Supplier potential bid surrounding and bid stringing analysis method and device and medium

Abstract

The invention provides a method and a device for analyzing potential bid enclosing and bid stringing of suppliers and a medium. The method comprises the following steps: collecting historical bidding and tendering data of a supplier and an association relationship data sample; determining network parameters of the first artificial neural network ANN1 by using the sample data; a supplier bidding file in a single purchase project is collected and structured; determining network parameters of a second artificial neural network ANN2 by using the bidding file structured data; calculating the potential bidding probability of the supplier combination; feature level fusion based on an artificial neural network is executed on the basis of the supplier combination potential bid surrounding and bid matching probability; decision level fusion based on Bayesian is executed; and according to the decision rule, carrying out supplier potential bid and confusion identification. According to the method and the device for analyzing the potential bid and consignment of the supplier and the medium provided by the invention, the reliability and the accuracy of the potential bid and consignment analysis of the supplier based on artificial intelligence and Bayesian fusion can be improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of artificial intelligence, and particularly relates to a method, device and medium for analyzing potential bid rigging and collusion among suppliers. Background Art

[0002] The bidding and tendering system is a relatively perfect trading method under the market economy system. Since the implementation of the bidding and tendering system in China, it has played a crucial role in the trading of engineering projects and has played a positive role. How to effectively identify bid rigging and collusion phenomena by using the rapidly developing big data technology is the key and difficult point in the bidding and tendering process; in addition, after identifying the phenomenon, what measures should be taken to control bid rigging and collusion and maintain the healthy development of bidding and tendering has also become a new consideration point at the present stage. Summary of the Invention

[0003] The purpose of the present invention is to provide a method, device and medium for analyzing potential bid rigging and collusion among suppliers, which can improve the reliability and accuracy of analyzing potential bid rigging and collusion among suppliers based on the fusion of artificial intelligence and Bayesian.

[0004] To achieve the above object, in a first aspect, the present invention provides a method for analyzing potential bid rigging and collusion among suppliers, the method comprising:

[0005] Collecting historical bidding and tendering data and associated relationship data samples of suppliers;

[0006] Constructing a first artificial neural network ANN1, and using the sample data to determine the network parameters of the first artificial neural network ANN1;

[0007] Collecting the bidding documents of suppliers in a single procurement project and structuring them;

[0008] Constructing a second artificial neural network ANN2, and using the structured data of the bidding documents to determine the network parameters of the second artificial neural network ANN2;

[0009] Calculating the potential bid rigging and collusion probability of supplier combinations;

[0010] Performing feature-level fusion based on artificial neural networks based on the potential bid rigging and collusion probability of supplier combinations;

[0011] Performing decision-level fusion based on Bayesian;

[0012] Identifying potential bid rigging and collusion among suppliers according to decision rules.

[0013] In some embodiments, collecting historical bidding and tendering data and associated relationship data samples of suppliers includes:

[0014] Collecting historical bidding and tendering data of suppliers, associated relationship data samples of suppliers, and bidding documents of suppliers for projects.

[0015] Collect the historical bidding behavior characteristics of suppliers and the characteristics of suppliers' project bidding documents;

[0016] Calculate the historical combined bidding frequency x of suppliers based on the historical bidding behavior characteristics of suppliers 1，1 , the combined supplier association relationship x 1，2 and the combined supplier joint registration confidence x 1，3 ;

[0017] Use the bidding document machine code detection device to collect the unique machine characteristics used by suppliers for bidding, and record whether there is a potential behavior of bid rigging and collusion in the supplier combination in the procurement sourcing R i ;

[0018] According to R i classify the potential probability of bid rigging and collusion of the bidding supplier combination according to the probability;

[0019] Compare the potential bid rigging and collusion situation of the supplier combination i with the obtained classification situation. When the two are the same, the data on the potential bid rigging and collusion of the supplier combination i is stored as valid sample data in the sample dataset, otherwise, it is excluded as invalid data.

[0020] In some embodiments, construct the first artificial neural network ANN1, and use the sample data to determine the network parameters of the first artificial neural network ANN1, including:

[0021] Initialization of ANN-1 network parameters;

[0022] Select a sample data of the historical bidding behavior characteristics of a supplier from the sample dataset;

[0023] Take as the input vector of ANN-1, and take as the expected output vector of ANN-1 and input it into ANN-1;

[0024] Define N = 1, w 1 i，j (1) is the connection weight between the input layer and the hidden layer of the first iteration of ANN- 1 1, θ 1 j (1) is the connection threshold between the input layer and the hidden layer of the first iteration of ANN- 1 1, v 1 j，t (1) is the connection weight between the hidden layer and the output layer of the first iteration of ANN-1, γ 1 t (1) is the connection threshold between the hidden layer and the output layer of the first iteration of ANN-1;

[0025] Execute the forward propagation process of the input mode;

[0026] Execute the backward propagation process of the error;

[0027] Calculate the new connection weight w 1 i，j (N + 1) and the new threshold θ 1 j (N + 1);

[0028] Calculate the new connection weight v 1 j，t (N + 1) and the new threshold γ 1 t (N + 1);

[0029] Judge Whether it is less than or equal to ε 1 , or whether N + 1 is equal to n M , if satisfied, the training ends and the next step is executed, otherwise, N = N + 1, and the forward propagation process of the input mode is executed again;

[0030] When the above conditions are met, use the connection weights and thresholds obtained from training as the optimal network parameters of ANN-1.

