Public resource transaction big data analysis method based on heterogeneous data and intelligent portraits

Abstract

The invention relates to the technical field of neural network topology data processing, in particular to a public resource transaction big data analysis method based on heterogeneous data and intelligent portray.The method comprises the steps that the data of the public resource transaction big data are analyzed on the basis of the equity association strength between transaction subjects in the heterogeneous data, the equipment feature consistency and the spherical distance of the geographic position information of the transaction subjects; calculating a static space correlation degree between any two transaction subjects; according to the static space correlation degree and the Euclidean distance between the current feature vector and the historical feature mean vector, calculating a comprehensive risk index between transaction subjects in the current bid; the state transition of the transaction subject portrait label is judged through the comprehensive risk index, so that the accuracy of abnormal behavior recognition in public resource transaction can be effectively improved.

Description

Technical Field

[0001] This invention relates to the field of neural network topology data processing technology. More specifically, this invention relates to a big data analysis method for public resource transactions based on heterogeneous data and intelligent profiling. Background Technology

[0002] In the current regulatory landscape of public resource transactions, regulatory authorities face massive and complex transaction data, requiring real-time risk monitoring of various market participants. Traditional regulatory methods often rely on static data such as business background and qualification levels to construct credit profiles of transaction entities, focusing on analyzing the inherent logical connections between entities.

[0003] However, with the continuous upgrading of illegal methods, modern illegal groups exhibit extremely strong concealment and dynamic evolution characteristics. These illegal entities typically maintain a superficially independent operating status during non-trading periods, only conducting highly synchronized collaborative operations within a very short time window during specific bidding project cycles. This pattern of disguise in peacetime and surprise attacks in wartime has become a challenge for industry supervision.

[0004] In such complex real-world application scenarios, existing analytical technologies face severe challenges in solving the problem of accurate early warning. First, because transaction data is scattered across different systems and has heterogeneous formats, directly reading business data can easily lead to the loss of key temporal characteristics due to system concurrency latency, making it impossible for the regulatory system to accurately capture millisecond-level instantaneous coordinated actions. Second, relying solely on explicit connections such as equity stakes for investigation makes it difficult to uncover implicit physical connections, resulting in missed detections of shadow groups without equity ties. Ultimately, when facing temporarily formed, covert bid-rigging groups, the system struggles to accurately capture the qualitative change process of transaction entities transitioning from a safe state to a suspicious state, resulting in a high false alarm rate and an inability to achieve real-time blocking.

[0005] Therefore, how to accurately supervise public resource transactions based on big data is an urgent problem to be solved. Summary of the Invention

[0006] To address the aforementioned technical challenge of accurately regulating public resource transactions based on big data, this invention proposes a public resource transaction big data analysis method based on heterogeneous data and intelligent profiling. This method includes the following steps: Heterogeneous data is collected during the public resource transaction process. This data includes the business registration data of the transaction entities, the equipment characteristics of the uploaded bid documents, geographical location information, and transaction operation timestamps. Based on this heterogeneous data, the static spatial correlation between any two transaction entities is calculated. The static spatial correlation is positively correlated with the strength of equity association and the consistency of equipment characteristics between the transaction entities, and negatively correlated with the spherical distance of the geographical location information of the transaction entities. Feature vectors reflecting the instantaneous behavioral state of each bid are obtained, and a comprehensive risk index between the transaction entities in the current bid is calculated. The comprehensive risk index is positively correlated with the static spatial correlation, the Euclidean distance between the current feature vector and the historical feature mean vector, and negatively correlated with the absolute value of the difference in operation timestamps between the transaction entities. The state transition of the transaction entity profile label is determined by the comprehensive risk index.

[0007] This invention provides a big data analysis method for public resource transactions based on heterogeneous data and intelligent profiling, which can effectively improve the accuracy of identifying abnormal behavior of transaction entities during the bidding process. In the process of identifying anomalies, this invention collects and integrates heterogeneous data from multiple dimensions such as business registration, equipment, geography, and time, reducing the limitations of traditional supervision that relies solely on explicit logical connections such as equity ownership. This allows for the uncovering of shadowy groups with implicit physical connections, such as shared office locations and shared bidding equipment. Furthermore, by combining static spatial correlation with a dynamic comprehensive risk index, this invention can accurately capture the instantaneous coordinated actions of bid-rigging groups within massive amounts of transaction data. By monitoring changes in the comprehensive risk index, it achieves dynamic transitions in the transaction entity's profile label from a safe state to a suspicious state, effectively reducing the problems of delayed early warnings and high false alarm rates caused by data fragmentation and loss of temporal features in existing technologies, thereby effectively improving the accuracy of public resource transaction supervision.

