Identification of risky platforms and prediction of risk propagation in commodity trading markets
Through complex networks and SIR infectious disease models, the risk platforms and predict risk transmission paths in the commodity trading market are identified, which solves the problems of risk identification and transmission prediction in the market, and improves regulatory efficiency and targeted policy formulation.
Patent Information
- Application Number
- CN202210532466.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-05-12
- Publication Date
- 2024-11-19
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
There are frequent risk events in the commodity trading market, and it is difficult for existing technologies to efficiently identify risk platforms and predict risk transmission paths, resulting in low regulatory efficiency and inadequate policy formulation.
Complex network structure and SIR infectious disease model are adopted to build a trading network, identify abnormal trading patterns, calculate the correlation coefficients between platforms, construct the minimum spanning tree, simulate the risk transmission process, and predict the risk transmission path and range.
It improves the efficiency of risk platform identification, reduces the harm caused by risk transmission, enhances the pertinence of policy formulation, and helps regulatory departments establish and improve systems and prevent risks in advance.
Smart Images

Figure CN115147217B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of bulk commodity e-commerce supervision, and specifically relates to a method for identifying risk platforms and predicting risk propagation in a bulk commodity trading market. Background Art
[0002] With the rapid development of Internet information technology and the continuous innovation of electronic finance, various types and trading modes of electronic commodity trading markets have emerged rapidly. The types of commodities involved involve many industries and fields such as chemicals, grain, steel, plastics, and precious metals. The distribution range covers 23 provinces, municipalities, and autonomous regions such as Shanghai, Shandong, Guangdong, Beijing, and Guangxi, which are of great strategic significance. Because of its diverse trading modes, large trading volume, and large price fluctuations, it has certain financial characteristics, which has led to frequent risk events in recent years, and then triggered the promulgation of various policies and measures to regulate the market related to commodity trading. In order to find trading platforms that violate policies and measures, a large amount of manpower is often required for screening. At the same time, the trading commodities in the trading platforms have overlapping parts. When a trading platform violates policies and regulations and is ordered to rectify, the impact of this abnormal trading mode may also spread to related trading platforms, which is easy to bring certain risks to other trading platforms. The final result may be possible spread, thereby exacerbating the outbreak of the original crisis. With the help of software system algorithms, helping regulatory authorities screen trading platforms with abnormal trading modes can improve the efficiency of supervision. At the same time, identifying risk transmission paths and possible infection platforms will help regulatory authorities establish and improve relevant systems, and will also help trading platforms take precautions against impending risk infections.
[0003] A complex network is composed of many nodes and edges between them. Nodes are used to represent different individuals in the real system, while edges are used to represent the relationship between individuals. The topological structure can well reflect the connection relationship between nodes. With the help of complex network structure, trading accounts are used as nodes and trading behaviors are used as edges to summarize the types of abnormal trading patterns and their topological structures prohibited in the current policy measures for the commodity market. Then, the trading network of each platform is constructed, and the power law distribution law existing in the trading network is used to analyze whether there are abnormal trading patterns in the trading network, so as to determine the trading platforms with transmission risks.
[0004] In the study of complex network problems, the problem of transmission dynamics is one of the important branches. This problem mainly studies the transmission mechanism on different types of complex networks such as natural networks and social networks, and explores feasible and effective methods to control transmission. Susceptible-Infectious-Recovered (SIR) is one of the classic models in the study of transmission dynamics. It was originally used to study the law of the Black Death in London, England in the 16th century. It was later widely applied to other practical problems and is the basis for the study of infectious disease dynamics. Combining complex networks with the SIR infectious disease model is conducive to predicting the infection-immunity process and intuitively measuring the risk contagion effect. Therefore, the method of transmission dynamics in complex networks is introduced into the risk propagation of commodity trading markets. If all trading platforms are regarded as nodes, then the entire commodity trading market can be regarded as a network. When a node has a problem, other nodes in the network may be infected through point-to-point connections. Under this network model, the spread of risks in commodity trading markets can be predicted, and the abnormal risk propagation paths and scopes can be explored, which is not only conducive to the supervision of regulatory authorities, but also conducive to crisis prevention. Summary of the invention
[0005] To solve the above problems, the present invention discloses a method for identifying risk platforms and predicting risk propagation in commodity trading markets, which is used to predict risks and the path and scope of risk propagation for platform practitioners and regulatory authorities in commodity electronic trading markets. The system first constructs a trading network for relevant trading data of commodities, then extracts and models the prohibited trading modes in the industry policies of some existing commodity trading markets, uses network features to find abnormal trading modes, and sets the platform to which it belongs as the source platform. At the same time, according to the correlation between the overlapping of trading commodities between trading platforms, the Pearson correlation coefficient between platforms is calculated, and then the minimum spanning tree is constructed to generate a complex network between trading platforms to show the strongest correlation between trading platforms. Finally, taking the source platform as the starting point and the complex network between platforms as the propagation network, the SIR infectious disease model is applied to predict the path and scope of risk propagation and display the trading platforms that may be affected by the risk. The risk platform identification method based on the power law distribution law and the risk propagation prediction method combining the complex network and the SIR infectious disease model proposed by the system can be used as a reference for risk prevention of trading platforms in commodity trading markets and a reference for the soundness of the regulatory authorities.
