A multi-armed bandit based dynamic pricing method for graph data

Abstract

This invention discloses a dynamic pricing method for graph data based on a multi-armed slot machine model, addressing the problem of rigid pricing in existing graph data transactions and susceptibility to arbitrage by buyers. The method first retrieves historical transaction records based on the buyer's graph query request: if the graph data is being traded for the first time, a static data price is calculated based on the topological characteristics of nodes and edges; if it is a repeat transaction, a dynamic data price is calculated using a multi-armed slot machine model, combining the historical highest transaction price and lowest rejection price. After the buyer indicates whether they intend to purchase, the data intermediary can calculate the transaction's profit surplus and comprehensively consider multiple indicators such as basic costs, network centrality, transaction frequency, and success rate to update the data price for each node and edge, maintaining price consistency for nodes and edges across different graph data. This method can flexibly adapt to changes in market demand and effectively prevent arbitrage by buyers through unified pricing of nodes and edges, ensuring the effectiveness and fairness of graph data transactions.

Description

Technical Field

[0001] This application relates to the field of data trading and data pricing, specifically to a dynamic pricing method based on graph data. Background Technology

[0002] With the development of information technology, data has become a key factor of production. Among them, graph data has become an important carrier for depicting multidimensional and complex data relationships in many fields such as social network analysis, knowledge graph construction, and financial risk control.

[0003] As a new type of production factor, the full release of data value highly depends on its free circulation and sharing in the market. Constructing a reasonable and efficient data trading and pricing mechanism is a core element in promoting the market-based allocation of data factors. The traditional data trading model, which trades data in units of raw datasets, often suffers from serious limitations: on the one hand, raw datasets are massive and expensive, exceeding the budget limits of most data consumers; on the other hand, consumers' data needs are often targeted, and the trading model based on raw datasets ignores consumers' personalized demands.

[0004] To address this issue, query-based data pricing models are emerging. In this model, consumers submit personalized graph query requests to a data intermediary, which then returns matching graph data and prices it accordingly. However, existing methods still face two major challenges: First, most existing graph data pricing methods employ static pricing strategies, assigning a fixed price to the graph data. However, static pricing cannot adapt to the heterogeneity of buyer valuations and changes in market demand, easily leading to failed transactions or suboptimal returns. Second, graph data transactions may involve overlapping topologies, allowing buyers to circumvent high costs by splitting complex queries into multiple low-priced subqueries, thus engaging in arbitrage.

[0005] Therefore, how to provide a query-based graph data pricing scheme that can dynamically adapt to market demand while strictly preventing arbitrage is an urgent problem to be solved in graph data market transactions. Summary of the Invention

[0006] To address the aforementioned shortcomings in existing technologies, this invention provides a graph data dynamic pricing method based on multi-armed slot machines, thereby solving the technical problem that existing methods cannot simultaneously achieve dynamic revenue optimization and arbitrage-free constraints.

[0007] The technical content of this invention includes:

[0008] A graph data-driven dynamic pricing method based on multi-armed slot machines, comprising the following steps:

[0009] S1, Graph-based query The query results are used to determine whether there is graph data in past data transactions that is related to the current transaction. Matching historical transaction records;

[0010] S2. Based on the matching results of historical records, determine the graph data. Whether it is the first transaction or a repeat transaction, and use different pricing strategies to price the data accordingly;

[0011] S3, Data intermediaries will price data. The data is delivered to the buyer, and the graph data is processed based on the buyer's binary decision of "whether to purchase". The price of each node and each edge in the graph is updated to maintain the consistency between static and dynamic prices in the graph data.

[0012] Furthermore, step S2 includes the following sub-steps:

[0013] S21, If ​​graph data If it is the first time it is queried, the static price of the graph data is calculated based on the graph topology.

[0014] S22, If graph data It is data that is repeatedly traded, and data intermediaries dynamically price graph data based on historical transaction prices using a multi-armed slot machine model.

[0015] Furthermore, the node- and edge-based graph data price calculation in S21 includes the following sub-steps:

[0016] S211. The data intermediary extracts graph topology feature vectors for each node and each edge in the target subgraph, which serve as indicators for evaluating node and edge prices. Specifically, for node elements, the extracted graph topology features include degree centrality and betweenness centrality; for edge elements, the extracted graph topology features mainly include edge betweenness.