[0008] According to the public resource transaction big data analysis method based on heterogeneous data and intelligent profiling provided by the present invention, the collection of heterogeneous data in the public resource transaction process includes: preprocessing the collected initial heterogeneous data, including at least using the mean imputation method to process missing values ​​in the initial heterogeneous data, using the Laida criterion to remove abnormal outliers in the initial heterogeneous data, and finally obtaining heterogeneous data.

[0009] This invention solves the common problems of missing values ​​and abnormal noise in the acquisition process of multi-source heterogeneous data. Through a standardized preprocessing process, it eliminates the impact of poor data quality caused by differences in formats of different business systems or transmission interference, and ensures the integrity and reliability of the input data for subsequent calculation models.

[0010] According to the public resource transaction big data analysis method based on heterogeneous data and intelligent profiling provided by the present invention, the method for obtaining the device characteristics of the uploaded bid documents includes: obtaining the MAC address and IP address of the terminal device of the transaction entity that uploaded the bid documents by accessing the background log of the bidding platform, as the corresponding device characteristics.

[0011] This invention can delve into the underlying hardware fingerprint level to identify the commonality between trading entities. By extracting unique identifiers such as MAC addresses and IP addresses, it can effectively identify physical traces of different trading entities that appear to have independent qualifications but actually use the same terminal device to collude in bidding. This prevents violators from circumventing supervision through simple business registration changes or qualification isolation, and enhances the ability to penetrate and identify hidden physical connections.

[0012] According to the public resource transaction big data analysis method based on heterogeneous data and intelligent profiling provided by the present invention, the method for obtaining the equity association strength includes: calculating the cross-shareholding ratio between the transaction entities through an equity penetration algorithm in the business registration data of the transaction entities, and using the cross-shareholding ratio as the equity association strength.

[0013] According to the public resource transaction big data analysis method based on heterogeneous data and intelligent profiling provided by the present invention, the step of calculating the static spatial correlation degree between any two transaction entities based on heterogeneous data includes: ; For the first The and the first The static spatial correlation between individual trading entities For the first The and the first The strength of equity connections between the transaction entities For the first The and the first The spherical distance between individual trading entities Let be the distance normalization constant. For the first The and the first Consistency of equipment characteristics among the trading entities , , These are the weighting coefficients for equity linkage strength, spherical distance, and equipment characteristic consistency, respectively. For An exponential function with base 0.

[0014] This invention provides a precise method for calculating static spatial correlation. By mapping geographical distance, hardware features, and equity relationships to a unified mathematical index, and using a negative exponential function to handle geographical distance, it conforms to the objective law that the greater the physical distance, the weaker the correlation. Based on this, the comprehensive tightness of the transaction entities in physical and logical space can be accurately calculated, providing an evaluation standard for subsequent identification of abnormal remote collaboration or close collusion.

[0015] According to the public resource transaction big data analysis method based on heterogeneous data and intelligent profiling provided by the present invention, the method for obtaining the consistency of equipment characteristics includes: if the equipment characteristics of the transaction entities are consistent, then the consistency of the equipment characteristics between the transaction entities is set to 1, otherwise it is set to 0.

[0016] According to the public resource transaction big data analysis method based on heterogeneous data and intelligent profiling provided by the present invention, the step of obtaining the feature vector reflecting the instantaneous behavioral state of each bid includes: taking the implementation location code of the current bid segment of the transaction entity as the regional component, taking the deviation rate of the bid price as the monetary component, taking the business classification code of the bid project as the industry component, and taking the duration of bid preparation as the timeliness component; and normalizing the regional component, monetary component, industry component, and timeliness component respectively to obtain the feature vector corresponding to each bid of the transaction entity.

[0017] This invention comprehensively characterizes the bidding behavior of trading entities from multiple dimensions such as regional preferences, pricing strategies, business scope, and production timeliness. It transforms abstract bidding actions into measurable numerical vectors, thereby sensitively detecting abnormal and sudden behaviors that deviate from the entity's daily business practices (such as suddenly bidding across industries or regions), providing rich and standardized feature inputs for subsequent anomaly detection models.

