.

5. The method for pricing training data assets according to claim 4, characterized in that, The privacy and security risk detection is performed on the training data asset according to the asset type to obtain the risk detection result. ,include: The training data assets are subjected to personal information risk detection, sensitive personal information risk detection, business sensitivity risk detection, and metadata risk detection respectively to obtain each sub-risk level; The risk detection result is determined based on a comprehensive assessment of each of the sub-risk levels. ; The risk detection results This includes handling recommendations corresponding to each sub-risk level; among which, the handling recommendations include: allowing continued basic pricing, restricting sample previews, restricting original downloads, restricting formal listing, and providing de-identification processing recommendations or providing manual review recommendations.

6. The method for pricing training data assets according to claim 1, characterized in that, The process of using a pricing model to constrain the basic pricing input to obtain the basic pricing result includes: The basic pricing input is fed into the pricing model to generate a preliminary price; By applying price constraints to the preliminary price, the basic pricing result is obtained. The price constraints include at least one of the following: source and exclusivity constraints, authorization boundary constraints, cost constraints, security risk constraints, and market transaction case constraints.

7. The method for pricing training data assets according to claim 1, characterized in that, The buyer's requirements information includes: task purpose, budget range, authorization requirements, verification preferences, and delivery preferences.

8. The method for pricing training data assets according to claim 1, characterized in that, The controlled verification includes at least one of sample preview verification, interface verification, sandbox verification, or model service verification.

9. The method for pricing training data assets according to claim 1, characterized in that, Also includes: The transaction feedback information is associated and stored with the basic pricing input, the basic pricing result, and the scenario pricing result to form associated information; The pricing model is updated using the associated information to obtain an updated pricing model.

10. A training data asset detection and pricing system, characterized in that, The training data asset detection and pricing system includes: an acquisition unit, a detection unit, a basic pricing unit, a constraint processing unit, a scenario pricing unit, a controlled verification unit, and an output unit; The acquisition unit is used to acquire the training data assets to be processed and the seller's basic information; The detection unit is used to detect the training data assets in a trusted detection environment and generate a data asset detection report. The basic pricing unit is used to generate basic pricing input using the data asset detection report and the seller's basic information; The constraint processing unit is used to apply constraints to the basic pricing input using a pricing model to obtain a basic pricing result; wherein the pricing model is a rule model, a statistical model, or a machine learning model. The scenario pricing unit is used to obtain buyer demand information and generate a scenario pricing result based on the basic pricing result and the buyer demand information; The controlled verification unit is used to perform controlled verification based on the scenario pricing result; The output unit is used to generate an authorized delivery result by using the scenario pricing result, the verification result of the controlled verification, and the authorization conditions contained in the seller's basic information after the controlled verification is passed or the buyer and seller confirm it, and to record the transaction feedback information associated with the authorized delivery result.

Images