[0017] S212. Calculate the graph topology feature vector based on the weighted linear function to obtain the base price of each node and each edge. Since the topological features of nodes and edges often do not change with the data transaction, the data intermediary calculates the node and edge prices before all graph data transactions. The calculated prices can be regarded as the initial static prices of nodes and edges.

[0018] S213, Plot Data The base prices of all nodes and edges included are linearly aggregated and summed to obtain the static graph data price based on the graph topology. .

[0019] Furthermore, step S22 includes the following sub-steps:

[0020] S221. Obtain graph data from historical transaction data. The highest transaction price in history With the lowest rejection price ;

[0021] S222, If graph data The lowest historical rejection price Highest transaction price If the difference between the two prices exceeds a preset market volatility threshold, the data intermediary will use the arithmetic mean of the historical highest accepted price and the historical lowest rejected price as the current round of chart data. dynamic price ;

[0022] S223, if and If the difference between the two is less than or equal to the market volatility threshold, then the historical highest transaction price will be used. As data in this round of graphs dynamic price ;

[0023] S224, If graph data Only transactions that have been completed will Add the preset boundary detection step size as the basis for this round of bidding. ; and if the graph data The history only contains records of failed transactions. Subtract the preset detection step size as the bid for this round. .

[0024] Furthermore, the method for updating the prices of nodes and edges in S3 includes the following steps:

[0025] S31, Calculate the quotation The difference between the cost of the graph data and the cost of the graph data is used as the distributable surplus if the difference is greater than zero; otherwise, the distributable surplus is truncated to zero, and no price update is performed.

[0026] S32. Obtain graph data The system contains static and dynamic metrics for all nodes and edges. Static metrics include base cost and network centrality, with weights assigned empirically. Dynamic metrics include transaction frequency and historical transaction success rate, with weights calculated using the entropy weighting method.

[0027] S33. Combine static and dynamic weights to calculate the comprehensive importance weight of each node and each edge, allocate the allocable surplus to the corresponding nodes and edges, and calculate the updated data price.

[0028] Furthermore, the calculation of dynamic weight coefficients using the entropy weight method in step S32 includes the following sub-steps:

[0029] S321. The original matrix of importance indicators of the graph data is linearly transformed using the range standardization method to obtain the standardized matrix;

[0030] S322. Calculate the feature weight of each node and edge under each metric;

[0031] S323. Calculate information entropy based on feature weight;

[0032] S324. Normalize the difference coefficients to obtain the objective weights, and assign the objective weights to the corresponding dynamic weight coefficients.

[0033] The beneficial effects of this invention are:

[0034] This invention provides a dynamic pricing method based on graph data. In the absence of historical transactions, pricing can be performed at the underlying data structure level, cutting off arbitrage opportunities through split queries. When historical transactions exist, this invention introduces a multi-armed slot machine mechanism to dynamically determine the current transaction price based on historical transaction prices, thus dynamically adapting to market demand. Furthermore, this invention distributes dynamic price changes of graph data inversely to nodes and edges according to their weights, allowing nodes and edges in high-frequency queries and high-value locations to dynamically command premiums based on market demand. This ensures dynamic alignment between the micro-graph topology dimension and the macro-graph data price, achieving accurate quantification and scientific updating of data value. Attached Figure Description

[0035] Figure 1 This is a flowchart illustrating a graph-based dynamic pricing method for multi-armed slot machines.

[0036] Figure 2 This is a demonstration diagram of a graph-based dynamic pricing method for multi-armed slot machines; Detailed Implementation

[0037] To make the objectives, technical solutions, and advantages of this invention clearer, the embodiments of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are merely specific embodiments of this invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without inventive effort are within the scope of protection of this invention.

[0038] Figure 1 This is a flowchart illustrating the overall process of the graph data dynamic pricing method based on multi-armed slot machines proposed in this invention, demonstrating the basic data pricing framework of this method in a query-based graph data transaction scenario.

[0039] S1. Determine if historical transaction information exists.