[0018] According to the public resource transaction big data analysis method based on heterogeneous data and intelligent profiling provided by the present invention, the calculation of the comprehensive risk index among the transaction entities in the current bidding includes: ; The first in the current bidding The and the first A comprehensive risk index among the trading entities. , The first The or the first Feature vectors of each trading entity , The first The or the first The historical characteristic mean vector of each trading entity For the first The and the first The static spatial correlation between individual trading entities It is a natural constant. For the first The and the first The absolute value of the difference in operation timestamps between individual transaction entities. Let be the time normalization constant. The L2 norm symbol, For Logarithmic function with base 0. , These are the distance weight and the time weight, respectively.

[0019] According to the public resource transaction big data analysis method based on heterogeneous data and intelligent profiling provided by the present invention, the step of determining the state transition of the transaction subject's profile label through a comprehensive risk index includes: constructing a profile update mechanism based on state transition energy logic, wherein the state transition energy is obtained in the following way: ; The first in the current bidding The and the first Energy for state transfer between trading entities For the first The and the first The rate of change of the comprehensive risk index of each trading entity The sampling period is For step activation function, The first in the current bidding The and the first A comprehensive risk index among the trading entities. For the first time in an adjacent sampling period The and the first The overall risk index difference between individual trading entities The preset threshold is used; in response to the state transition energy between trading entities exceeding the energy threshold, the profile label of the trading entity is updated to a suspected state.

[0020] According to the public resource transaction big data analysis method based on heterogeneous data and intelligent profiling provided by the present invention, the method for obtaining the preset threshold includes: extracting the upper bound of the 95% confidence interval of the comprehensive risk index in the historical violation case library as the preset threshold.

[0021] The present invention has the following beneficial effects: Based on the above technical solutions, the public resource transaction big data analysis method based on heterogeneous data and intelligent profiling provided by this invention collects and integrates heterogeneous data from multiple dimensions such as business registration, equipment, geography, and time. This reduces the limitations of traditional supervision that relies solely on explicit logical connections such as equity, thereby uncovering shadow groups with implicit physical connections such as shared office locations and shared bidding equipment. Furthermore, by combining static spatial correlation with a dynamic comprehensive risk index, this invention can accurately capture the instantaneous coordinated actions of bid-rigging groups within massive transaction data. By monitoring changes in the comprehensive risk index, it achieves dynamic transitions in the transaction entity profile label from a safe state to a suspicious state, effectively reducing the problems of delayed early warnings and high false alarm rates caused by data fragmentation and loss of temporal characteristics in existing technologies, thus effectively improving the accuracy of public resource transaction supervision. Attached Figure Description

[0022] Figure 1 A flowchart illustrating the steps of a public resource transaction big data analysis method based on heterogeneous data and intelligent profiling provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of a public resource transaction big data analysis architecture provided in an embodiment of the present invention; Figure 3 A schematic diagram illustrating the relationship between transaction entities provided in an embodiment of the present invention; Figure 4 This invention provides a schematic diagram of a static spatial correlation heatmap of transaction entities. Figure 5 This is a schematic diagram of a three-dimensional surface plot of state transition energy provided in an embodiment of the present invention. Detailed Implementation

[0023] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0024] Please see Figure 1 , Figure 1 This is a flowchart illustrating the steps of a public resource transaction big data analysis method based on heterogeneous data and intelligent profiling provided in an embodiment of the present invention. The method includes the following steps: S1: Collect heterogeneous data during the public resource transaction process and perform feature preprocessing.

[0025] The heterogeneous data includes the business registration data of the transaction entities, the characteristics of the equipment used to upload the bidding documents, geographical location information, and transaction operation timestamps.

[0026] It should be noted that modern violations exhibit strong dynamic evolution and temporal coordination, with violators typically demonstrating a high degree of temporal synchronization and spatial overlap only within a specific bidding project cycle. This invention, by executing an independent data acquisition program, ensures that data scattered across different business systems avoids data quality issues caused by format differences, inconsistent sampling frequencies, and varying storage media, thereby reducing the possibility of severe analytical biases in subsequent calculation models.

[0027] Specifically, if business database data is read directly in real time, latency or table locking issues caused by high-concurrency access may lead to packet loss of time-series data, causing dynamic collaborative analysis to lose its accurate time reference. Therefore, embodiments of the present invention can clean and standardize heterogeneous data through a dedicated channel, thereby ensuring the authenticity of the transaction entity profile.