[0031] In some embodiments, calculating the potential probability of bid rigging and collusion among supplier combinations includes:

[0032] Collect the historical bidding behavior characteristics of suppliers and vehicle behavior characteristic signals of the supplier combination, denoted as and

[0033] According to Calculate the historical combined bidding frequency of suppliers Probability of supplier combination association relationship and confidence of joint registration of supplier combination

[0034] According to Calculate the IP address characteristics of the uploaded bidding documents Similarity of bidding texts and abnormal consistency of bidding texts

[0035] In some embodiments, based on the potential probability of bid rigging and collusion among supplier combinations, perform feature-level fusion based on artificial neural networks, including: feature-level fusion identification of ANN-1 and feature-level fusion identification of ANN-2.

[0036] In some embodiments, Bayesian-based decision-level fusion is performed, including:

[0037] Normalize the output of the artificial neural network;

[0038] Basic probability assignment of the evidence theory recognition framework;

[0039] Decision-level fusion based on D-S evidence theory.

[0040] In some embodiments, the expression of the decision rule is:

[0041]

[0042] where ε T1 and ε T2 are set thresholds.

[0043] To achieve the above object, in a second aspect, the present invention also provides a device for analyzing potential bid rigging of suppliers, and the device includes:

[0044] A first acquisition module for acquiring historical bidding data and associated relationship data samples of suppliers;

[0045] A first construction module for constructing a first artificial neural network ANN1 and determining the network parameters of the first artificial neural network ANN1 using the sample data;

[0046] A second acquisition module for acquiring the bidding documents of suppliers in a single procurement project and structuring them;

[0047] A second construction module for constructing a second artificial neural network ANN2 and determining the network parameters of the second artificial neural network ANN2 using the structured data of the bidding documents;

[0048] A probability calculation module for calculating the potential bid rigging probability of supplier combinations;

[0049] A first fusion module for performing feature-level fusion based on the artificial neural network based on the potential bid rigging probability of supplier combinations;

[0050] A second fusion module for performing Bayesian-based decision-level fusion;

[0051] An identification module for identifying potential bid rigging of suppliers according to the decision rule.

[0052] To achieve the above object, in a third aspect, the present invention also provides an electronic device, and the electronic device includes: a processor, a memory, and a communication bus;

[0053] The processor is used to execute one or more programs stored in the memory to implement the steps of the method for analyzing potential bid rigging by suppliers as described above.

[0054] To achieve the above object, in a fourth aspect, the present invention further provides a computer-readable storage medium storing one or more programs, which can be executed by one or more processors to implement the steps of the method for analyzing potential bid rigging by suppliers as described above.

[0055] The embodiments of the present invention overcome the limitations of the method for analyzing potential bid rigging by suppliers based on the fusion of artificial intelligence and Bayesian using single-category features or single-information-source features, reduce its false detection rate and missed detection rate, and improve the reliability and accuracy of the analysis of potential bid rigging by suppliers based on the fusion of artificial intelligence and Bayesian, which is suitable for the occasion of analyzing and evaluating potential bid rigging in multi-supplier bidding. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 It is a flowchart of the method for analyzing potential bid rigging by suppliers provided by an embodiment of the present invention;

[0057] Figure 2 It is a flowchart of the acquisition process provided by an embodiment of the present invention;

[0058] Figure 3 It is a flowchart of determining the network parameters of ANN-1 provided by an embodiment of the present invention through BP algorithm training;

[0059] Figure 4 It is a flowchart of determining the network parameters of ANN-2 provided by an embodiment of the present invention through BP algorithm training;

[0060] Figure 5 It is a flowchart of the decision-level fusion provided by an embodiment of the present invention;

[0061] Figure 6 It is a structural diagram of the device for analyzing potential bid rigging by suppliers provided by an embodiment of the present invention;

[0062] Figure 7 It is a structural diagram of the electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0063] In order to make the objectives, features, and advantages of the present invention more obvious and understandable, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative efforts belong to the scope of protection of the present invention.

[0064] Figure 1 It shows the flow of the method for analyzing potential bid rigging and collusion among suppliers provided by the embodiments of the present invention.

[0065] See Figure 1 , the method for analyzing potential bid rigging and collusion among suppliers includes the following steps:

[0066] S11, collect samples of suppliers' historical bidding data and associated relationship data.

[0067] S12, construct the first artificial neural network ANN1, and use the sample data to determine the network parameters of the first artificial neural network ANN1.

[0068] S13, collect the bidding documents of suppliers in a single procurement project and structure them.

[0069] S14, construct the second artificial neural network ANN2, and use the structured data of the bidding documents to determine the network parameters of the second artificial neural network ANN2.

[0070] S15, calculate the potential bid rigging and collusion probability of the supplier combination.

[0071] S16, based on the potential bid rigging and collusion probability of the supplier combination, perform feature-level fusion based on the artificial neural network.

[0072] S17, perform decision-level fusion based on Bayesian.