[0006] To achieve the above object, the technical solution of the present invention is as follows:
[0007] The method for identifying risk platforms and predicting risk propagation in commodity trading markets includes the following steps:
[0008] (1) After the system is initialized, the database reads historical transaction data and builds a user transaction network;
[0009] (2) Conduct network modeling of prohibited transaction patterns in the policy and summarize the power law distribution rules;
[0010] (3) Use the least squares method to fit the power law distribution to the transaction account nodes and calculate the outliers;
[0011] (4) Calculate the threshold value and screen out the trading patterns formed by trading accounts with outliers greater than the threshold value, which are abnormal trading patterns and the platforms they belong to are risk source platforms;
[0012] (5) Calculate the correlation coefficients between platforms based on the platform's rating matrix for trading products, and finally obtain the correlation coefficient matrix between trading platforms;
[0013] (6) Taking the trading platforms as nodes of the tree, the minimum spanning tree is constructed based on the correlation coefficient matrix between the trading platforms;
[0014] (7) treating the minimum spanning tree constructed in step (6) as the initial complex network structure and establishing the SIR epidemic model;
[0015] (8) All platform nodes are added to the susceptible state set S, and then according to the screening results of abnormal transaction patterns in (4), the risk source platform is transferred from the susceptible state set S to the infected state set I;
[0016] (9) Traverse the trading platforms in the susceptible state set S. If there is an infected node among the adjacent nodes, that is, in set I, then according to the infection rate, there is a certain probability that it will develop into an infected node. If infected, it will be removed from set S and added to set I;
[0017] (10) Traverse the trading platforms in the infected state set I. The infected platform nodes have a certain probability of changing to the recovery state. The cured nodes are removed from set I and added to set R;
[0018] (11) Determine whether the number of infected nodes has changed compared with the previous iteration and whether the number of iterations is less than the maximum value. If so, return to step (9); otherwise, go to step (12);
[0019] (12) After the risk source platform is identified and the risk infection scope and path are predicted, the risk source platform and its risk infection scope and path are adaptively displayed based on the results.
[0020] Furthermore, in step (1), the trading account in the platform is defined as U = {u1, u2, ..., u n}, a k Indicates the number of transactions, s kIndicates the type of traded goods, then each transaction can be written in the form of a four-tuple i ,u j ,a k ,s k >, where u i For the seller, u j For the buyer, a k Indicates the number of transactions, s k Indicates the type of traded goods. Map the transaction set in the commodity trading market into a directed graph model G = {V, E, W}, where the node set V represents all trading accounts involved in the commodity trading market, that is, V = {u1,u2,…,u n}, the edge set E is the set of all transaction behaviors, that is, E = { i ,u j ,a1,s1>,…, i ,u j ,a n ,s n >}, with transaction volume as the attribute of each edge, the directionality of the edge represents the buying and selling of the transaction, forming a directed weighted graph with multiple edges.
[0021] Furthermore, in step (2), the abnormal transaction patterns specified in the policy are summarized to establish a model and extract features, which are expressed by using the related structure of complex network nodes, and the corresponding three power law distribution laws are sorted out:
[0022] Number of edges ~ number of neighbors: The total number of connected edges E and the number of neighbors N of most normal transaction nodes follow a power law distribution, that is, the logarithmic coordinates are linear:
[0023] E i ~N i a ,1≤a≤2
[0024] Weight and number of edges: The total edge weight and number of edges of most normal transaction nodes follow a power law distribution:
[0025] W i ~N i c ,c≥1
[0026] Main eigenvalue ~ weight sum: The main eigenvalue of the central network of the node and the central network weight satisfy a certain power law relationship:
[0027] λ i ~W i γ ,0.5≤γ≤1
[0028] Furthermore, in step (3), according to the summarized power law distribution law that satisfies the linear relationship, a linear regression model anomaly detection algorithm is used to detect abnormal transaction subjects, and the power law distribution is parameter-fitted by the least squares method, and an outlier (os i ) represents node u i The degree of deviation from the normal mode under a given power law distribution characteristic is formally expressed as That is, the distance between the predicted value and the actual value, to represent the node u i The difference between the observed value and the predicted value under a given power law distribution, node u i Outlier values of os i It is necessary to reflect the degree to which the node deviates from the normal mode. i The closer the observed value is to the predicted value, the better the node u i The closer the corresponding transaction behavior is to the normal transaction behavior pattern of most transactions, the lower the abnormal value associated with the node. Otherwise, it means that the transaction pattern formed by the transaction account represented by the node and its related transaction parties is a possible abnormal transaction pattern.