[0028] It is understandable that in the bidding process of large-scale projects, the overall project is divided into several independent parts according to profession, region, or construction stage, and each independent part is a bidding section. Each bidding section is tendered separately to improve efficiency and optimize resource allocation. This embodiment of the invention uses one of the bidding sections as the current bidding analysis target for processing. The transaction subject in the public resource transaction process is the individual bidding unit. Since some individual bidding units have few bidding records, for ease of processing, this embodiment of the invention can use individual bidding units with no less than 3 bidding records among all individual bidding units as the transaction subject.

[0029] Furthermore, this embodiment of the invention acquires four types of core data through a categorized data collection mechanism. Specifically, it can obtain the business registration data of the transaction entity by calling government service interfaces; this business registration data may include the shareholding data of the transaction entity. It can also obtain the terminal device from which the transaction entity uploaded its bid documents, such as the MAC address and IP address of the terminal device, by accessing the backend logs of the bidding platform. Finally, it can obtain the geographical location information of the transaction entity, such as the latitude and longitude of its office address, through a geographic information system. Finally, it can collect the transaction operation timestamps through a time-series database.

[0030] After obtaining the initial heterogeneous data based on the above steps, preprocessing procedures can be performed. Specifically, missing values ​​in the initial heterogeneous data can be handled using the mean imputation method; outliers in the initial heterogeneous data can be removed using the Laida criterion, ultimately resulting in high-quality heterogeneous data. The specific preprocessing method can be set according to actual needs.

[0031] For details, please refer to Figure 2 As shown, Figure 2This is a schematic diagram of a big data analysis architecture for public resource transactions provided in an embodiment of the present invention. It allows for the construction of profile tags for each transaction entity using big data technology modules, resulting in a transaction entity profile database. There are two types of profile tags: a safe status corresponding to normal operation and a suspected status corresponding to collaborative violations.

[0032] Figure 3 This diagram illustrates the relationship between trading entities according to an embodiment of the present invention, where a rectangular area represents one trading entity. As shown in the diagram, during the bidding process, trading entities may engage in joint bidding activities. For example, trading entity 5 may jointly bid with trading entities 0, 1, 2, 3, and 4.

[0033] Thus, by classifying and collecting heterogeneous data from multiple sources and standardizing preprocessing, the embodiments of the present invention can effectively solve the problems of data dispersion and time-series packet loss, thereby providing an accurate data foundation for subsequently constructing the underlying topological association strength between transaction entities.

[0034] S2: Based on heterogeneous data, calculate the static spatial correlation between any two trading entities.

[0035] Among them, the static spatial correlation is positively correlated with the strength of equity association and the consistency of equipment characteristics between the transaction entities, and negatively correlated with the spherical distance of the geographical location information of the transaction entities.

[0036] It should be noted that after completing data collection based on the above steps, it is necessary to construct the underlying topological relationship strength between the transaction entities. In practice, illegal connections are often hidden at the intersection of multiple dimensions. However, the transaction entity profile can only identify explicit equity holding relationships, ignoring implicit physical connections such as the same office location or data uploaded from the same device. By mapping multi-dimensional features such as geography, equipment, and equity to unified quantitative indicators, the regulatory system can sense the physical distance between entities, thereby enabling accurate profile analysis.

[0037] Based on this, embodiments of the present invention can transform abstract implicit relationships into specific weights by comprehensively considering equity penetration, geographical spherical distance, and hardware feature consistency.

[0038] Specifically, since the overlap of physical spaces is usually the physical basis for coordinated violations, while the correlation of logical spaces is the source of the motivation for violations, when two independent trading entities exhibit a high degree of overlap in both physical and logical spaces, their static spatial correlation will show a non-linear increase, thus providing a logical basis for subsequent identification of abnormal actions.

[0039] Optionally, in this embodiment of the invention, the static spatial correlation degree between any two trading entities is calculated, as shown in the following formula: ; For the first The and the first The static spatial correlation between individual trading entities For the first The and the first The strength of equity connections between the transaction entities For the first The and the first The spherical distance between individual trading entities Let be the distance normalization constant. For the first The and the first Consistency of equipment characteristics among the trading entities , , These are the weighting coefficients for equity linkage strength, spherical distance, and equipment characteristic consistency, respectively. For An exponential function with base 0.

[0040] The strength of equity association is obtained by calculating the cross-shareholding ratio between the two parties using an equity penetration algorithm in the business registration data.

[0041] The Havesing formula can be used to calculate the spherical distance between the trading entities based on the latitude and longitude of the collected office addresses.