[0073] S18, identify potential bid rigging and collusion among suppliers according to the decision rules.

[0074] The present invention overcomes the limitations of the method for analyzing potential bid rigging and collusion among suppliers based on artificial intelligence and Bayesian fusion with single-category features or single-information-source features, reduces its false detection rate and missed detection rate, and improves the reliability and accuracy of the analysis of potential bid rigging and collusion among suppliers based on artificial intelligence and Bayesian fusion, and is suitable for the occasions of analyzing and evaluating potential bid rigging and collusion in multi-supplier bidding.

[0075] Figure 2 It shows the flow of the collection process provided by the embodiments of the present invention. See Figure 2 , collecting samples of suppliers' historical bidding data and associated relationship data includes the following steps:

[0076] S21. Collect historical tendering and bidding data of suppliers, data samples of supplier association relationships, and suppliers' project tendering documents.

[0077] S22. Collect the characteristics of suppliers' historical tendering and bidding behaviors and the characteristics of suppliers' project tendering documents.

[0078] The characteristics of suppliers' historical tendering and bidding behaviors include suppliers' combined tendering characteristics, suppliers' combined association relationship characteristics, and suppliers' combined joint registration confidence characteristics. For example, the combined tendering characteristic of Supplier A and Supplier B is the number of times the two suppliers jointly register for a bid package; the combined association relationship characteristic of suppliers is whether there is an association in the equity penetration between the supplier combinations, and the combined joint registration confidence characteristic of the supply chain combination is the number of times the combination registers for the same bid package / the total number of bid packages registered by Supplier A.

[0079] The characteristics of suppliers' project tendering documents include the IP address characteristics of the uploaded tendering documents, the similarity of tendering texts, and the abnormal characteristics of tendering texts. For example, the similarity of tendering texts will be calculated using the cosine distance after text vectorization; the abnormal characteristics of tendering texts will be judged whether there are abnormally consistent characteristics in the tendering texts through characteristics such as text spelling mistakes, punctuation marks, and font intervals.

[0080] S23. Calculate the historical combined tendering frequency x of suppliers according to the characteristics of suppliers' historical tendering and bidding behaviors 1，1 the combined association relationship x of suppliers 1，2 and the combined joint registration confidence x of suppliers 1，3 .

[0081] S24. Use the machine code detection device for tendering documents to collect the unique characteristics of the machines used by suppliers for tendering, and record whether there is a potential behavior of bid rigging and collusion in the supplier combination in procurement sourcing as R i .

[0082] R i =P in

[0083] where P in is the potential bid rigging and collusion probability of supplier combination i in the nth project.

[0084] S25. Classify the potential bid rigging and collusion probability of the tendering supplier combination according to the probability of R i .

[0085] When R i <s 1 , the potential bid rigging and collusion probability of supplier combination i is "low".

[0086] When s 1≤R i <s 2 When, the potential probability of bid rigging and collusion of supplier combination i is "medium".

[0087] When R i ≥s 2 When, the potential probability of bid rigging and collusion of supplier combination i is "high".

[0088] s 1 <s 2 ,s 1 and s 2 are all natural numbers.

[0089] S26. Compare the potential bid rigging and collusion situation of supplier combination i with the classification situation obtained in step S25. When they are the same, the data of the potential bid rigging and collusion of this supplier combination i is stored as valid sample data in the sample dataset; otherwise, it is excluded as invalid data.

[0090] It should be noted that there are two artificial neural networks constructed in the embodiments of the present invention. The first artificial neural network is denoted as ANN-1, and the second artificial neural network is denoted as ANN-2.

[0091] The structure of ANN-1 is: the historical combination bidding frequency x of suppliers 1，1 , the correlation relationship x of supplier combinations 1，2 and the joint registration confidence x of supplier combinations 1，3 as the input vectors of the three neurons in the input layer of ANN-1. The neurons in the output layer of ANN-1 are denoted by y 1，1 , y 1，2 and y 1，3 representing "low", "medium" and "high" respectively. The number of neurons in the hidden layer of ANN-1 Among them, a 1 is the number of neurons in the input layer of ANN-1, b 1 is the number of neurons in the output layer of ANN-1, and c 1 is a constant.

[0092] The structure of ANN-2 is: the IP address feature x of the uploaded bidding documents 2，1 , the similarity x of bidding texts 2，2 and the abnormal consistency x of bidding texts 2，3 as the input vectors of the three neurons in the input layer of ANN-2. The neurons in the output layer of ANN-2 are denoted by y 2，1 , y 2，2 and y 2，3 representing "low", "medium" and "high" respectively. The number of neurons in the hidden layer of ANN-2 Among them, a2 is the number of neurons in the input layer of ANN-2, b2 is the number of neurons in the output layer of ANN-2, and c 2 is a constant.

[0093] In Figure 2 the illustrated embodiment, the acquisition process of the input data of ANN-1 is taken as an example for illustration. The acquisition process of the input data of ANN-2 is similar to that of the input data of ANN-1.

[0094] The difference between the two is that, in the acquisition process of the input data of ANN-2, the action performed at step S23 is to calculate the IP address feature x of the bid document upload, the similarity x of the bid text 2，1 and the abnormal consistency x of the bid text 2，2 from the features of the supplier project bid document. 2，3 .