[0029] Furthermore, in step (4), it is assumed that the outliers of all nodes approximately satisfy a condition with a mean of μ and a variance of σ. 2 The normal distribution N(μ,σ 2 ). Select μ+3σ as the threshold of the anomaly score. If the anomaly value is greater than μ+3σ, the transaction pattern formed by the account group to which the account belongs is an abnormal transaction pattern, and the platform to which the account belongs is set as a risk platform.
[0030] Furthermore, in step (5), the commodity scores are performed based on the proportion of each commodity transaction volume in the total transaction volume of the platform. Assuming that on platform P i The total amount of commodity transactions is sum i , Product B j The transaction volume is b j , then product B j On platform P i The scores on are:
[0031]
[0032] Among them, α is a constant that controls the rating within a certain range. Thus, the platform-product rating matrix R is obtained, and then the correlation between two platforms is calculated based on the Pearson correlation coefficient, and finally the correlation coefficient matrix ρ between trading platforms is obtained.
[0033] Furthermore, in step (6), the trading platforms are used as nodes of the tree, and a minimum spanning tree is constructed based on the correlation between the trading platforms. The weights in the minimum spanning tree are defined by the distance between the correlations between the trading platforms. For example, platform P iand Platform P k The similarity between ik , then the distance between the two trading platforms is:
[0034]
[0035] The distance matrix D between trading platforms is obtained, and the minimum spanning tree is constructed according to the Prim algorithm.
[0036] Furthermore, in step (7), the minimum spanning tree constructed in (6) is regarded as the initial complex network structure, G =<V,E> , where V represents the set of points in the network, E represents the set of edges in the network, and the node set V = {P1, P2, ..., P m}, is a platform node, and the set of edges can be expressed as E = { <P i ,P j ,d ij >,… <P q ,P r ,d qr >}, where d ij is the edge weight, which is also the distance between the two platforms calculated in step (6), forming the most strongly correlated complex network structure between trading platforms.
[0037] At the same time, a susceptible state set S is established to store nodes in a healthy state, which belongs to the set of platforms that are not affected by the risk of abnormal trading patterns; an infected state set I is established to store nodes that are infected and have a certain chance of recovery, which belongs to the set of platforms that are affected by the risk of abnormal trading patterns; and a recovered state set R is established, that is, nodes that have recovered from the infection period, which belong to the platforms that are affected by the risk of abnormal trading patterns but adjust their trading patterns according to market rules and return to normal.
[0038] Furthermore, in step (8), all platform nodes are added to the susceptible state S set, and then according to the screening results of abnormal transaction patterns in (4), assuming that an infectious disease occurs, that is, there is a platform with abnormal transaction patterns screened in (4), it is transferred from the susceptible state S set to the infected state set I.
[0039] Furthermore, in step (9), the trading platforms in the susceptible state set S are traversed. If there is an infected node among the adjacent nodes, that is, in set I, then according to the infection rate, there is a certain probability that it will develop into an infected node. If infected, it will be removed from set S and added to set I.
[0040] Furthermore, in step (10), the trading platforms in the infected state set I are traversed, and the infected platform nodes have a certain probability of becoming a recovered state, and the cured nodes are removed from the set I and added to the set R.
[0041] Furthermore, in step (11), it is determined whether the number of infected nodes has changed compared with the previous iteration and whether the number of iterations is less than the maximum value. If so, iterate (9)-(10). Otherwise, the risk source platform is screened out, and the risk range caused by it and the trading platforms on the path can also be predicted. Therefore, screening the risk source platform, finding the risk transmission path and possible infected platforms, is conducive to the establishment and improvement of relevant systems by the regulatory authorities, and also helps the trading platform to take precautions against the risk infection that may be suffered in advance.
[0042] The beneficial effects of the present invention are:
[0043] (1) Improve the efficiency of identifying risky platforms: With the help of algorithm technology in the Internet field, there is no need for human screening and interference. By analyzing historical transaction data and summarizing the characteristics of abnormal transaction patterns, trading platforms with abnormal transaction patterns can be identified and set as risk source platforms.