[0042] The distance normalization constant is used to eliminate the influence of the dimension of physical distance. Optionally, in this embodiment of the invention, it can be set to 100km.

[0043] Specifically, if the device characteristics (MAC address and / or IP address) are identical, the device characteristic consistency is set to 1; otherwise, it is set to 0. Therefore, if either the MAC address or the IP address is the same, the device characteristic consistency can be directly set to 1.

[0044] When setting the weight coefficients for equity association strength, spherical distance, and equipment characteristic consistency, since the probability of different transaction entities using the same equipment characteristic is small, the weight coefficient for equipment characteristic consistency can be greater than the weight coefficients for equity association strength and spherical distance.

[0045] Optionally, in this embodiment of the invention, the weighting coefficients for equity association strength, spherical distance, and equipment feature consistency can be set to 0.3, 0.3, and 0.4, respectively, and can be set according to actual needs.

[0046] For example, in an embodiment of the present invention, the ratio of the number of joint bids among the current trading entities to the number of joint bids among all trading entities in the current bid can also be used as the weighting coefficient for the consistency of the equipment characteristics of the current trading entity.

[0047] In this relation, Item and Each item is treated as an additive item. Since the closer the equity relationship or the more consistent the equipment characteristics, the higher the logical and physical overlap between the entities, therefore… As it increases, a positive correlation is observed.

[0048] The greater the distance between the transacting parties, the weaker the connection may be. (Intermediate term) A negative exponential function was used to handle physical distances. Due to the monotonically decreasing property of the exponential function, when dealing with spherical distances... As the value increases, the score for this item will rapidly decrease, leading to a final result... The decrease in data value indicates a negative correlation, which is consistent with the objective law that the correlation weakens with increasing physical distance.

[0049] For example: Setting The distance is 100km. Transaction entity. and The cross-shareholding ratio is 0.2, and the equipment characteristics of the transaction entities are inconsistent. The spherical distance between the two is 10 km. Substituting into the above formula, we can calculate: .

[0050] If the distance is shortened to 1km, then This shows that the value of static spatial correlation increases as the distance decreases.

[0051] See also Figure 4 As shown, Figure 4 This is a schematic diagram of the static spatial correlation of transaction entities provided in an embodiment of the present invention. The diagram shows the static spatial correlation between all transaction entities. For transaction entities with high static spatial correlation, although they have no explicit relationship at the business level, they may have a high degree of overlap in the latitude and longitude of their office addresses or the fingerprints of the devices used to upload tender documents, requiring close monitoring.

[0052] Thus, through the fusion calculation of multi-dimensional spatial weights, the embodiments of the present invention can accurately assess the implicit physical and logical connections between trading entities, thereby providing a logical basis for subsequent identification of abnormal dynamic collaborative actions.

[0053] S3: Obtain the feature vector reflecting the instantaneous behavioral state of the transaction in each bid, calculate the comprehensive risk index among the trading entities in the current bid. The comprehensive risk index is positively correlated with the static spatial correlation degree, the Euclidean distance between the current feature vector and the historical feature mean vector, and negatively correlated with the absolute value of the difference between the operation timestamps of the trading entities.

[0054] For example, when obtaining the historical feature mean vector of a transaction entity, all completed bidding records of the transaction entity can be obtained within a preset historical period (e.g., 24 months), and the corresponding feature vectors can be extracted to form a historical sample set. The historical sample set is then processed using mean clustering logic to obtain the historical feature mean vector corresponding to each transaction entity. The historical feature mean vector represents the centroid of the transaction entity's behavioral profile in the past historical period. The historical feature mean vector includes regional components, monetary components, industry components, and time-sensitivity components.

[0055] It's important to note that after establishing the static background, a time dimension needs to be introduced for two-way monitoring. Since the true profile of violations evolves dynamically, the system may be unable to distinguish between long-term, stable partners and temporary bid-rigging groups organized for specific projects. Therefore, it's crucial to organically combine the historical behavioral patterns of individuals with the dynamic interactions between groups. Vertically, observe whether individuals deviate from their historical patterns; horizontally, observe whether groups exhibit high levels of coordination within a narrow time window. This allows the profile to capture sudden motives for violations. Only when individuals who don't usually interact suddenly perform actions within a very short time window, and these actions significantly deviate from their usual business scope (such as cross-industry or cross-regional bidding), can such actions be identified as abnormal bidding behavior.

