Data transaction method for driving multi-party benefit balance based on data value

Abstract

The invention discloses a data transaction method for driving multi-party benefit balance based on data value, and the method is characterized in that the method comprises the following steps: S1, a data owner submits data to an intermediate platform for consignment sale, the intermediate platform carries out the evaluation of the consignment sale data, obtains the data value, and publishes the data value in a system; the user purchases data most conforming to own benefits according to the data value; s2, a Stackelberg game economic framework is established, and the Stackelberg game is divided into two stages: in the first stage, an intermediate platform is used as a leader, and a data owner is used as a follower; in the second stage, the data owner serves as a leader, and the user serves as a follower; and S3, a Nash equilibrium point participating in the Stackelberg game of the three parties is sought. The problem that in the prior art, a transaction dominant party maliciously improves profit by utilizing the dominant position of the transaction dominant party, and pricing cannot be carried out according to the actual value of data is solved.

Description

Technical Field

[0001] The present invention relates to a data transaction method based on data value-driven multi-party interest balance, belonging to the technical field of data transaction. Background Art

[0002] With the popularity of smart devices, a large amount of data is generated by all kinds of users every day. In the information age, these "big data" have shown unprecedented value. Considering these economic values, "big data" has now become part of market transactions. Data trading refers to the behavior of enterprises or individuals selling or exchanging their data assets to third parties in exchange for economic benefits or other value returns. In recent years, with the development of data trading, a large amount of data assets are "exchanged" in the data market every year.

[0003] Considering the huge commercial value of the data market, many companies have begun to build data markets and are committed to providing data to data demanders to make profits. For example, there are data trading platforms such as "DataStreamX" abroad and the Guiyang Big Data Trading Center in China. However, a large number of data trading issues have also led to many problems, such as:

[0004] It is easy for the party with the advantage in the transaction to maliciously increase profits by taking advantage of its own advantageous position. In data transactions, the transactions between data owners, intermediary platforms and users are based on the free pricing of buyers, the free setting of rates by intermediary platforms and the free choice of buyers. However, in this transaction mechanism, intermediary platforms may maliciously increase rates by taking advantage of their monopoly position, resulting in damage to the interests of data owners and users.

[0005] Pricing cannot be based on the actual value of the data. In the current mainstream data trading platforms, data users cannot perceive the actual value of the data before purchasing the data. Many malicious sellers take advantage of this feature of the data platform to sell low-quality data that does not meet user expectations, causing huge financial losses to users. Summary of the invention

[0006] The technical problem to be solved by the present invention is to provide a data trading method based on data value-driven multi-party interest balance to overcome the shortcomings of the prior art.

[0007] The technical solution of the present invention is: a data trading method based on data value-driven multi-party interest balance, comprising the following steps:

[0008] S1. The data owner submits the data to the intermediary platform for consignment. The intermediary platform evaluates the consignment data to obtain the data value and publishes the data value in the system. Users purchase the data that best suits their own interests based on the data value.

[0009] S2. A Stackelberg game economic framework was established. The Stackelberg game is divided into three stages: In the first stage, the intermediate platform, as the leader, determines the service fee charged to the data owner; in the second stage, the data owner, as the follower of the intermediate platform and the leader of the user, determines the data sales value; in the third stage, the user, as the follower, purchases the data that best suits his own interests by referring to the data value;

[0010] S3. Seek the Nash equilibrium point of the Stackelberg game involving three parties.

[0011] Furthermore, the method by which the intermediate platform evaluates the consignment data to obtain the data value is:

[0012]

[0013] Among them, u k Indicates the value of data, represents the data size of data set k, θ k represents the accuracy of data set k, σ k represents the data availability of dataset k, Represents the relationship between data size and data value; Represents the relationship between data accuracy and data value; represents the relationship between data availability and data value; ε, γ and ρ are weight coefficients, is the bias term;

[0014]

[0015] λ1, λ2, λ3, ω1, ω2, ω3, ν1, ν2 and ν3 are fitting parameters.

[0016] Furthermore, the method for finding the optimal λ1, λ2 and λ3 is as follows:

[0017] The function of the relationship between data value and data size is set as:

[0018]

[0019] In order to find the optimal λ1, λ2 and λ3, under the same data accuracy and data availability, a set of training sets with different sizes and the same test set are set. The data sizes of the training sets are {(φ1),...(φ n ),...,(φ N-1 ),(φ N)}; the training sets are input into the machine learning model for training; after training, the same test set is input into the machine learning model to obtain the effect of different data sizes on the accuracy of the same test set; the model accuracy corresponding to these training sets is

[0020]

[0021] Enter the data size as a parameter to the function The data values ​​obtained based on the linear regression method are By using the gradient descent method, the optimal λ1, λ2 and λ3 are obtained, and the λ1, λ2 and λ3 with the smallest minimum mean square error formula are used as the optimal fitting parameters to satisfy the minimum mean square error formula.