[0044] (2) Reduce the harm caused by risk transmission: By constructing a complex network for the trading platform, simulating the risk transmission process, predicting the path of risk transmission and the scope of the affected platform, and issuing early warnings to the trading platform for risk occurrence, it can serve as a reference for the trading platform to prevent risks.
[0045] (3) Enhance the targeted nature of policy making: By identifying possible loopholes in risk platform outcomes and the scope of risk transmission, regulatory authorities can use this as a reference when adjusting policy measures, which will help regulatory authorities establish and improve relevant systems. BRIEF DESCRIPTION OF THE DRAWINGS
[0046] Figure 1 Main flow chart of the present invention.
[0047] Figure 2 , 3 4 is a schematic diagram of an abnormal transaction mode of the present invention.
[0048] in Figure 2 , 3 The circular nodes in 4 represent trading account nodes, the direction of the connecting edge represents the buying and selling direction, the starting point of the arrow represents the seller, and the end of the arrow represents the buyer. The thickness of the connecting edge represents the size of the transaction volume. The thicker the connecting edge, the larger the transaction volume. The marking numbers 1, 2, 3, ..., 13 represent the sequence numbers of the nodes and connecting edges. DETAILED DESCRIPTION
[0049] The present invention will be further explained below in conjunction with the accompanying drawings and specific embodiments. It should be understood that the following specific embodiments are only used to illustrate the present invention and are not used to limit the scope of the present invention.
[0050] The method for identifying risk platforms and predicting risk propagation in the commodity trading market described in the present invention does not require further human analysis when historical trading information and information related to the trading platform are known. Instead, the system analyzes the trading status of the entire commodity trading market based on a complex network analysis method, identifies the risk source platform, and predicts the path of risk propagation and the scope of the affected platform.
[0051] Identification of risk platforms and risk propagation prediction methods in commodity trading markets, such as Figure 1 As shown, the specific steps include:
[0052] (1) Define the transaction account in the platform as U = {u1,u2,…,u n}, a k Indicates the number of transactions, s k Indicates the type of traded goods, then each transaction can be written in the form of a four-tuple i ,u j ,a k ,s k >, where u i For the seller, u j For the buyer, a k Indicates the number of transactions, s k Indicates the type of transaction commodity. There may be multiple different transactions between two trading accounts, so the transaction set of the two trading accounts can be expressed as:
[0053] p ij ={ i ,u j ,a1,s1>,…, i ,u j ,a n ,s n >}
[0054] The transaction set in the commodity trading market is mapped into a directed graph model G = {V, E, W}, where the node set V represents all the trading accounts involved in the commodity trading market, that is, V = {u1,u2,…,u n}, the edge set E is the set of all transaction behaviors, that is, E = { i ,u j ,a1,s1>,…, i ,u j ,a n ,s n >}, with transaction volume as the attribute of each edge, the directionality of the edge represents the buying and selling of the transaction, forming a directed weighted graph with multiple edges.
[0055] (2) At present, the state has no clear regulations on abnormal trading patterns in the commodity trading market. With the help of the state's regulations on abnormal trading patterns in securities and futures, that is, Article 70 of the Futures Management Regulations, it can be expanded into the following categories:
[0056] 1) Individually or in collusion, concentrating financial advantages, position advantages, or using information advantages to jointly or continuously buy and sell contracts to manipulate futures trading prices;
[0057] 2) Intentionally colluding to conduct futures transactions with each other at a pre-agreed time, price and method, thereby affecting the futures transaction price or volume;
[0058] 3) Taking oneself as the trading partner, buying and selling by oneself, thus affecting the futures trading price or futures trading volume;
[0059] 4) Hoarding spot goods to influence the futures market;
[0060] 5) Other acts of manipulating futures trading prices as prescribed by the futures regulatory authority of the State Council.
[0061] The model is established and features are extracted for the abnormal transaction patterns specified in the above policies, and they are expressed using the related structure of complex network nodes, such as Figure 2 (Among them, abnormal trading mode 3) has been strictly controlled by the trading platform and is not considered.) At the same time, in the commodity trading market, normal trading nodes often satisfy a certain power-law distribution. The formulas of these power-law distribution characteristics are composed of the characteristic attributes of the trading nodes, which can represent normal trading modes to a certain extent. Therefore, the degree of deviation from the power-law distribution can be used to determine whether the node has an abnormal trading mode:
[0062] Number of edges ~ number of neighbors: The total number of connected edges E and the number of neighbors N of most normal transaction nodes follow a power law distribution, that is, the logarithmic coordinates are linear:
[0063] E i ~N i a ,1≤a≤2
[0064] For nodes with abnormal transaction patterns, there may be situations where neighbors are closely connected or rarely connected. Through the buying and selling relationships with multiple neighbors, the purpose of hoarding is achieved, or multiple neighbors are united to let them engage in reselling, conduct joint hoarding, and manipulate prices. The typical structure is as follows: Figure 2 The star structure and polygon structure in the graph can be screened out by the degree to which the nodes deviate from this power-law distribution, and the nodes with abnormal trading patterns 4) or 1) can be obtained.