[0095] Figure 3 shows the process of determining the network parameters of ANN-1 provided by the embodiment of the present invention through BP algorithm training. Refer to Figure 3 , construct the first artificial neural network ANN1, and use the sample data to determine the network parameters of the first artificial neural network ANN1, including the following steps:

[0096] S301, initialize the network parameters of ANN-1.

[0097] Define w 1 i，j as the connection weight between the input layer and the hidden layer of ANN-1, θ 1 j as the connection threshold between the input layer and the hidden layer of ANN-1, v 1 j，t as the connection weight between the hidden layer and the output layer of ANN-1, γ 1 t as the connection threshold between the hidden layer and the output layer of ANN-1, ε 1 as the error precision of ANN-1 training, n M as the maximum number of iterations.

[0098] wherein, i, t, and j are all integers, i ∈ [1, 3], t ∈ [1, 3], j ∈ [1, n 1 .

[0099] S302, select a sample data of the historical bidding behavior characteristics of a supplier from the sample data set.

[0100] Use I 1 k to represent the k-th sample data of the historical bidding behavior characteristics of a supplier,

[0101]

[0102] Among them, x k 1，1 is the number of combined bids of the k-th supplier in the historical bidding and tendering behavior characteristics, and x k 1，2 is the correlation of the supplier combination in the historical bidding and tendering behavior characteristics of the k-th supplier, and x k 1，3 is the confidence level of joint registration of the supplier combination in the historical bidding and tendering behavior characteristics of the k-th supplier.

[0103] y k 1，1 is the state where the potential probability of bid rigging and collusion of the k-th supplier combination is "low", and y k 1，2 is the state where the potential probability of bid rigging and collusion of the k-th supplier combination is "medium", and y k 1，3 is the state where the potential probability of bid rigging and collusion of the k-th supplier combination is "high".

[0104] S303. Take as the input vector of ANN-1, and take as the expected output vector of ANN-1 and input it into ANN-1.

[0105] S304. Define N = 1, w 1 i，j (1) is the connection weight between the input layer and the hidden layer of ANN-1 in the first iteration, and θ 1 j (1) is the connection threshold between the input layer and the hidden layer of ANN-1 in the first iteration, and v 1 j，t (1) is the connection weight between the hidden layer and the output layer of ANN-1 in the first iteration, and γ 1 t (1) is the connection threshold between the hidden layer and the output layer of ANN-1 in the first iteration;

[0106] S305. Execute the forward propagation process of the input pattern.

[0107] The forward propagation process is as follows:

[0108] a1) Calculate the output of each neuron in the hidden layer of ANN-1

[0109]

[0110] a2) Calculate the actual output of each neuron in the output layer of ANN-1

[0111]

[0112] a3) Calculate the error between the expected output and the actual output of ANN-1

[0113]

[0114] S306. Perform the backpropagation process of the error.

[0115] The backpropagation process is as follows:

[0116] b1) Calculate the correction error of each unit in the output layer of ANN-1

[0117]

[0118] b2) Calculate the correction error of each unit in the hidden layer of ANN-1

[0119]

[0120] S307. Calculate the new connection weights w 1 i，j (N + 1) and the new threshold θ 1 j (N + 1) between the input layer and the hidden layer of ANN-1 for the next iteration.

[0121]

[0122]

[0123] where β 1 is the learning coefficient of ANN-1.

[0124] S308. Calculate the new connection weights v 1 j，t (N + 1) and the new threshold γ 1 t (N + 1) between the hidden layer and the output layer of ANN-1 for the next iteration.

[0125]

[0126]

[0127] where α 1 is also the learning coefficient of ANN-1;

[0128] S309. Judge Is it less than or equal to ε? 1 , or whether N + 1 is equal to n M , if satisfied, the training ends, go to step S310, otherwise, N = N + 1, go to step S305.

[0129] S310, when the conditions of step S309 are met, take the connection weights and thresholds obtained during training as the optimal network parameters of ANN-1.

[0130] Among them, the connection weights as the optimal network parameters of ANN-1 are respectively denoted as and The connection thresholds as the optimal network parameters of ANN-1 are respectively denoted as and

[0131] Figure 4 Shows the process of determining the network parameters of ANN-1 provided by the embodiment of the present invention through BP algorithm training. See Figure 4 , construct the first artificial neural network ANN1, and use the sample data to determine the network parameters of the first artificial neural network ANN1, including the following steps:

[0132] S401, initialize the network parameters of ANN-2.

[0133] Define w 2 i，j′ as the connection weight between the input layer and the hidden layer of ANN-2, θ 2 i′ as the connection threshold between the input layer and the hidden layer of ANN-2, v 2 j′，t as the connection weight between the hidden layer and the output layer of ANN-2, γ 2 t as the connection threshold between the hidden layer and the output layer of ANN-2, ε 2 as the error precision of ANN-2 training;

[0134] Among them, j′ is an integer, j′ ∈ [1, n 2 ;

[0135] S402, select a sample data of the characteristics of the supplier portfolio bidding documents from the sample dataset.