[0056] Based on this, embodiments of the present invention can analyze the instantaneous behavioral state of trading entities by constructing feature vectors and calculate a comprehensive risk index by combining static correlation.

[0057] Furthermore, we first obtain the feature vector corresponding to each bid by the transaction entity. We take the implementation location code of the current bid section as the regional component, the bid price deviation rate as the monetary component, the business classification code as the industry component, and the bid preparation time as the timeliness component, and then perform normalization processing on each of them to obtain the feature vector corresponding to each bid by the transaction entity.

[0058] The bid price deviation rate is the difference between the actual total bid price submitted by the trading entity in the bid documents for the current bid section and the bid control price published in the bidding documents for that bid section.

[0059] For example, the normalization method can be max-min normalization, and the specific method can be set according to actual needs.

[0060] Optionally, in this embodiment of the invention, the comprehensive risk index among the trading entities in the current bid is calculated, as shown in the following formula: ; The first in the current bidding The and the first A comprehensive risk index among the trading entities. , The first The or the first Feature vectors of each trading entity , The first The or the first The historical characteristic mean vector of each trading entity For the first The and the first The static spatial correlation between individual trading entities It is a natural constant. For the first The and the first The absolute value of the difference in operation timestamps between individual transaction entities. Let be the time normalization constant. The L2 norm symbol, For Logarithmic function with base 0. , These are the distance weight and the time weight, respectively.

[0061] The distance weight is used to capture individuals exhibiting abnormal behavior across dimensions, while the time weight is used to identify individuals exhibiting abnormal clustering over a short period. Optionally, in this embodiment of the invention, both the distance weight and the time weight can be set to 0.5, and can be adjusted according to actual needs.

[0062] In the above formula, the L2 norm is used to calculate the Euclidean distance between the current feature vector and the historical feature mean vector to characterize the deviation of individual behavior. The time normalization constant can optionally be set to 60s, which can be set according to actual needs.

[0063] It is the first The and the first The mean deviation of each trading entity is used to measure the geometric distance between the trading entity and its historical center of gravity. For example, the mean deviation of the first trading entity... If the historical feature mean vector of a transaction entity indicates that the industry component corresponding to that transaction entity is the decoration and renovation industry, and the geographical component has been active locally for a long time, then the numerical distribution of its feature vector will be relatively concentrated. The larger the value, the greater the deviation from the historical mean (the greater the Euclidean distance), indicating a cross-dimensional deviation in the business behavior of the trading entity in the current bidding process, and a greater likelihood of risk. It will also increase, showing a positive correlation.

[0064] As a collaborative item, due to the time difference in operation... Located in the logarithmic term of the denominator, as the time difference increases, the denominator... As it increases, the overall value of the cooperating terms decreases, ultimately leading to The risk index decreases, showing a negative correlation. This means that a non-linear peak will only appear when there is coordinated action within a short period of time, accompanied by abnormal individual behavior.

[0065] For example: Let If the mean deviation between the two trading entities is 0.8, the static correlation degree... The time difference is 10 seconds (assuming...) Substituting into the above formula, we can obtain: .

[0066] Thus, by deeply integrating time-series slicing with individual deviation correction, this embodiment of the invention can effectively filter compliant industry-wide common behaviors, thereby accurately capturing the instantaneous risk fluctuations of entities engaging in illegal transactions.

[0067] S4: Determine the status transition of the transaction entity's profile label through the comprehensive risk index.

[0068] It's important to note that risk monitoring cannot merely focus on numerical fluctuations; it needs to develop clear classification and judgment mechanisms, transforming computational burdens into state transitions of profile tags. By determining the state transitions of transaction entity profile tags through a comprehensive risk index, the impact of massive numerical fluctuations can be reduced, leading to effective action orders. By executing judgments based on energy changes, the system can record the process of an entity transitioning from a normal operating state to a state of coordinated violation, thereby ensuring a closed-loop chain of evidence.

[0069] Based on this, embodiments of the present invention can construct a profile update mechanism through state transition energy logic, and use energy triggering to shield against minor random disturbances.

[0070] Furthermore, when determining the state transition of a trading entity's profile label using the comprehensive risk index, a profile update mechanism can be constructed based on the state transition energy logic of the comprehensive risk index. If the state transition energy obtained based on the cumulative amount or instantaneous growth rate of the comprehensive risk index exceeds the corresponding physical threshold, the trading entity's profile will transition from a safe state to a suspicious state. This ensures that a state reversal can only be triggered when the trading entity exhibits a continuous and strong tendency to coordinate deviations, thereby effectively shielding against minor random disturbances and accurately capturing hidden groups launching time-series attacks at the last moment.