[0022] The method to find the optimal ω1, ω2 and ω3 is as follows:

[0023] The function of the relationship between data value and data accuracy is set as:

[0024] In order to find the optimal ω1, ω2 and ω3, under the same data size and data availability, a set of training sets with different data accuracy and the same test set are set. The data accuracy of the training sets is {(θ1),...(θ h ), ...., (θ H-1 ), (θ H )}; input the training sets into the machine learning model for training; after the training is completed, input the same test set into the machine learning model to obtain the impact of different data accuracy on the accuracy of the same test set; the actual data values ​​corresponding to this set of training sets are Enter data accuracy as a parameter to the function The data values ​​obtained based on the linear regression method are Through the gradient descent method, we can find the optimal ω1,

[0025] ω2 and ω3, so that Established;

[0026] The method to find the optimal ν1, ν2 and ν3 is as follows:

[0027] The function of the relationship between data value and data availability is set as:

[0028]

[0029] In order to find the optimal ν1, v2 and ν3, under the same data size and data accuracy, a set of training sets with different data availability and the same test set are set. The data availability of the training sets is {(σ1),...(σ t ), ...., (σ T-1 ), (σ T )}; input the training set into the machine learning model for training; after the training is completed, input the test set into the machine learning model to obtain the impact of different data availability on the actual accuracy of the same test set. The actual data values ​​of this group of training sets with different availability are Input data availability as a parameter to the function The data values ​​obtained based on the linear regression method are By gradient descent method, we search for the optimal v1, ν2 and ν3 so that Established;

[0030] Find the optimal ε,γ,ρ and The method is as follows:

[0031] Set up multiple sets of data sets with different data sizes, different data accuracy, and different data availability in proportion, and input them into the machine learning model for training. After the training is completed, input the test set into the machine learning model to obtain the actual accuracy of the same test set for different factors;

[0032] Based on the optimal fitting parameters λ1, λ2, λ3, ω1, ω2, ω3, v1, v2 and v3, data sets with different data sizes, different data accuracy and different data availability are respectively input into the machine learning model as training sets. The accuracy of each training set corresponding to the same test set is {(D1), ... (D i ),....,(D M-1 ), (D M )},(D i ) represents the accuracy corresponding to the i-th training set, and the gradient descent method is used to find the most suitable ε, γ, ρ and Make the data value and usage function obtained by the test set The error between the calculated results is minimal.

[0033] Furthermore, the data set in the method for obtaining ω1, ω2 and ω3 is realized by replacing the labels of the pictures in the original database;

[0034] The data sets in the method for obtaining v1, v2 and ν3 are realized by replacing the pictures in the original database.

[0035] Furthermore, the profit function of the intermediate platform of the Stackelberg game economic framework is:

[0036]

[0037] Among them, 0 <r i <1, the profit function consists of two parts, the first part [(p ik *r i +wc j )*x ijk ] is charged to the data owner for consignment, and the second part l(1+u k ) is the cost of storing dataset k on the platform, where p ik is the price at which data owner i sells dataset k, l is the cost coefficient of platform storage and operation, and u k is the data value of data set k, r i is the rate that the platform charges the data owner i when selling the dataset, w is the fixed fee for consignment, and x ijk is the deterministic function for buyer j to purchase data k from data owner i, c j is the transmission cost of the platform sending data to user j, and J represents the total number of users in the data trading market.

[0038] Furthermore, the profit function of the data owner in the Stackelberg game economic framework is:

[0039]

[0040] Where 0≤u k ≤1, the data owner's profit function cost consists of three parts. The first part p ik *u k -r i *p ik -w is the reward for selling data, where p ik is the price that data owner i sells dataset k, u k is the data value of dataset k, w is the fixed rate x paid to the data seller ijk is the determination function for user j to purchase data from data owner i. When j decides to purchase data k from i, x ijk =1, otherwise x ijk =0; the second part c i The transmission cost of data owner i to transfer data to the intermediate platform for consignment, the third part e i *u k The cost of collecting dataset k for data owner i, e i is the correlation coefficient of cost.