[0065] Weight and number of edges: The total edge weight and number of edges of most normal transaction nodes follow a power law distribution:
[0066] W i ~N i c ,c≥1
[0067] In normal trading mode, the total weight and edges conform to a certain power law distribution relationship. However, in abnormal trading mode, there may be a huge total weight that does not match the number of edges. That is, there are users who create false market conditions by maintaining close trading relationships with their neighbors. The typical structure is that the total weight is too large, such as Figure 3 The transaction pattern structure in can be screened out by the degree to which the nodes deviate from this power-law distribution, and nodes with abnormal transaction patterns 2) can be obtained.
[0068] Main eigenvalue ~ weight sum: The main eigenvalue of the central network of the node and the central network weight satisfy a certain power law relationship:
[0069] λ i ~W i γ ,0.5≤γ≤1
[0070] In normal transaction mode, the total weight and edge conform to a certain power law distribution relationship, but in abnormal transaction mode, there may be a huge total weight that does not match the number of edges, that is, there is a single large transaction. The typical structure is that the weight of a certain edge is too large, such as Figure 4 The transaction pattern structure in can be screened out by the degree to which the nodes deviate from this power-law distribution, and the nodes with abnormal transaction patterns 1) can be obtained.
[0071] (3) According to the power law distribution law that satisfies the linear relationship summarized in (2), the linear regression model anomaly detection algorithm is used to detect abnormal transaction entities. For each type of power law distribution, the input of each power law distribution is first defined as follows:
[0072] 1) Density power law distribution
[0073]
[0074] 2) Power-law distribution of weights
[0075]
[0076] 3) Power-law distribution of main eigenvalues
[0077]
[0078] Then the power law distribution is fitted by the least squares method. Given the power law distribution output X = {x1, x2, ..., xn},Y={y1,y2,…,y n}, assuming that the data satisfies the linear regression model Y = aX + C + ε, all we need to do is to use the x in the data sample i and i The value of is used to fit this linear relationship, that is, to estimate the parameters a and C (ε is the error). This paper uses the least squares method to estimate the parameters and obtain the estimated value:
[0079]
[0080]
[0081] in
[0082] In order to separate abnormal transaction patterns, the outlier value (os i ) represents node u i The degree of deviation from the normal mode under a given power law distribution characteristic is formally expressed as Represents node u i The difference between the observed value and the predicted value under a given power law distribution, node u i Outlier values of os i It is necessary to reflect the degree to which the node deviates from the normal mode, so the node u is defined i The outlier calculation formula is as follows:
[0083]
[0084] When node u i The closer the observed value is to the predicted value, the better the node u i The closer the corresponding transaction behavior is to the normal transaction behavior pattern of most transactions, the lower the abnormal value associated with the node. Otherwise, it means that the transaction pattern formed by the transaction account represented by the node and its related transaction parties is a possible abnormal transaction pattern.
[0085] (4) Assume that the outliers of all nodes approximately satisfy a condition with mean μ and variance σ. 2 The normal distribution N(μ,σ 2 ). In a normal distribution, the probability that data falls in [μ-3σ,μ+3σ] is 99%. Based on the existing conditions, μ+3σ is selected as the threshold of the abnormal score. If the abnormal value is greater than μ+3σ, the transaction pattern formed by the account group to which the account belongs is an abnormal transaction pattern, and the platform where the account belongs is set as a risk platform.