[0071] Specifically, if the overall risk index among the trading entities is high but the rate of change is low, it may indicate normal fluctuations in long-term related enterprises, and the situation remains safe, but continued monitoring is warranted. If the overall risk index among the trading entities is low but the rate of change is high, it may indicate normal business rhythm adjustments by low-risk trading entities, requiring no intervention. If the overall risk index among the trading entities is both high and the rate of change is high, it indicates potential abnormal risks, triggering a transition from a safe state to a suspected state.

[0072] Optionally, in this embodiment of the invention, the state transition energy can be determined by referring to the following relationship: ; The first in the current bidding The and the first Energy for state transfer between trading entities For the first The and the first The rate of change of the comprehensive risk index of each trading entity The sampling period is For step activation function, The first in the current bidding The and the first A comprehensive risk index among the trading entities. For the first time in an adjacent sampling period The and the first The overall risk index difference between individual trading entities This is a preset threshold.

[0073] The preset threshold can be obtained by extracting the upper bound of the 95% confidence interval of the comprehensive risk index from the historical violation case database. The specific threshold can be set according to actual needs. The sampling period can also be set according to actual needs.

[0074] In the relation, the first The and the first The rate of change of the comprehensive risk index of each trading entity represents the speed of risk growth; the larger the value, the greater the risk. The and the first The more intense the collusion tendency of individual trading entities in the current bidding process becomes in a short period of time, the more the state transition energy increases synchronously as the risk index grows, exhibiting a positive correlation, due to the multiplier effect.

[0075] Step function It serves as a threshold protection mechanism. By calculating the difference between the current comprehensive risk index and the corresponding threshold, the item is set to 1 only when the rate of change of the comprehensive risk index exceeds the preset threshold; otherwise, it is set to 0. By directly reducing the state transition energy to zero, the low-risk fluctuation item can be zeroed out. The kinetic energy of the risk change rate is released only when the comprehensive risk index exceeds the abnormal limit, thereby triggering the state transition judgment to ensure that it does not react to small fluctuations in the low-risk area.

[0076] For example, in an embodiment of the present invention, in response to the state transition energy between transaction entities being higher than the energy threshold, the system automatically updates the profile label of the transaction entity to a suspected state.

[0077] The energy threshold can be determined through historical sample distribution analysis. Specifically, when determining the energy threshold, historical data on confirmed bid-rigging and collusion cases can be retrieved to obtain the peak energy of state transitions within one hour before the bid closing time; historical data on normal bids can be obtained to determine their state transition energy data; and an ROC curve can be constructed based on the peak energy of state transitions and the state transition energy data to finally obtain the corresponding energy threshold. The specific settings can be configured according to actual needs, and this embodiment of the invention does not impose excessive limitations.

[0078] See also Figure 5 As shown, Figure 5 This is a schematic diagram of a three-dimensional surface plot of state transition energy provided in an embodiment of the present invention. As shown in the figure, the main body of the surface plot remains flat, close to the zero energy region. However, at specific coordinate points of the trading entities, such as around coordinates 20-30, the comprehensive risk index of the trading entities before the closing price shows a sudden change in slope, which activates the step function and causes multiple extremely high vertical energy peaks to appear abruptly. Based on this, abnormal collusion behavior between trading entities can be accurately captured.

[0079] Thus, by dynamically determining the state transition energy, this embodiment of the invention can effectively capture hidden groups that launch timing attacks at the last moment, thereby achieving automated updating of profile tags and accurate early warning.

[0080] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A big data analysis method for public resource transactions based on heterogeneous data and intelligent profiling, characterized in that: include: Collect heterogeneous data during the public resource transaction process. The heterogeneous data includes the business registration data of the transaction entities, the characteristics of the equipment used to upload the bid documents, geographical location information, and transaction operation timestamps. Based on heterogeneous data, the static spatial correlation degree between any two transaction entities is calculated. The static spatial correlation degree is positively correlated with the equity correlation strength and equipment characteristic consistency between the transaction entities, and negatively correlated with the spherical distance of the geographical location information of the transaction entities. Obtain the feature vector reflecting the instantaneous behavioral state of the transaction in each bid, calculate the comprehensive risk index among the trading entities in the current bid, and the comprehensive risk index is positively correlated with the static spatial correlation degree, the Euclidean distance between the current feature vector and the historical feature mean vector, and negatively correlated with the absolute value of the difference between the operation timestamps of the trading entities. The status transition of the transaction entity's profile label is determined by the comprehensive risk index.