[0041] Furthermore, the user profit function of the Stackelberg game economic framework is:

[0042] U j =[δ 1j +δ 2j ln(1+δ 3j *u k )]-u k *p ik

[0043] Among them, δ 1j >0,δ 3j > 0, the user’s profit function is divided into two parts, the first part δ 1j +δ 2j ln(1+δ 3j *u k )] is the revenue obtained by users from purchasing data, δ 1j , δ 2j and δ 3j is the profit coefficient of user j, which is positively correlated with the actual profitability of user j; the second part u k *p ik The cost of purchasing dataset k for the user.

[0044] Furthermore, the Stackelberg game economic framework also includes the following constraints:

[0045] S≥0

[0046] T i ≥0

[0047] U j ≥0.

[0049] Furthermore, the Nash equilibrium point of the Stackelberg game among the three parties in step S4 is:

[0050] The user's decision is: Let the value of data k

[0051] The data owner makes the following decision: let data owner i sell data set k at a price

[0052] The platform decision is: the rate at which the platform charges data owner i

[0053] The beneficial effects of the present invention are: compared with the prior art,

[0054] 1) The present invention divides the Stackelberg game into three stages, includes all three parties involved in the data transaction in the data transaction, and then finds the point where all three parties can obtain the maximum profit by seeking the Nash equilibrium point of the Stackelberg game. On the one hand, it maximizes the profits of all transaction participants and motivates all transaction participants. On the other hand, since all transaction parties are priced according to the Nash equilibrium point, it prevents the transaction parties from taking advantage of their dominant position in the transaction to maliciously increase their profits.

[0055] 2) The present invention fully considers the measurement of data value, considers the actual value of data in the Stackelberg game, and sets the profit function based on the actual value of data, thereby obtaining the Nash equilibrium point that takes into account the value of data, thereby maximizing the profits of all parties involved in the transaction based on the value of data and effectively protecting the interests of all parties;

[0056] 3) The present invention adopts machine learning methods to fit the three dimensions of data size, data accuracy and data availability respectively, and then uses machine learning methods to fit the data sets of the three dimensions again, thereby obtaining a fitting function with an average error of less than 4%, and the prediction accuracy is higher;

[0057] 4) This patent is based on the "consignment" transaction method. The data owner entrusts the data to the platform for sale, and no longer needs to waste a lot of time processing the sale and delivery of goods. At the same time, in this patent, the platform will first conduct a quality assessment on the data consigned by the data owner, and make the assessed data value level public, which guarantees the user's right to know the value of the data before purchasing the data. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] Figure 1 The data transaction framework of the present invention;

[0059] Figure 2 This is a comparison chart of the public data value function and machine learning accuracy under different data sizes;

[0060] Figure 3 This is a comparison chart of the disclosed data value function and machine learning accuracy under different data accuracy;

[0061] Figure 4 This is a comparison chart of the public data value function and machine learning accuracy under different data availability;

[0062] Figure 5 A comparison chart of the data value function disclosed in the present invention and the machine learning accuracy;

[0063] Figure 6 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0064] In order to make the objectives, technical solutions and advantages of the present invention more clear, the present invention will be further described in detail below with reference to the accompanying drawings of this specification.

[0065] Embodiment 1: As attached Figure 1 and Figure 6 As shown, a data transaction method based on data value-driven multi-party interest balance is provided, including the following steps:

[0066] S1. The data owner submits the data to the intermediary platform for consignment. The intermediary platform evaluates the consignment data to obtain the data value and publishes the data value in the system. Users purchase the data that best suits their own interests based on the data value.

[0067] S2. A Stackelberg game economic framework was established. The Stackelberg game is divided into three stages: In the first stage, the intermediate platform, as the leader, determines the service fee charged to the data owner; in the second stage, the data owner, as the follower of the intermediate platform and the leader of the user, determines the data sales value; in the third stage, the user, as the follower, purchases the data that best suits his own interests by referring to the data value;

[0068] S3. Seek the Nash equilibrium point of the Stackelberg game involving three parties.

[0069] In actual transactions, the value of data is difficult to measure directly through the dimensions of the data set (data size, accuracy, availability, etc.). With the outstanding performance of machine learning technology in the field of data analysis, this article provides a new way of thinking for evaluating the value of data. In the field of machine learning, the value of the data set is a key factor in determining the accuracy of the model. Specifically, the performance of the model on the test set (i.e., the accuracy of the model) is directly related to the value of its training data set, and the model accuracy often maps out the actual value of the data set. Therefore, this article believes that the accuracy of the machine learning model represents the actual value of the data.

[0070] However, in the transaction process, training the data set of each transaction to obtain the performance of the trained model takes a lot of time and computing power, and the evaluation of data value is not easy to implement without explanation. Therefore, this paper designs a data value evaluation function based on the linear regression method to approximate the prediction accuracy of the machine learning model.