[0086] (5) There is often a certain correlation between trading platforms due to the overlap of trading commodities, and trading platforms have their own special preferences for trading commodities. Commodity scores are calculated based on the proportion of each commodity's trading volume in the total trading volume of the platform, and a commodity scoring matrix is obtained. Assume that on platform P i The total amount of commodity transactions is sum i , Product B j The transaction volume is b j , then product B j On platform P i The scores on are:
[0087]
[0088] Where α is a constant that controls the score within a certain range. Assuming there are m trading platforms and n types of trading products, we can get the platform-product score matrix R:
[0089]
[0090] Using R i Indicates platform P i The product rating vector, R i =[R i,1 R i,2 …R i,n ], and then calculate the correlation between platforms according to the Pearson correlation coefficient. i and Platform P k The similarity between can be expressed as:
[0091]
[0092] Where E is the mathematical expectation or mean, D is the variance, E{(P i -EP i )(P k -EP k )} is the covariance, and finally, the similarity formula can be expressed as:
[0093]
[0094] Finally, the correlation coefficient matrix ρ between trading platforms is obtained:
[0095]
[0096] (6) Taking the trading platforms as nodes of the tree, the minimum spanning tree is constructed based on the correlation between the trading platforms. The weights in the minimum spanning tree are defined by the distance between the correlations between the trading platforms. For example, platform P i and Platform P k The similarity between ik , then the distance between the two trading platforms is:
[0097]
[0098] The distance d under this definition is ik The three axioms of measurement are fully satisfied:
[0099] (1) If and only if i = k, d i,k =0
[0100] (2)d i,k =d k,i
[0101] (3)d i,k ≤d i,j +d j,k
[0102] The distance matrix D between trading platforms is obtained, and the minimum spanning tree is constructed according to the Prim algorithm. According to the construction principle of the minimum spanning tree, the minimum spanning tree is an acyclic graph with N-1 edges connecting N nodes. N-1 shorter edges are selected to connect all nodes. In the minimum spanning tree of trading platforms, the distance d between trading platforms is i,k As the correlation coefficient ρ between trading platforms i,k Therefore, the structure of the minimum spanning tree gives the strongest correlation between each trading platform.
[0103] (7) Consider the minimum spanning tree constructed in (6) as the initial complex network structure, G =<V,E> , where V represents the set of points in the network, E represents the set of edges in the network, and the node set V = {P1, P2, ..., P m}, is a platform node, and the set of edges can be expressed as E = { <P i ,P j ,d ij >,… <P q ,P r ,d qr >}, where d ij is the edge weight, which is also the distance between the two platforms calculated in (6), forming a complex network structure with the strongest correlation between trading platforms.
[0104] At the same time, a susceptible state set S is established to store nodes in a healthy state, which belongs to the set of platforms that are not affected by the risk of abnormal trading patterns; an infected state set I is established to store nodes that are infected and have a certain chance of recovery, which belongs to the set of platforms that are affected by the risk of abnormal trading patterns; and a recovered state set R is established, that is, nodes that have recovered from the infection period, which belong to the platforms that are affected by the risk of abnormal trading patterns but adjust their trading patterns according to market rules and return to normal.
[0105] Initially, all nodes are in the susceptible (S) state, and it is assumed that:
[0106] Trading platforms exist for a long time, and the time scale affected by abnormal trading patterns is shorter than the platform's life cycle.
[0107] Risk propagation occurs along the connecting edges, that is, only when two nodes are connected by an edge will the platforms be affected by risks to each other.
[0108] The probability that each node contacts its neighbor node and infects it is consistent, and the probability that the infected node recovers is also consistent.
[0109] The total number of nodes does not change with the infection process, that is, at time t: S(t)+I(t)+R(t)=M.
[0110] The above assumptions constitute the risk propagation environment between trading platforms. The trading platform network is structured as an undirected complex network, with the SIR model as the basis of the risk propagation model. The susceptible nodes connected to the infected nodes have a probability of β (infection rate) to become infected, and the infected nodes have a probability of γ (cure rate) to become recovered, that is, the trading platform controls the risk through certain measures.
[0111] The density of each state at time t is expressed in lowercase letters as: Then the differential equation of the SIR model is expressed as follows:
[0112]
[0113] (8) All platform nodes are added to the susceptible state S set. Then, according to the screening results of abnormal transaction patterns in (4), assuming that an infectious disease occurs, that is, there are platforms with abnormal transaction patterns screened in (4), they are transferred from the susceptible state S set to the infected state set I.
[0114] (9) Traverse the trading platforms in the susceptible state set S. If there is an infected node among the adjacent nodes, that is, in the set I, then according to the infection rate, there is a certain probability that it will develop into an infected node. If infected, it will be removed from the S set and added to the set I.
[0115] (10) Traversing the trading platforms in the infected state set I, the infected platform nodes have a certain probability of changing to the recovery state. The cured nodes are removed from set I and added to set R.
[0116] (11) Iterate steps (9) to (10) until the number of infected nodes stops changing or the number of iterations reaches the maximum number of iterations τ. Figure 4As shown in the figure, at this time, the risk source platform is screened out, and the risk range and trading platforms on the path caused by it can also be predicted. Therefore, screening the risk source platform, finding the risk transmission path and possible infection platforms will help the regulatory authorities to establish and improve relevant systems, and also help trading platforms to prevent the risk infection that may be coming in advance.
[0117] It should be noted that the above content only illustrates the technical idea of the present invention and cannot be used to limit the protection scope of the present invention. For ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications all fall within the protection scope of the claims of the present invention.