[0040] Specifically, the data intermediary receives the query request. Graph data is obtained by querying the graph database. Subsequently, data-mediated computation graph data. The corresponding graph hash signature is used to retrieve historical transaction summaries based on its hash identifier, thus determining the graph data. Whether or not the data has been traded determines the subsequent pricing strategy for the graph data.

[0041] Figure 2 This diagram illustrates an example of graph query matching in this invention, demonstrating the matching process between a data intermediary and a buyer's query in a query-based graph data trading scenario. In a query-based graph data marketplace, the graph query submitted by a data buyer can be formally represented as... ,in: Represents the target set of buyer's needs; Indicates the relationships between nodes; This represents the attributes and labels of nodes or edges. When a data buyer submits a graph query request... At that time, the data intermediary performs graph matching in the graph database to obtain the query result graph. After each data transaction, the data broker dynamically maintains a historical transaction set of the data based on the buyer's purchasing decision, including the graph data queried for each transaction. Each historical transaction summary corresponds to a single historical transaction. To achieve efficient retrieval and reduce computational complexity, the historical transaction summary uses a graph hash signature as the unique identifier for the graph data. The specific content recorded includes:

[0042] (1) Graph hash signature: It serves as a unique identifier for graph data and has a one-to-one correspondence with graph data, which is used for fast retrieval;

[0043] (2) The highest historical acceptance price : Represents graph data The highest price ever paid in a previous transaction;

[0044] (3) Historical lowest rejection price : Represents graph data The lowest offer that was explicitly rejected by the buyer in previous transactions;

[0045] (4) Transaction frequency statistics: including graph data The number of completed and failed transactions in the past reflects the stability of market demand for this data.

[0046] If the data intermediary retrieved it from the historical data summary The corresponding graph hash signature indicates that It has been traded; if not found, it means... This is the first time it has been searched in this round of transactions.

[0047] S2, Generate Data Quotation

[0048] Based on the search results from S1, data intermediaries will adopt different pricing strategies:

[0049] S21. If no match exists The historical transaction summary illustrates the graph data. When a graph is queried for the first time, the data intermediary will price the graph data based on the graph topology.

[0050] S211. In data transactions, the data intermediary will always maintain the price set of each node and edge contained in the graph database, that is, the price set of each node. Corresponding to a price Each node Corresponding to a price Before data transactions, intermediaries extract graph topology feature vectors for each node and edge in the target subgraph, using them as indicators to evaluate node and edge prices. For node elements, the extracted graph topology features include degree centrality and betweenness centrality; for edge elements, the extracted graph topology features mainly include edge betweenness.

[0051] S212. Calculate the graph topology feature vector based on the weighted linear function to obtain the base price of each node and each edge. Since the topological features of nodes and edges often do not change with the data transaction, the data intermediary calculates the node and edge prices before all graph data transactions. The calculated prices can be regarded as the initial static prices of nodes and edges.

[0052] S213, Plot Data The base prices of all nodes and edges included are linearly aggregated and summed to obtain the static graph data price based on the graph topology. :

[0053]

[0054] in, For graph data The set of nodes, For graph data The set of edges. This pricing mechanism strictly satisfies subadditivity mathematically, thus guaranteeing the no-arbitrage nature of graph data pricing.

[0055] S22. If a match exists The historical transaction summary illustrates the graph data. If a query is repeated, the data intermediary will then dynamically price the service based on a multi-armed slot machine model.

[0056] S221. Based on the graph data retrieved in S1 Historical transaction summaries allow intermediaries to obtain the highest historical acceptance price. and the lowest historical rejection price ;

[0057] S222, Exploration Phase: If Graph Data The lowest historical rejection price Highest transaction price The difference between This indicates that the buyer's psychological expectation range is relatively wide. The data intermediary then uses the arithmetic mean of the historical highest accepted price and the historical lowest rejected price. As graph data Dynamic prices in this round ;

[0058] S223, Utilization Stage: If the difference The data intermediary will then use the highest historical transaction price. As data in this round of graphs dynamic price To ensure a high success rate for transactions;

[0059] S224, Information Missing Stage: If the historical record only contains accepted records, then probe upwards and select... As graph data In this round of bidding; if there are only rejection records, then try to lower the bid, choosing... As a quote .