[0071] Specifically, the method for the intermediate platform to evaluate the consignment data and obtain the data value is to evaluate it through the following function for evaluating the data value:

[0072]

[0073] Among them, u kIndicates the value of data, represents the data size of data set k, θ k represents the accuracy of data set k, σ k represents the data availability of dataset k, Represents the relationship between data size and data value; Represents the relationship between data accuracy and data value; represents the relationship between data availability and data value; ε, γ and ρ are weight coefficients, is the bias term.

[0074]

[0075]

[0076] λ1, λ2, λ3, ω1, ω2, ω3, v1, v2 and v3 are fitting parameters.

[0077] Specifically, the method for finding the optimal λ1, λ2 and λ3 is as follows:

[0078] The function of the relationship between data value and data size is set as:

[0079]

[0080] In order to find the optimal λ1, λ2 and λ3, under the same data accuracy and data availability, a set of training sets with different sizes and the same test set are set. The data sizes of the training sets are {(φ1),...(φ n ),...,(φ N-1 ),(φ N )}; input the training set into the machine learning model for training; after the training is completed, input the test set into the machine learning model to obtain the impact of different data sizes on the data value. The actual values ​​corresponding to the data sets are These accuracies are considered to be the actual data values ​​of the data;

[0081] Enter the data size as a parameter to the function The data values ​​obtained based on the linear regression method are By using the gradient descent method, the optimal λ1, λ2 and λ3 are obtained, and the λ1, λ2 and λ3 with the smallest minimum mean square error formula are used as the optimal fitting parameters to satisfy the minimum mean square error formula.

[0082] The method to find the optimal ω1, ω2 and ω3 is as follows:

[0083] The function of the relationship between data value and data accuracy is set as:

[0084] In order to find the optimal ω1, ω2 and ω3, under the same data size and data availability, a set of training sets with different data accuracy and the same test set are set. The data accuracy of the training sets is {(θ1),...(θ h ), ...., (θ H-1 ), (θ H )}; input the training set into the machine learning model for training; after the training is completed, input the test set into the machine learning model to obtain the impact of different data accuracy on the accuracy of the same test set; the actual data values ​​corresponding to this set of training sets are Enter data accuracy as a parameter to the function The data values ​​obtained based on the linear regression method are By using the gradient descent method, we can find the optimal ω1, ω2 and ω3 so that

[0085] The method to find the optimal v1, ν2 and ν3 is as follows:

[0086] The function of the relationship between data value and data availability is set as:

[0087]

[0088] In order to find the optimal ν1, ν2 and ν3, under the same data size and data accuracy, a set of training sets with different data availability and the same test set are set. The data availability of the training sets is {(σ1),...(σ t ), ...., (σ T-1 ), (σ T )}; input the training set into the machine learning model for training; after the training is completed, input the test set into the machine learning model to obtain the impact of different data availability on the actual accuracy of the same test set. The actual data values ​​of this group of training sets with different availability are Input data availability as a parameter to the function The data values ​​obtained based on the linear regression method are By gradient descent method, we search for the optimal ν1, ν2 and ν3 so that Established;

[0089] Find the optimal ε,γ,ρ and The method is as follows:

[0090] Set up multiple sets of data sets with different data sizes, different data accuracy, and different data availability in proportion, and input them into the machine learning model for training. After the training is completed, input the test set into the machine learning model to obtain the actual accuracy of the same test set for different factors;

[0091] Based on the optimal fitting parameters λ1, λ2, λ3, ω1, ω2, ω3, ν1, ν2 and ν3, data sets with different data sizes, different data accuracy and different data availability are respectively input into the machine learning model as training sets. The accuracy of each training set corresponding to the same test set is {(D1), ... (D i ),....,(D M-1 ), (D M )},(D i ) represents the accuracy corresponding to the i-th training set, and the gradient descent method is used to find the most suitable ε, γ, ρ and Make the data value and usage function obtained by the test set The error between the calculated results is minimal.

[0092] Specifically, the data set in the method for obtaining ω1, ω2 and ω3 is realized by replacing the labels of the pictures in the original database;

[0093] The data sets in the method for obtaining ν1, v2 and v3 are realized by replacing the pictures in the original database.

[0094] Specifically, the profit function of the intermediate platform of the Stackelberg game economic framework is:

[0095]

[0096] Among them, 0 <r i <1, the profit function consists of two parts, the first part [(p ik *r i +wc j )*x ijk ] is charged to the data owner for consignment, and the second part l(1+u k ) is the cost of storing dataset k on the platform, where p ik is the price at which data owner i sells dataset k, l is the cost coefficient of platform storage and operation, and u k is the data value of data set k, r i is the rate that the platform charges the data owner i when selling the dataset, w is the fixed fee for consignment, and x ijk is the deterministic function for buyer j to purchase data k from data owner i, c j is the transmission cost of the platform sending data to user j, and J represents the total number of users in the data trading market.