[0136] Use I 2 k to represent the sample data of the kth supplier portfolio bidding document characteristics,

[0137]

[0138] Among them, x k2，1 is the feature of the k-th bid document upload IP address, x k 2，2 is the similarity of the k-th supplier portfolio's bid text, x k 2，3 is the abnormal consistency of the bid text;

[0139] y k 2，1 is the state where the potential probability of bid rigging and collusion for the k-th supplier portfolio is "low", y k 2，2 is the state where the potential probability of bid rigging and collusion for the k-th supplier portfolio is "medium", y k 2，3 is the state where the potential probability of bid rigging and collusion for the k-th supplier portfolio is "high".

[0140] S403, take as the input vector of ANN-2, and take as the expected output vector of ANN-2 and input it into ANN-2.

[0141] S404, define N = 1, w 2 i，j′ (1) is the connection weight between the input layer and the hidden layer of ANN-2 in the first iteration, θ 2 j′ (1) is the connection threshold between the input layer and the hidden layer of ANN-2 in the first iteration, v 2 j′，t (1) is the connection weight between the hidden layer and the output layer of ANN-2 in the first iteration, γ 2 t (1) is the connection threshold between the hidden layer and the output layer of ANN-2 in the first iteration.

[0142] S405, execute the forward propagation process of the input pattern.

[0143] The forward propagation process is as follows:

[0144] c1) Calculate the output of each neuron in the hidden layer of ANN-2

[0145]

[0146] c2) Calculate the actual output of each neuron in the output layer of ANN-2

[0147]

[0148] c3) Calculate the error between the expected output and the actual output of ANN-2

[0149]

[0150] S406. Perform the reverse propagation process of the execution error.

[0151] The reverse propagation process is as follows:

[0152] d1) Calculate the correction error of each unit in the output layer of ANN-2

[0153]

[0154] d2) Calculate the correction error of each unit in the hidden layer of ANN-2

[0155]

[0156] S407. Calculate the new connection weights w 2 i，j′ (N + 1) between the input layer and the hidden layer of ANN-2 for the next iteration and the new threshold θ 2 j′ (N + 1).

[0157]

[0158]

[0159] where β 2 is the learning coefficient of ANN-2;

[0160] S408. Calculate the new connection weights v 2 j′，t (N + 1) between the hidden layer and the output layer of ANN-2 for the next iteration and the new threshold γ 2 t (N + 1);

[0161]

[0162]

[0163] where α 2 is also the learning coefficient of ANN-2;

[0164] S409. Judge whether it is less than or equal to ε 2 , or whether N + 1 is equal to n M . If satisfied, the training ends and go to step S410. Otherwise, N = N + 1 and go to step S405.

[0165] S410. When the conditions of step S409 are met, the connection weights and thresholds obtained through training are used as the optimal network parameters of ANN-2.

[0166] Among them, the connection weights used as the optimal network parameters of ANN-2 are respectively denoted as and The connection thresholds used as the optimal network parameters of ANN-2 are respectively denoted as and

[0167] In some embodiments of the present invention, the process of calculating the potential probability of bid rigging for supplier combinations is as follows:

[0168] D1) Collect the historical bidding behavior characteristics of suppliers and vehicle behavior characteristic signals of the supplier combination, and denote them as and

[0169] D2) According to Calculate the historical combined bidding frequency of suppliers The probability of supplier combination association And the confidence level of joint registration of supplier combinations

[0170] D3) According to Calculate the IP address characteristics of the uploaded bidding documents The similarity of bidding texts And the abnormal consistency of bidding texts

[0171] In some embodiments of the present invention, the feature-level fusion based on artificial neural networks includes the feature-level fusion identification of ANN-1 and the feature-level fusion identification of ANN-2.

[0172] Feature-level fusion identification of ANN-1:

[0173] Take and As the inputs of ANN-1, perform feature-level fusion on the facial behavior characteristic parameters according to the BP artificial neural network model. The actual output of ANN-1 is

[0174]

[0175] Among them,

[0176] Feature-level fusion identification of ANN-2:

[0177] Take and As the input of ANN-2, the facial behavior feature parameters are fused at the feature level according to the BP artificial neural network model, and the actual output of ANN-2 is

[0178]

[0179] Among them,

[0180] Figure 5 shows the process of decision-level fusion provided by the embodiments of the present invention. Refer to Figure 5 , and perform decision-level fusion based on Bayesian, including the following steps:

[0181] S51, perform normalization processing on the output of the artificial neural network.

[0182] Normalize the output of ANN-1 to obtain the normalized result

[0183] Normalize the output of ANN-2 to obtain the normalized result

[0184] S52, perform basic probability assignment for the evidence theory recognition framework.

[0185] Let the recognition framework Θ of the evidence theory = {A 1 , A 2 , A 3}, A 1 , A 2 and A 3 respectively represent the three probability states of "low", "medium" and "high" of the potential bid rigging by supplier combinations. The evidence set e = {e 1 , e 2}, e 1 and e 2 respectively represent the evidence body based on the historical bidding behavior characteristics of suppliers and the evidence body based on the characteristics of suppliers' project bidding documents. m 1 (A 1 ), m 1 (A 2 ) and m 1 (A 3 ) represent the basic probability assignment of the evidence body based on the historical bidding behavior characteristics of suppliers on the recognition framework Θ. m 2 (A 1 ), m 2 (A 2 ) and m 2 (A 3) represents the basic probability assignment of the evidence body based on the characteristics of the supplier project bidding documents. The result after normalizing the artificial neural network is used as the basic probability assignment on the identification framework Θ, that is, let

[0186] S53, decision-level fusion based on D-S evidence theory.