[0060] In summary, the dynamic data pricing calculation formula based on historical transaction prices is as follows:

[0061]

[0062] in, The system is preset with a market volatility sensitivity threshold. The price testing step size set for data intermediaries.

[0063] S3, Update node and edge prices

[0064] When data buyers submit graph queries, they have a mental price expectation for the data they wish to purchase. That is, for graph data The highest price a buyer is willing to pay. Data brokers will use graph data. Quotation After informing the buyer, the buyer will base their decision on their own expected price. Making a purchase decision: When When the price is less than or equal to the buyer's expected price, the buyer accepts the offer and completes the transaction; conversely, if the offer exceeds the buyer's tolerance limit, the buyer rejects the transaction. Since a buyer's purchase decision only involves two outcomes—acceptance or rejection—the buyer's purchase decision feedback can be defined as a... ,like ,but ;like ,but The data broker will base its decisions on buyer feedback. Update and maintain historical transaction summaries, and calculate graph data. The system assigns weights to the metrics of each node and each edge, and updates the prices of the nodes and edges.

[0065] S31. Non-negative profit constraint and earnings calculation

[0066] Data brokers first calculate the quote. With graph data cost The difference between them, to ensure the basic costs of the intermediary, will be used to allocate the surplus profit. Defined as:

[0067]

[0068] in, Representative graph data The cost. If the transaction fails or the offer is lower than the total cost, then The threshold is 0, and no price update is performed.

[0069] S32. Calculate the comprehensive importance weight of nodes and edges based on the combined weighting method.

[0070] Graph data price updates refer to data intermediaries updating prices based on distributable profit surpluses. For graph data The process involves updating the prices of the nodes and edges contained within the graph; therefore, the intermediary needs to update the prices for each underlying graph element participating in this transaction. (i.e., nodes and edges) Calculate their importance weights. :

[0071]

[0072] in, Image elements The cost, It is an element Trading frequency, Representative elements Network centrality, It is the historical transaction success rate of this element. , , and These are the weighting coefficients corresponding to the four indicators mentioned above.

[0073] Among the above indicators, cost and network centrality These are static indicators. They reflect the inherent attributes and topological importance of graph elements, and are relatively stable in the short term. Their corresponding weight coefficients... and Pre-set based on experience. And the trading frequency... and historical transaction success rate These are dynamic indicators that fluctuate continuously as market transactions proceed, directly reflecting the real-time demand from buyers and the market's acceptance of the data. The weighting coefficients corresponding to dynamic indicators... and The entropy weight method will be used for calculation, which is dynamically calculated based on the objective information entropy of the current data matrix to eliminate the bias of subjective assignment.

[0074] Furthermore, the calculation of dynamic weight coefficients using the entropy weight method in S32 includes the following steps:

[0075] S321. Data standardization and proportion calculation

[0076] Assuming the current query contains By extracting the basic elements of each graph and considering transaction frequency and historical transaction success rate, an original evaluation matrix is ​​constructed. The number of indicators in this article To eliminate the computational impact of differences in dimensions and orders of magnitude between different indicators, the system employs a range normalization method for the matrix. Perform a linear transformation to obtain the normalized matrix. .

[0077]

[0078] S322. Calculate the feature weight of each node and edge under each metric.

[0079] Based on the standardized matrix, calculate the... The basic elements of the diagram are in the first... Feature proportions under dynamic indicators The calculation formula is as follows:

[0080]

[0081] S323. Calculating information entropy based on feature weight.

[0082] For the A dynamic indicator, its information entropy The calculation formula is defined as follows:

[0083]

[0084] in, This is an adjustment constant related to the sample size, typically set to a value of [value missing]. To ensure the calculated information entropy Always fall Within the range. Specifically, when the characteristic proportion... At that time, it was stipulated .

[0085] S324. Normalize the difference coefficients to obtain the objective weights.