Claims
1. A method for identifying risk platforms and predicting risk propagation in commodity trading markets, characterized by: The following steps are involved: (1) After the system is initialized, the database reads historical transaction data and builds a user transaction network; (2) Conduct network modeling of prohibited transaction patterns in the policy and summarize the power law distribution rules; (3) Use the least squares method to fit the power law distribution to the transaction account nodes and calculate the outliers; (4) Calculate the threshold value and screen out the trading patterns formed by trading accounts with outliers greater than the threshold value, which are abnormal trading patterns and the platforms they belong to are risk source platforms; (5) According to the rating matrix of the trading products by the platform, the correlation coefficient between the platforms is calculated, and finally the correlation coefficient matrix ρ between the trading platforms is obtained; (6) Taking the trading platforms as nodes of the tree, the minimum spanning tree is constructed based on the correlation coefficient matrix ρ between the trading platforms; (7) treating the minimum spanning tree constructed in step (6) as the initial complex network structure and establishing the SIR epidemic model; (8) Add all platform nodes to the susceptible state set S, and then transfer the risk source platform from the susceptible state set S to the infected state set I according to the screening results of abnormal transaction patterns in step (4); (9) Traverse the trading platforms in the susceptible state set S. If there is an infected node among the adjacent nodes, that is, in the set I, then according to the infection rate, there is a certain probability that it will develop into an infected node. If infected, it will be removed from the S set and added to the set I; (10) Traverse the trading platforms in the infected state set I. The infected platform nodes have a certain probability of changing to the recovery state. The cured nodes are removed from set I and added to set R; (11) Determine whether the number of infected nodes has changed compared with the previous iteration and whether the number of iterations is less than the maximum value. If so, return to step (9); otherwise, go to step (12); (12) After the risk source platform is identified and the risk infection scope and path are predicted, the risk source platform and its risk infection scope and path are adaptively displayed based on the results.
2. The method for identifying risk platforms and predicting risk propagation in a commodity trading market according to claim 1, characterized in that: In step (1), the trading account in the platform is defined as U = {u1,u2,…,u n }, a k Indicates the number of transactions, s k Indicates the type of traded goods, then each transaction can be written in the form of a four-tuple i ,u j ,a k ,s k >, where u i For the seller, u j For the buyer, a k Indicates the number of transactions, s k Indicates the type of traded goods; maps the transaction set in the commodity trading market into a directed graph model G = {V, E, W}, where the node set V represents all trading accounts involved in the commodity trading market, that is, V = {u1,u2,…,u n }, the edge set E is the set of all transaction behaviors, that is, E = { i ,u j ,a1,s1>,…, i ,u j ,a n ,s n >}, with transaction volume as the attribute of each edge, the directionality of the edge represents the buying and selling of the transaction, forming a directed weighted graph with multiple edges. 3. The method for identifying risk platforms and predicting risk propagation in a commodity trading market according to claim 1, characterized in that: In step (2), the abnormal transaction patterns specified in the policy are summarized for model building and feature extraction, which are expressed using the related structure of complex network nodes, and the corresponding three power law distribution laws are sorted out: Number of edges ~ number of neighbors: The total number of connected edges E and the number of neighbors N of most normal transaction nodes follow a power law distribution, that is, the logarithmic coordinates are linear: E i ~N i a ,1≤a≤2 Weight and number of edges: The total edge weight and number of edges of most normal transaction nodes follow a power law distribution: W i ~N i c ,c≥1 Main eigenvalue ~ weight sum: The main eigenvalue of the central network of the node and the central network weight satisfy a certain power law relationship: l i ~W i γ ,0.5≤γ≤1。 4. The method for identifying risk platforms and predicting risk propagation in the commodity trading market according to claim 3, characterized in that: In step (3), according to the summarized power law distribution law that satisfies the linear relationship, the linear regression model anomaly detection algorithm is used to detect abnormal transaction entities, the power law distribution is fitted by the least squares method, and the outlier (os i ) represents node u i The degree of deviation from the normal mode under a given power law distribution characteristic is formally expressed as That is, the distance between the predicted value and the actual value, to represent the node u i The difference between the observed value and the predicted value under a given power law distribution, node u i Outlier values of os i It is necessary to reflect the degree to which the node deviates from the normal mode; when the node u i The closer the observed value is to the predicted value, the better the node u i The closer the corresponding transaction behavior is to the majority of normal transaction behavior patterns, the lower the abnormal value associated with the node; otherwise, it means that the transaction pattern formed by the transaction account represented by the node and its related transaction parties is a possible abnormal transaction pattern.