[0187] The rule of decision-level fusion based on D-S evidence theory is:

[0188]

[0189] m(A z ) is the basic probability assignment of the result after decision-level fusion on the identification framework Θ. z is an integer.

[0190] In some embodiments of the present invention, the expression of the decision rule is:

[0191]

[0192] where ε T1 and ε T2 are set thresholds.

[0193] When A f represents that the potential probability of bid rigging in a combination of suppliers is A 1 , then the identification result of decision-level fusion is "low"; when A f represents that the potential probability of bid rigging in a combination of suppliers is A 2 , then the identification result of decision-level fusion is "medium"; when A f represents that the potential probability of bid rigging in a combination of suppliers is A 3 , then the identification result of decision-level fusion is "high".

[0194] The embodiments of the present invention provide a device for analyzing potential bid rigging by suppliers. Refer to Figure 6 As shown, the device for analyzing potential bid rigging by suppliers includes: a first acquisition module 601, a first construction module 602, a second acquisition module 603, a second construction module 604, a probability calculation module 605, a first fusion module 606, a second fusion module 607, and an identification module 608.

[0195] The first acquisition module 601 is used to acquire historical bidding data and associated relationship data samples of suppliers.

[0196] The first construction module 602 is used to construct a first artificial neural network ANN1 and determine the network parameters of the first artificial neural network ANN1 using the sample data.

[0197] The second acquisition module 603 is used to acquire the tender documents of suppliers in a single procurement project and structure them.

[0198] The second construction module 604 is used to construct the second artificial neural network ANN2, and determine the network parameters of the second artificial neural network ANN2 by using the structured data of the tender documents.

[0199] The probability calculation module 605 is used to calculate the potential probability of bid rigging and collusion among supplier combinations.

[0200] The first fusion module 606 is used to perform feature-level fusion based on artificial neural network based on the potential probability of bid rigging and collusion among supplier combinations.

[0201] The second fusion module 607 is used to perform decision-level fusion based on Bayesian.

[0202] The identification module 608 is used to identify the potential bid rigging and collusion of suppliers according to the decision rules.

[0203] In some embodiments, the first acquisition module 601 includes: a historical data acquisition unit, a feature acquisition unit, a confidence acquisition unit, a recording unit, a classification unit, and a comparison unit.

[0204] The historical data acquisition unit is used to acquire the historical bidding data of suppliers, the data samples of supplier association relationships, and the tender documents of supplier projects.

[0205] The feature acquisition unit is used to acquire the historical bidding behavior characteristics of suppliers and the characteristics of tender documents of supplier projects.

[0206] The confidence acquisition unit is used to calculate the historical combined bidding frequency x of suppliers according to the historical bidding behavior characteristics of suppliers 1，1 , the association relationship x of supplier combinations 1，2 , and the joint registration confidence x of supplier combinations 1，3 .

[0207] The recording unit is used to acquire the unique features of the machines used by suppliers for bidding by using the tender document machine code detection device, and record whether there is a potential bid rigging and collusion behavior R in the supplier combination in the procurement sourcing. i .

[0208] The classification unit is used to classify the potential probability of bid rigging and collusion of the tendering supplier combination according to the probability of R i .

[0209] The comparison unit is used to compare the potential bid rigging and collusion situation of supplier combination i with the obtained classification situation. When the two are the same, the data of the potential bid rigging and collusion of the supplier combination i is stored as valid sample data in the sample dataset, otherwise, it is excluded as invalid data.

[0210] In some embodiments, the first construction module 602 includes: an initialization unit, a selection unit, an input unit, a parameter definition unit, a forward propagation unit, a backward propagation unit, a first parameter calculation unit, a second parameter calculation unit, a judgment unit, and a training unit.

[0211] The initialization unit is used for initializing the ANN-1 network parameters.

[0212] The selection unit is used to select a sample data of the historical bidding behavior characteristics of a supplier from the sample dataset.

[0213] The input unit is used to as the input vector of ANN-1, and as the expected output vector of ANN-1 and input it into ANN-1.

[0214] The parameter definition unit is used to define N = 1, w 1 i，j (1) as the connection weight between the input layer and the hidden layer of the first iteration of ANN-1, θ 1 j (1) as the connection threshold between the input layer and the hidden layer of the first iteration of ANN-1, v 1 j，t (1) as the connection weight between the hidden layer and the output layer of the first iteration of ANN-1, γ 1 t (1) as the connection threshold between the hidden layer and the output layer of the first iteration of ANN-1.

[0215] The forward propagation unit is used to perform the forward propagation process of the input pattern.

[0216] The backward propagation unit is used to perform the backward propagation process of the error.

[0217] The first parameter calculation unit is used to calculate the new connection weight w 1 i，j (N + 1) and the new threshold θ 1 j (N + 1) between the input layer and the hidden layer of the next iteration of ANN-1.