[0097] Specifically, the profit function of the data owner in the Stackelberg game economic framework is:

[0098]

[0099] Where 0≤u k ≤1, the data owner's profit function cost consists of three parts. The first part p ik *u k -r ik *p ik -w is the reward for selling data, where p ik is the price at which data owner i sells dataset k, u k is the data value of data set k, r ik is the rate that the platform charges data owner i for selling dataset k, and w is the fixed rate x for the data seller. ijk is the determination function for user j to purchase data from data owner i. When j decides to purchase data from i, x ijk =1, otherwise x ijk =0; the second part c i The cost of data owner i transferring data to the intermediary platform for consignment, the third part e i *u k The cost of collecting dataset k for data owner i, e i is the correlation coefficient of cost.

[0100] Specifically, the user profit function of the Stackelberg game economic framework is:

[0101] U j =[δ 1j +δ 2j ln(1+δ 3j *u k )]-u k *p ik

[0102] Among them, δ 1j >0,δ 3j > 0, the user’s profit function is divided into two parts, the first part δ 1j +δ 2j ln(1+δ 3j *u k )] is the benefit brought to users by unit data value, δ 1j , δ 2j and δ 3j is the profit coefficient of user j, which is positively correlated with the actual profitability of user j; the second part u k *pik The cost of purchasing dataset k for the user, u k is the data value of data set k, p ik is the price that data owner i sells dataset k.

[0103] Specifically, the Stackelberg game economic framework also includes the following constraints:

[0104] S≥0

[0105] T i ≥0

[0106] U j ≥0.

[0108] Specifically, the Nash equilibrium point of the Stackelberg game among the three parties in step S4 is:

[0109] The user's decision is: Let the value of data k

[0110] The data owner makes the following decision: let data owner i sell data set k at a price

[0111] The platform decision is: the rate at which the platform charges data owner i

[0112] The derivation process of Nash equilibrium point is as follows:

[0113] The profits of the intermediary platform, data owner, and user are formalized, and then the backward induction method is used to mathematically derive the profit function of the three parties after quantification. The results show that there is a unique equilibrium point in the Stackelberg game of the three parties. At this equilibrium point, all three parties can obtain the maximum profit. The details are as follows:

[0114] The profit of the intermediate platform is formalized. For the sake of convenience, this paper only studies the optimal decision of the intermediate platform for data owner i. This method can be extended to all data owners in the system. Therefore, its profit function is shown in formula (3-5):

[0115]

[0116] Where 0<r i <1

[0117] The profit function consists of two parts: the first part [(p ik *r ik +wc j )*x ijk ] is the seller's fee for consignment, the second part l(1+u k) is the cost of storing dataset k on the platform, where p ik is the price at which data owner i sells data set k, l is the storage cost coefficient, and u k is the data value of data set k, r ik is the rate that the platform charges data owner i when selling dataset k, w is the fixed fee for consignment, and x ijk is the deterministic function for buyer j to purchase data from data owner i, c j is the transmission cost of the platform sending data j to the user, 0 represents the platform, and J represents the number of users (the number of users who purchased data k).

[0118] Similarly, the profits of data owners and users are formalized, and the formalized profit formulas are shown in Equations (3-6) and (3-7) respectively.

[0119]

[0120] U j =[δ 1j +δ 2j ln(1+δ 3j *u k )]-u k *p ik (3-7)

[0121] Where 0≤u k ≤1, the data owner's profit function cost consists of three parts. The first part [(p ik *r ik +wc j )*x ijk ] charges the data owner a rate for consignment, where p ik is the price at which data owner i sells dataset k, u k is the data value of data set k, r ik is the rate that the platform charges data owner i when selling dataset k, x ijk is the determination function for user j to purchase data from data owner i. When j decides to purchase data from i, x ijk =1, otherwise x ijk =0; the second part c i0 The cost of data owner i transferring data to the intermediary platform for consignment, the third part e i *u k The cost of collecting dataset k for data owner i, e i is the correlation coefficient of cost.

[0122] The user's profit function is divided into two parts. The first part δ 1j +δ 2jln(1+δ3j*u k )] is the benefit brought to users by unit data value, δ 1j , δ 2j and δ 3j is the profit coefficient of user j, which is positively correlated with the actual profitability of user j; the second part u k *p ik The cost of purchasing dataset k for the user, u k is the data value of data set k, p ik is the price at which data owner i sells data set k; δ 1j >0,δ 3j >0.