2. The public resource transaction big data analysis method based on heterogeneous data and intelligent profiling according to claim 1, characterized in that, The collection of heterogeneous data during the public resource transaction process includes: The initial heterogeneous data is preprocessed, including at least using the mean imputation method to handle missing values ​​in the initial heterogeneous data and using the Laida criterion to remove outliers in the initial heterogeneous data, and finally obtaining the heterogeneous data.

3. The public resource transaction big data analysis method based on heterogeneous data and intelligent profiling according to claim 2, characterized in that, The methods for obtaining the device characteristics for uploading bid documents include: By accessing the backend logs of the bidding platform, the MAC address and IP address of the terminal device on which the trading entity uploaded the bidding documents can be obtained as the corresponding device characteristics.

4. The public resource transaction big data analysis method based on heterogeneous data and intelligent profiling according to claim 1, characterized in that, The methods for obtaining the equity linkage strength include: The cross-shareholding ratio between the transaction entities is calculated using an equity penetration algorithm based on the business registration data of the transaction entities, and this cross-shareholding ratio is used as the strength of equity association.

5. The public resource transaction big data analysis method based on heterogeneous data and intelligent profiling according to claim 1, characterized in that, The calculation of the static spatial correlation between any two trading entities based on heterogeneous data includes: ； For the first The and the first The static spatial correlation between individual trading entities For the first The and the first The strength of equity connections between the transaction entities For the first The and the first The spherical distance between individual trading entities Let be the distance normalization constant. For the first The and the first Consistency of equipment characteristics among the trading entities , , These are the weighting coefficients for equity linkage strength, spherical distance, and equipment characteristic consistency, respectively. For An exponential function with base 0.

6. The public resource transaction big data analysis method based on heterogeneous data and intelligent profiling according to claim 5, characterized in that, Methods for obtaining device characteristic consistency include: If the device characteristics of the trading entities are consistent, then the consistency of device characteristics between the trading entities is set to 1; otherwise, it is set to 0.

7. The public resource transaction big data analysis method based on heterogeneous data and intelligent profiling according to claim 1, characterized in that, The process of obtaining the feature vector reflecting the instantaneous behavioral state of each bid includes: The implementation location code of the transaction entity in the current bidding segment is used as the regional component, the deviation rate of the bid price is used as the monetary component, the business classification code of the bidding project is used as the industry component, and the duration of bid preparation is used as the timeliness component. The regional component, monetary component, industry component, and timeliness component are normalized respectively to obtain the feature vector corresponding to each bid of the transaction entity.

8. The public resource transaction big data analysis method based on heterogeneous data and intelligent profiling according to claim 1, characterized in that, The calculation of the comprehensive risk index among the trading entities in the current bidding includes: ； The first in the current bidding The and the first A comprehensive risk index among the trading entities. , The first The or the first Feature vectors of each trading entity , The first The or the first The historical characteristic mean vector of each trading entity For the first The and the first The static spatial correlation between individual trading entities It is a natural constant. For the first The and the first The absolute value of the difference in operation timestamps between individual transaction entities. Let be the time normalization constant. The L2 norm symbol, For Logarithmic function with base 0. , These are the distance weight and the time weight, respectively.

9. The public resource transaction big data analysis method based on heterogeneous data and intelligent profiling according to claim 1, characterized in that, The process of determining the status transition of the transaction entity profile label through a comprehensive risk index includes: A profile update mechanism is constructed based on state transition energy logic, wherein the state transition energy is obtained in the following way: ; The first in the current bidding The and the first Energy for state transfer between trading entities For the first The and the first The rate of change of the comprehensive risk index of each trading entity The sampling period is For step activation function, The first in the current bidding The and the first A comprehensive risk index among the trading entities. For the first time in an adjacent sampling period The and the first The overall risk index difference between individual trading entities The preset threshold is used; in response to the state transition energy between trading entities exceeding the energy threshold, the profile label of the trading entity is updated to a suspected state.

10. The public resource transaction big data analysis method based on heterogeneous data and intelligent profiling according to claim 9, characterized in that, The method for obtaining the preset threshold includes: Based on the distribution of the comprehensive risk index in the historical violation case database, the upper bound of its 95% confidence interval is extracted as the preset threshold.

Images