[0218] The second parameter calculation unit is used to calculate the new connection weight v 1 j，t (N + 1) and the new threshold γ 1 t (N + 1) between the hidden layer and the output layer of the next iteration of ANN-1.

[0219] The judgment unit is used to judge whether it is less than or equal to ε1 or whether N + 1 is equal to n M If the condition is satisfied, the training ends and the next step is executed; otherwise, N = N + 1, and the forward propagation process of the input mode is executed again.

[0220] The training unit is used to use the connection weights and thresholds obtained through training when the above conditions are met as the optimal network parameters of ANN-1.

[0221] In some embodiments, the probability calculation module 605 includes: an acquisition unit, a first calculation unit, and a second calculation unit.

[0222] The acquisition unit is used to acquire the historical bidding behavior characteristics of suppliers and the vehicle behavior characteristic signals of the supplier combination, denoted as and

[0223] The first calculation unit is used to calculate the historical combined bidding frequency of suppliers the probability of the supplier combination association relationship and the confidence level of the combined registration of the supplier combination

[0224] The second calculation unit is used to calculate the IP address feature of the uploaded bidding document the similarity of the bidding text and the abnormal consistency of the bidding text

[0225] In some embodiments, based on the potential probability of bid rigging by the supplier combination, feature-level fusion based on an artificial neural network is performed, including: feature-level fusion identification of ANN-1 and feature-level fusion identification of ANN-2.

[0226] In some embodiments, the second fusion module 607 includes: a normalization unit, an assignment unit, and a fusion unit.

[0227] The normalization unit is used to perform normalization processing on the output of the artificial neural network.

[0228] The assignment unit is used for basic probability assignment of the evidence theory identification framework.

[0229] The fusion unit is used for decision-level fusion based on the D-S evidence theory.

[0230] In some embodiments, the expression of the decision rule is:

[0231]

[0232] where ε T1 and ε T2is a set threshold value.

[0233] An embodiment of the present invention provides an electronic device. Refer to Figure 7 As shown, it includes a processor 701, a memory 702, and a communication bus 703, where: The communication bus 703 is used to implement connection communication between the processor 701 and the memory 702; The processor 701 is configured to execute one or more computer programs stored in the memory 702 to implement at least one step in the multi-target data association method in the first embodiment above.

[0234] This embodiment also provides a computer-readable storage medium, which includes volatile or non-volatile, removable or non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, computer program modules, or other data). Computer-readable storage media include, but are not limited to, RAM (Random Access Memory), ROM (Read-Only Memory), EEPROM (Electrically Erasable Programmable Read Only Memory), flash memory, or other memory technologies, CD-ROM (Compact Disc Read-Only Memory), digital versatile discs (DVDs), or other optical disc storage, magnetic cassettes, tapes, magnetic disk storage, or any other medium that can be used to store the desired information and can be accessed by a computer.

[0235] The computer-readable storage medium in this embodiment can be used to store one or more computer programs, and the one or more computer programs stored therein can be executed by the processor to implement at least one step of the method in the first embodiment above.

[0236] This embodiment also provides a computer program, which can be distributed on a computer-readable medium and executed by a computable device to implement at least one step of the method in the first embodiment above; and in some cases, at least one step shown or described can be executed in a different order from that described in the above embodiment.

[0237] This embodiment also provides a computer program product, including a computer-readable device, on which the computer program as shown above is stored. In this embodiment, the computer-readable device may include the computer-readable storage medium as shown above.

[0238] As can be seen, those skilled in the art should understand that all or some of the steps in the methods disclosed above, and the functional modules / units in the systems and devices, can be implemented as software (which can be realized by computer program codes executable by a computing device), firmware, hardware, or a suitable combination thereof. In the hardware implementation, the division between the functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, a single physical component can have multiple functions, or a function or step can be executed by the cooperation of several physical components. Some or all of the physical components can be implemented as software executed by a processor, such as a central processing unit, a digital signal processor, or a microprocessor, or implemented as hardware, or implemented as an integrated circuit, such as an application-specific integrated circuit.

[0239] In addition, as is well known to those of ordinary skill in the art, a communication medium typically includes computer-readable instructions, data structures, computer program modules, or other data in a modulated data signal such as a carrier wave or other transmission mechanism, and can include any information delivery medium. Therefore, the present invention is not limited to any specific combination of hardware and software.

[0240] The above content is a further detailed description of the embodiments of the present invention in conjunction with specific implementation manners, and it cannot be determined that the specific implementation of the present invention is only limited to these descriptions. For those of ordinary skill in the technical field to which the present invention belongs, without departing from the concept of the present invention, several simple deductions or substitutions can be made, and all should be regarded as belonging to the protection scope of the present invention.

Claims

1. A method for analyzing potential bid rigging among suppliers, including: Collecting samples of suppliers' historical bidding data and associated relationship data; Constructing a first artificial neural network ANN1 and determining the network parameters of the first artificial neural network ANN1 using the sample data; Collecting suppliers' bidding documents in a single procurement project and structuring them; Constructing a second artificial neural network ANN2 and determining the network parameters of the second artificial neural network ANN2 using the structured data of the bidding documents; Calculating the potential bid rigging probability of supplier combinations; Performing feature-level fusion based on an artificial neural network based on the potential bid rigging probability of supplier combinations; Performing decision-level fusion based on Bayesian; Identifying potential bid rigging among suppliers according to decision rules.