[0123] Considering that all participants in data transactions are rational, the decisions of all parties should enable them to obtain positive benefits, that is:

[0124] S>=0

[0125] T i >=0

[0126] U j >=0

[0127] In the first stage, the intermediate platform acts as a leader and determines the consignment rate to maximize its profit. Its strategy can be expressed as:

[0128]

[0129] As a follower of the intermediate platform and a leader of users, the data owner decides the selling price of the data to maximize his own profit. His strategy can be expressed as

[0130]

[0131] In the second stage, all users form a Stackelberg game follower. After observing the decision of the leader, the data owner, they decide the value of the data to purchase to maximize their own profits:

[0132]

[0133] From the backward induction method, we can know that to study the equilibrium point of the Stackelberg game, we need to first consider the followers, that is, to first explore the optimal strategy of the user.

[0134] Therefore, this paper first considers the user's decision, which determines the optimal data value u to purchase k * By taking the second-order derivative of formula (3-7), we can get

[0135]

[0136] From formula (3-9), we can see that formula (3-7) is a strict convex function. Therefore, when When k * is the optimal strategy for each user. Let Available

[0137]

[0138] That is, at a given unit data price p ik and the resale rate r i , the optimal strategy for each seller is

[0139] Based on the optimal strategy of the third stage of the Stackelberg game, we can derive the optimal strategy of the data owner in the second stage. First, substitute equation (3-10) into equation (3-6) to obtain

[0140]

[0141] For the convenience of research, the second-order derivative of formula (3-11) is obtained. Easy to know T j is a strictly monotone convex function, let And assume that the number of users who purchase data from data owner j in the system is J, that is

[0142] Available

[0143]

[0144] Considering When , the data seller's profit is less than 0, which is obviously not true.

[0145] Therefore, the unit value data pricing of the data seller can be When , they can obtain the maximum profit. Combining (3-10) and (3-12), we can get:

[0146]

[0147] Based on the optimal decision of the second stage, the optimal decision of the intermediate platform can be obtained. First, substitute formula (3-12) and (3-13) into formula (3-5) to obtain

[0148]

[0149] Taking the first-order derivative of equation (3-14), we can get

[0150]

[0151] make Available

[0152]

[0153] When i ≤0, then S≤0, which obviously does not hold. At the same time, when r i ≥1, T j ≤0, which is obviously not true. Therefore, this section only considers 0<r i <1. When 0<r i When <1, we can get

[0154] Taking the second-order derivative of equation (3-14), we can get

[0155]

[0156] because Appropriate hour, again It is always established, so there is always r i ∈(0,1), such that Therefore, in the first stage of the Stackelberg game, the optimal decision of the intermediate platform is Substituting formula (3-15) into (3-10) and (3-13), we can obtain:

[0157]

[0158] In summary, we can finally draw the following conclusion: When the user makes a decision The data owner decides Platform decision making This is the only equilibrium point of the Stackelberg game, and all three parties can obtain maximum profit.

[0159] Test results:

[0160] In order to verify the rationality of the proposed data value function, this paper conducts experiments on a real dataset called Mnist based on the machine learning model Keras. Keras is an open source artificial neural network framework written in Python, which is widely used in model construction and training in the field of deep learning. The Mnist dataset contains 60,000 handwritten digits ranging from 0 to 9, of which 50,000 are training sets and 10,000 are test sets. The machine learning model Keras is used to identify which number from 0 to 9 the picture is. This experiment uses datasets with different data sizes, accuracy, and availability to train the keras model, and uses the trained model to identify the same test set. The experimental results are the average accuracy of 10,000 test sets in 10 experiments.

[0161] Figure 2 The figure shows the comparison between the method of predicting data value based on data size and the actual value of data when λ1=0.9750, λ2=0.3154, λ3=0.0890. It can be observed that the relationship function α between data size and data value proposed in this paper is φ The actual results can be well approximated, which proves the accuracy of the proposed precision function.

[0162] The accuracy experiment in this article is achieved by replacing the labels of the pictures in the original database. Specifically, according to the accuracy ratio, the labels of the pictures in the original dataset are replaced with random numbers from 0 to 9. Assuming the accuracy is 0.9, the dataset sample is 5000, and the labels of 500 pictures are randomly replaced with 0 to 9. Figure 3 The figure shows the comparison between the method of predicting data value based on data accuracy and the actual value of data when ω1=0.9616, ω2=0.6934, ω3=5.2850. As can be seen from the figure, the data value function β proposed in this paper h Close to the actual value of the data, with an error within 4%.