[0086] Based on the calculated information entropy, the first... Coefficient of difference of the items The larger the coefficient of variation, the richer the effective information contained in the indicator, and the more significant its contribution to the overall evaluation. Then, through normalization, the final objective weight is obtained. The calculation formula is as follows:

[0087]

[0088] The objective weights calculated above Directly and precisely assign values ​​to the dynamic weight coefficients defined above. and This provides a scientific quantitative basis for updating the prices of nodes and edges.

[0089] S33. Complete the price update of the underlying nodes and edges.

[0090] After obtaining the weights of each node and edge, the intermediary will determine the weights based on the profit surplus. Calculate the update price of each graph element based on the normalized comprehensive weight ratio. Given the graph data... Includes Each graph element (i.e., the sum of nodes and edges), element Updated prices The calculation method is as follows:

[0091]

[0092] This mechanism ensures that nodes and edges with high-frequency queries and high success rates can reasonably accumulate price premiums, achieving dynamic alignment between the dynamic pricing of macro graph data and the basic prices of micro graph elements.

Claims

1. A graph data-based dynamic pricing method for multi-armed slot machines includes the following steps: S1, Graph-based query The query results are used to determine whether there is graph data in past data transactions that is related to the current transaction. Matching historical transaction records; S2. Based on the matching results of historical records, determine the graph data. Whether it's the first transaction or a repeat transaction, and then using different pricing strategies to price the data accordingly: S21, If ​​graph data If this is the first query, the static price of the graph data is calculated based on the graph topology: S211. The data intermediary extracts the graph topology feature vector of each node and each edge in the target subgraph as an indicator to evaluate the price of nodes and edges. For node elements, the extracted graph topological features include degree centrality and betweenness centrality; for edge elements, the extracted graph topological features mainly include edge betweenness. S212. Calculate the graph topology feature vector based on the weighted linear function to obtain the base price of each node and each edge. Since the topological features of nodes and edges often do not change with the data transaction, the data intermediary calculates the node and edge prices before all graph data transactions. The calculated prices can be regarded as the initial static prices of nodes and edges. S213, Plot Data The base prices of all nodes and edges included are linearly aggregated and summed to obtain the static graph data price based on the graph topology. , S22, If graph data The data is the data of repeatedly queried transactions. The data intermediary then dynamically prices the graph data based on historical transaction prices using a multi-armed slot machine model. S221. Obtain graph data from historical transaction data. The highest transaction price in history With the lowest rejection price ; S222, If graph data The lowest historical rejection price Highest transaction price If the difference between the two prices exceeds a preset market volatility threshold, the data intermediary will use the arithmetic mean of the historical highest accepted price and the historical lowest rejected price as the current round of chart data. dynamic price ; S223, if and If the difference between the two is less than or equal to the market volatility threshold, then the historical highest transaction price will be used. As data in this round of graphs dynamic price ; S224, If graph data Only transactions that have been completed will Add the preset boundary detection step size as the basis for this round of bidding. ; and if the graph data The history only contains records of failed transactions. Subtract the preset detection step size as the bid for this round. , S3, Data intermediaries will price data. The data is delivered to the data buyer, and the graph data is processed based on the buyer's binary decision of "whether to purchase". The price of each node and each edge in the graph is updated to maintain the consistency between static and dynamic prices in the graph data. S31, Calculate the quotation The difference between the cost of the graph data and the cost of the graph data is used as an allocable surplus if the difference is greater than zero; otherwise, the allocable surplus is truncated to zero, and no price update is performed. S32. Obtain graph data The static and dynamic metrics values ​​for all nodes and edges in the network are provided. Static metrics include basic cost and network centrality, with weights obtained through empirical weighting. Dynamic metrics include transaction frequency and historical transaction success rate, with weights calculated using the entropy weighting method. S321. The original matrix of importance indicators of the graph data is linearly transformed using the range standardization method to obtain the standardized matrix; S322. Calculate the feature weight of each node and edge under each metric; S323. Calculate information entropy based on feature weight; S324. Normalize the difference coefficients to obtain the objective weights, and assign the objective weights to the corresponding dynamic weight coefficients. S33. Combine static and dynamic weights to calculate the comprehensive importance weight of each node and each edge, allocate the allocable surplus to the corresponding nodes and edges, and calculate the updated data price.

Images