2. The method according to claim 1, wherein, Collecting samples of suppliers' historical bidding data and associated relationship data includes: Collecting suppliers' historical bidding data, samples of suppliers' associated relationship data, and suppliers' project bidding documents; Collecting the characteristics of suppliers' historical bidding behaviors and the characteristics of suppliers' project bidding documents; Calculate the historical combined bidding frequency x of suppliers based on the characteristics of the suppliers' historical bidding and tendering behaviors 1，1 , the supplier portfolio correlation x 1，2 and the confidence level of joint registration of the supplier portfolio x 1，3 ; Use the machine code detection device of the bidding documents to collect the unique features of the machines used by suppliers for bidding, and record whether there is a potential behavior of bid rigging and collusive bidding in the supplier combination during procurement sourcing R i ; According to the probability of R i classify the potential probability of bid rigging by the combination of tendering suppliers; Comparing the potential bid rigging situation of supplier combination i with the obtained classification situation. When the two are the same, the data of the potential bid rigging of the supplier combination i is stored as valid sample data in the sample dataset. Otherwise, it is excluded as invalid data.

3. The method according to claim 1, wherein, Constructing a first artificial neural network ANN1 and determining the network parameters of the first artificial neural network ANN1 using the sample data includes: Initializing the network parameters of ANN-1; Selecting a sample data of the characteristics of a supplier's historical bidding behavior from the sample dataset; Take as the input vector of ANN--1, and take as the expected output vector of ANN-1 and input it into ANN-1; Define N = 1, w 1 i，j (1) is the connection weight between the input layer and the hidden layer of ANN-1 in the first iteration, θ 1 j (1) is the connection threshold between the input layer and the hidden layer of ANN-1 in the first iteration, v 1 j，t (1) is the connection weight between the hidden layer and the output layer of ANN-1 in the first iteration, γ 1 t (1) is the connection threshold between the hidden layer and the output layer of ANN--1 in the first iteration; Performing the forward propagation process of the input pattern; Performing the backward propagation process of the error; Calculate the new connection weights w 1 between the input layer and the hidden layer of ANN-1 for the next iteration i，j and the new threshold θ 1 j (N+1); 1 i，j (N+1) and the new threshold θ 1 j (N+1); Calculate the new connection weights v between the hidden layer and the output layer of ANN-1 for the next iteration 1 j，t (N + 1) and the new threshold γ 1 t (N + 1); Determine whether it is less than or equal to ε 1 , or whether N + 1 is equal to n M , if satisfied, the training ends and the next step is executed; otherwise, N = N + 1, and the forward propagation process of the input mode is executed again; When the above conditions are met, the obtained connection weights and thresholds after training are used as the optimal network parameters of ANN-1.

4. The method according to claim 1, wherein, Calculating the potential bid rigging probability of supplier combinations includes: Collect the supplier historical bidding behavior characteristics and vehicle behavior characteristic signals of the supplier portfolio, denoted as and According to calculate the historical combined bidding frequency of suppliers the probability of the combined supplier association and the confidence level of combined supplier joint registration According to Calculate the IP address feature for uploading the bidding documents Similarity of bidding texts And abnormal consistency of bidding texts 5. The method according to claim 1, wherein, Performing feature-level fusion based on an artificial neural network based on the potential bid rigging probability of supplier combinations includes: Feature-level fusion identification of ANN-1 and feature-level fusion identification of ANN-2.

6. The method according to claim 1, wherein, Performing decision-level fusion based on Bayesian includes: Normalizing the output of the artificial neural network; Basic probability assignment of the evidence theory identification framework; Decision-level fusion based on D-S evidence theory.

7. The method according to claim 1, wherein, The expression of the decision rule is: where ε T1 and ε T2 are set thresholds.

8. A device for analyzing potential bid rigging among suppliers, including: A first collection module for collecting samples of suppliers' historical bidding data and associated relationship data; A first construction module for constructing a first artificial neural network ANN1 and determining the network parameters of the first artificial neural network ANN1 using the sample data; A second collection module for collecting suppliers' bidding documents in a single procurement project and structuring them; A second construction module, configured to construct a second artificial neural network ANN2 and determine network parameters of the second artificial neural network ANN2 by using structured data of tender documents; A probability calculation module, configured to calculate the potential probability of bid rigging by a supplier combination; A first fusion module, configured to perform feature-level fusion based on an artificial neural network based on the potential probability of bid rigging by a supplier combination; A second fusion module, configured to perform decision-level fusion based on Bayesian theory; An identification module, configured to identify potential bid rigging by a supplier according to decision rules.

9. An electronic device, comprising: a processor, a memory, and a communication bus; The processor is configured to execute one or more programs stored in the memory to implement the steps of the method for analyzing potential bid rigging by a supplier according to any one of claims 1 to 7.

10. A computer-readable storage medium storing one or more programs, the one or more programs being executable by one or more processors to implement the steps of the method for analyzing potential bid rigging by a supplier according to any one of claims 1 to 7.

Images