[0163] The data truth rate experiment in this paper is achieved by replacing the pictures in the dataset. Specifically, the pictures in the original Mnist handwriting database are randomly replaced with cat and dog pictures according to the true data rate. The Keras model is used to train the replaced dataset to obtain the experimental results. Figure 4 The comparison chart between the data value based on data availability and the actual value of the data is shown when ν1=0.9574, ν2=0.5028, ν3=5.0775. It can be found that the scheme proposed in this paper can better fit the actual situation.

[0164] Finally, this paper uses 125 sets of data sets with increasing data size, accuracy, and availability to conduct simulation experiments to determine the appropriate fitting functions ε, γ, ρ and Multiple experimental results show that when ε=1.6117, γ=0.8784, ρ=0.8754 and η=-2.0171, the data value function proposed in this paper is close to the actual value of the data, and the average error is controlled at about 6%. Figure 5 The experimental results of the proposed method are compared with the actual value, where the blue dots represent the data value based on machine learning, and the blue bar graph represents the data value based on the linear regression method.

[0165] The parts not described in detail in the present invention are all known technologies to those skilled in the art. Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention is described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention can be modified or replaced by equivalents without departing from the purpose and scope of the technical solutions of the present invention, which should be included in the scope of the claims of the present invention.

Claims

1. A data trading method based on data value driving the balance of interests among multiple parties. It is characterized in that The steps include: S1. The data owner submits the data to the intermediary platform for consignment. The intermediary platform evaluates the consignment data to obtain the data value and publishes the data value in the system. Users purchase the data that best suits their own interests based on the data value. S2. A Stackelberg game economic framework is established. The Stackelberg game is divided into three stages: In the first stage, the intermediate platform acts as a leader and decides the service fee to be charged to the data owner; In the second stage, data owners, as followers of the intermediary platform and leaders of users, determine the data sales value; in the third stage, users, as followers, purchase the data that best suits their own interests by referring to the data value; S3. Seek the Nash equilibrium point of the Stackelberg game involving three parties.

2. The data transaction method based on data value driven and used to balance the interests of multiple parties according to claim 1, It is characterized in that The method by which the intermediate platform evaluates the consignment data to obtain the data value is: Among them, u k Indicates the value of data, represents the data size of data set k, θ k represents the accuracy of data set k, σ k represents the data availability of dataset k, Represents the relationship between data size and data value; Represents the relationship between data accuracy and data value; represents the relationship between data availability and data value; ε, γ and ρ are weight coefficients, is the bias term; λ 1 , 2 , 3 ,ω 1 ,ω 2 ,ω 3 , ν 1 、v 2 and v 3 is the fitting parameter.

3. The data transaction method based on data value driven and used to balance the interests of multiple parties according to claim 2, It is characterized in that Finding the optimal λ 1 , 2 and λ 3 The method is as follows: The function of the relationship between data value and data size is set as: To find the optimal λ 1 , 2 and λ 3 , under the same data accuracy and data availability, a set of training sets of different sizes and the same test set are set, and the data sizes of the training sets are {(φ 1 ),...(φ n ),...,(φ N-1 ),(φ N )}; the training sets are input into the machine learning model for training; after training, the same test set is input into the machine learning model to obtain the effect of different data sizes on the accuracy of the same test set; the model accuracy corresponding to these training sets is Enter the data size as a parameter to the function The data values ​​obtained based on the linear regression method are By using the gradient descent method, we can find the optimal λ 1 , 2 and λ 3 , which minimizes the minimum mean square error formula 1 , 2 and λ 3 As the optimal fitting parameter, the minimum mean square error formula is satisfied Finding the optimal ω 1 ,ω 2 and ω 3 The method is as follows: The function of the relationship between data value and data accuracy is set as: To find the optimal ω 1 ,ω 2 and ω 3 , under the same data size and data availability, a set of training sets with different data accuracy and the same test set are set. The data accuracy of the training sets is {(θ 1 ),...(θ h ), ...., (θ H-1 ), (θ H )}; input the training sets into the machine learning model for training; after the training is completed, input the same test set into the machine learning model to obtain the impact of different data accuracy on the accuracy of the same test set; the actual data values ​​corresponding to this set of training sets are Enter data accuracy as a parameter to the function The data values ​​obtained based on the linear regression method are Through the gradient descent method, we can find the optimal ω 1 ,ω 2 and ω 3 , so that Established; Finding the optimal ν 1 , ν 2 and v 3 The method is as follows: The function of the relationship between data value and data availability is set as: To find the optimal v 1 、v 2 and v 3 , under the same data size and data accuracy, set a set of training sets with different data availability and the same test set. The data availability of the training sets is {(σ 1 ),......(σ t ), ...., (σ T-1 ), (σ T )}; input the training set into the machine learning model for training; after the training is completed, input the test set into the machine learning model to obtain the impact of different data availability on the actual accuracy of the same test set. The actual data values ​​of this group of training sets with different availability are Input data availability as a parameter to the function The data values ​​obtained based on the linear regression method are Through the gradient descent method, we can find the optimal v 1 、v 2 and v 3 , so that Established; Find the optimal ε,γ,ρ and The method is as follows: Set up multiple sets of data sets with different data sizes, different data accuracy, and different data availability in proportion, and input them into the machine learning model for training. After the training is completed, input the test set into the machine learning model to obtain the actual accuracy of the same test set for different factors; Based on the optimal fitting parameter λ 1 , 2 , 3 ,ω 1 ,ω 2 ,ω 3 、v 1 、v 2 and v 3 , data sets of different data sizes, different data accuracy, and different data availability are input into the machine learning model as training sets. The accuracy of each training set corresponding to the same test set is {(D 1 ),......(D i ),....,(D M-1 ), (D M )},D i represents the accuracy corresponding to the i-th training set, and uses the gradient descent method to find the most suitable ε, γ, ρ and Make the data value and usage function obtained by the test set The error between the calculated results is minimal.