5. The method for identifying risk platforms and predicting risk propagation in the commodity trading market according to claim 4, characterized in that: In step (4), it is assumed that the outliers of all nodes approximately satisfy a condition with mean μ and variance σ 2 The normal distribution N(μ,σ 2 ); select μ+3σ as the threshold of the anomaly score. If the anomaly value is greater than μ+3σ, the transaction pattern formed by the account group to which the account belongs is an abnormal transaction pattern, and the platform to which the account belongs is set as a risk platform.
6. The method for identifying risk platforms and predicting risk propagation in a commodity trading market according to claim 1, characterized in that: In step (5), the commodity scores are calculated based on the proportion of each commodity's transaction volume in the total transaction volume of the platform. Assume that on platform P i The total amount of commodity transactions is sum i , Product B j The transaction volume is b j , then product B j On platform P i The scores on are: Where α is a constant that controls the score within a certain range; thus, the platform-product score matrix R is obtained, and then the correlation between the two platforms is calculated based on the Pearson correlation coefficient, and finally the correlation coefficient matrix ρ between the trading platforms is obtained.
7. The method for identifying risk platforms and predicting risk propagation in the commodity trading market according to claim 6, characterized in that: In step (6), the trading platforms are used as nodes of the tree, and the minimum spanning tree is constructed based on the correlation between the trading platforms. The weights in the minimum spanning tree are defined by the distance between the correlation between the trading platforms. For example, the similarity between platform Pi and platform Pk is ρ ik , then the distance between the two trading platforms is: The distance matrix D between trading platforms is obtained, and the minimum spanning tree is constructed according to the Prim algorithm.
8. The method for identifying risk platforms and predicting risk propagation in the commodity trading market according to claim 7, characterized in that: In step (7), the minimum spanning tree constructed in step (6) is regarded as the initial complex network structure, G =<V,E> , where V represents the set of points in the network, E represents the set of edges in the network, and the node set V = {P1, P2, ..., P m }, is a platform node, and the set of edges can be expressed as E = { <P i ,P j ,d ij >,… <P q ,P r ,d qr >}, where d ij is the edge weight, which is also the distance between the two platforms calculated in (6), forming the most strongly correlated complex network structure between trading platforms; At the same time, a susceptible state set S is established to store nodes in a healthy state, which belongs to the set of platforms that are not affected by the risk of abnormal trading patterns; an infected state set I is established to store nodes that are infected and have a certain chance of recovery, which belongs to the set of platforms that are affected by the risk of abnormal trading patterns; and a recovered state set R is established, that is, nodes that have recovered from the infection period, which belong to the platforms that are affected by the risk of abnormal trading patterns but adjust their trading patterns according to market rules and return to normal.
9. The method for identifying risk platforms and predicting risk propagation in the commodity trading market according to claim 8, characterized in that: In step (8), all platform nodes are added to the susceptible state S set. Then, according to the screening results of abnormal transaction patterns in step (4), assuming that an infectious disease occurs, that is, there is a platform with abnormal transaction patterns screened in step (4), it is transferred from the susceptible state S set to the infected state set I.
10. The method for identifying risk platforms and predicting risk propagation in the commodity trading market according to claim 9, characterized in that: In step (9), the trading platforms in the susceptible state set S are traversed. If there is an infected node among the adjacent nodes, that is, in set I, then according to the infection rate, there is a certain probability that it will develop into an infected node. If infected, it will be removed from set S and added to set I.
11. The method for identifying risk platforms and predicting risk propagation in a commodity trading market according to claim 10, characterized in that: In step (10), the trading platforms in the infected state set I are traversed. The infected platform nodes have a certain probability of changing to a recovery state. The cured nodes are removed from set I and added to set R.
12. The method for identifying risk platforms and predicting risk propagation in a commodity trading market according to claim 1, characterized in that: In step (11), it is determined whether the number of infected nodes has changed compared with the previous iteration and whether the number of iterations is less than the maximum value. If so, iterate (9)-(10). Otherwise, the risk source platform is screened out at this time, and the risk range caused by it and the trading platforms on the path can also be predicted. Therefore, screening the risk source platform and finding the risk transmission path and possible infected platforms are conducive to the establishment and improvement of relevant systems by the regulatory authorities, and also help the trading platform to take precautions against the risk infection that may be suffered in advance.
Citation Information
Patent Citations
Risk dissemination information evaluation method and system, storage medium and computer equipment
CN112132441A
Self-adaptive recommendation method and system for bulk commodity transaction platform associated supervision range
CN112926999A
Graphical structure model-based method for transaction risk control, and device and equipment
WO2019114434A1
Bulk commodity transaction information sharing method and system, and storage medium
WO2022036505A1