4. The data transaction method based on data value driven and used to balance the interests of multiple parties according to claim 3, It is characterized in that The ω 1 ,ω 2 and ω 3 The dataset in the acquisition method is realized by replacing the labels of the pictures in the original database; The v 1 、v 2 and v 3 The data set in the acquisition method is achieved by replacing the pictures in the original database.

5. The data transaction method based on data value driven and used to balance the interests of multiple parties according to claims 1 to 4, It is characterized in that The profit function of the intermediate platform in the Stackelberg game economic framework is: Among them, 0 <r i <1, the profit function consists of two parts, the first part [(p ik *r i +wc j )*x ijk ] is charged to the data owner for consignment, and the second part l(1+u k ) is the cost of storing dataset k on the platform, where p ik is the price at which data owner i sells dataset k, l is the cost coefficient of platform storage and operation, and u k is the data value of data set k, r i is the rate that the platform charges the data owner i when selling the dataset, w is the fixed fee for consignment, and x ijk is the deterministic function for buyer j to purchase data k from data owner i, c j is the transmission cost of the platform sending data to user h, and J represents the total number of users in the data trading market.

6. The data transaction method based on data value driven and used to balance the interests of multiple parties according to claim 5, It is characterized in that The data owner profit function of the Stackelberg game economic framework is: Among them, 0≤uk ≤ 1. The cost of the data owner's profit function consists of three parts. The first part is p ik *u k -r i *p ik -w is the reward for selling data, where p ik is the price that data owner i sells dataset k, u k is the data value of dataset k, w is the fixed rate x paid to the data seller ijk is the determination function for user j to purchase data from data owner i. When j decides to purchase data k from i, x ijk =1, otherwise x ijk =0; the second part c i The transmission cost of data owner i to transfer data to the intermediate platform for consignment, the third part e i *u k The cost of collecting dataset k for data owner i, e i is the correlation coefficient of cost.

7. The data transaction method based on data value driven and used to balance the interests of multiple parties according to claim 6, It is characterized in that The user profit function of the Stackelberg game economic framework is: THE j =[δ 1j +δ 2j ln(1+δ 3j *the k )]-the k *p ik Among them, δ 1h >0,δ 3j > 0, the user’s profit function is divided into two parts, the first part δ 1j +δ 2j ln(1+δ 3j *u k )] is the revenue obtained by users from purchasing data, δ 1j , δ 2j and δ 3j is the profit coefficient of user j, which is positively correlated with the actual profitability of user j; the second part u k *p ik The cost of purchasing dataset k for the user.

8. The data transaction method based on data value driven and used to balance the interests of multiple parties according to claim 7, It is characterized in that The Stackelberg game economic framework also includes the following constraints: S≥0 T i ≥0 U j ≥0。 9. The data transaction method based on data value driven and used to balance the interests of multiple parties according to claim 8, It is characterized in that The Nash equilibrium point of the Stackelberg game of the three parties in step S4 is: The user's optimal decision is: u k * The optimal data value for purchasing data k for the user. The optimal decision of the data owner is: p ik * The optimal price for data seller i to sell data k; The optimal decision of the platform is: r i * is the optimal consignment rate charged by the platform to data owner i.

Images