Bidding behavior anomaly detection method and system based on artificial intelligence

Abstract

The present invention discloses an artificial intelligence-based bidding behavior anomaly detection method and system, which includes data collection, anomaly scoring, bid clustering optimization, and bidding behavior anomaly detection. The present invention belongs to the field of anomaly detection, and specifically refers to an artificial intelligence-based bidding behavior anomaly detection method and system. This solution uses dual-category feature construction for complementary detection, comprehensively evaluates bidding behavior, and improves the comprehensiveness of anomaly detection; sets anomaly scores to reflect the degree of anomaly of bidding samples, effectively distinguishing between normal and abnormal bidding behaviors; suppresses the optimal threshold of extreme noise through entropy tuning and extreme correction, reducing false positives and missed reports; introduces a maximum proportion penalty to ensure that a small number of abnormal samples will not affect normal bidding behavior; and based on the separation term and search strategy, prevents falling into local optimality and ensures the reliability of the final anomaly detection results.

Description

Technical Field

[0001] The present invention relates to the field of anomaly detection technology, and specifically to an artificial intelligence-based bidding behavior anomaly detection method and system. Background Art

[0002] Bidding anomaly detection methods are a series of technologies and methods used to identify bidding behaviors that do not conform to normal rules and may involve illegal operations during the bidding process. Their purpose is to maintain a fair and just market competition environment and prevent the occurrence of unfair behaviors such as bid rigging, collusion, and bidding with others. However, general bidding anomaly detection methods cannot simultaneously detect extreme anomalies related to both time and amount, and are unable to adapt to complex behavioral patterns, resulting in poor accuracy in detecting bidding anomalies. General bidding anomaly detection methods are also sensitive to extremely abnormal samples and are significantly affected by noise, causing normal samples to be misclassified as anomalies. Homogenization can mask small anomalies, resulting in poor detection of bidding anomalies. Summary of the Invention

[0003] In view of the above situation, in order to overcome the defects of the existing technology, the present invention provides a bidding behavior anomaly detection method and system based on artificial intelligence. The general bidding behavior anomaly detection method has the problem that it cannot simultaneously take into account the detection of extreme anomalies of time and amount, and cannot adapt to complex behavior patterns, which leads to poor accuracy in bidding behavior anomaly detection. This scheme uses dual-category feature construction to perform complementary detection, comprehensively evaluate bidding behavior, and improve the comprehensiveness of anomaly detection; sets anomaly scores to reflect the degree of abnormality of bidding samples, and effectively distinguishes normal and abnormal bidding behaviors; the general bidding behavior anomaly detection method is sensitive to extreme abnormal samples and has a large impact of noise, resulting in normal samples also being misjudged as abnormal, and homogenization covering up a small number of anomalies, which leads to poor bidding behavior anomaly detection effect. This scheme suppresses the optimal threshold of extreme noise through entropy tuning and extreme correction, reduces false positives and missed reports; introduces a maximum proportion penalty to ensure that a small number of abnormal samples will not affect normal bidding behavior; based on the separation term and search strategy, it prevents falling into local optimality and ensures the reliability of the final anomaly detection result.

[0004] The technical solution adopted by the present invention is as follows: The present invention provides an artificial intelligence-based bidding behavior anomaly detection method, which includes the following steps:

[0005] Step S1: data collection;

[0006] Step S2: abnormality scoring;

[0007] Step S3: bid clustering optimization;

[0008] Step S4: Detection of abnormal bidding behavior.

[0009] Furthermore, in step S1, the data collection is to collect historical bidding record data; construct strategy category features and behavior category features based on the historical bidding record data; perform standardization processing on each dimension of the features to obtain a bidding sample set; and use a sliding window to divide the bidding sample set into different batches of bidding sample subsets.

[0010] Furthermore, in step S2, the anomaly scoring is performed by performing density peak clustering on all bidding samples in the same batch of bidding sample subsets; the special congestion degree within the cluster is calculated for the clustering results, and for the j-th sample, the nearest neighbor h and the farthest neighbor g are found in the cluster to which it belongs, and the strategy category isolation degree is calculated. The formula used is: ;in, is the isolation degree of the jth bidding sample in the strategic category; d is the number of strategic category dimensions, and k is the strategic category dimension index; 、 and are the characteristic values ​​of the bidding sample, the nearest neighbor of the bidding sample, and the farthest neighbor of the bidding sample in the kth dimension of the strategy category; similarly, in the behavior category, we get ;in, is the degree of isolation of the j-th bidding sample in the behavior category; p is the number of behavior category dimensions, l is the behavior category dimension index; 、 and are the characteristic values ​​of the bidding sample, the nearest neighbor of the bidding sample, and the farthest neighbor of the bidding sample in the kth dimension of the behavior category respectively; first calculate the average threshold of the two categories, and the formula used is: ; ;in, and are the average thresholds for the strategy category and the behavior category respectively; B is the number of samples in the current batch; and the anomaly score is calculated using the following formula: ;in, is the abnormal score of the jth sample; set the abnormal score threshold, and determine that the bidding samples with abnormal scores higher than the abnormal score threshold in the clustering results are abnormal behavior samples, otherwise they are normal behavior samples.

[0011] Furthermore, in step S3, the bidding clustering optimization is to optimize the parameters of the cutoff distance, cluster center selection threshold and minimum cluster size used for density peak clustering based on the optimized particle swarm; specifically, the following steps are included:

[0012] Step S31: fitness function design; construct entropy tuning term , and introduce anomaly correction terms , expressed as: ; ; ; ; Where V is the total number of intervals for the anomaly score, and v is the interval index; is the sample proportion of the vth interval; is the weight coefficient of the extreme anomaly correction term; M is the total number of dimensions; N is the total number of samples, and n is the sample index; is the eigenvalue of the mth dimension of the nth sample; abs(·) is the absolute value; I(·) is the indicator function; is the extreme threshold; is the normalized entropy; is the sample outlier adjustment value; and are the minimum and maximum values ​​of the sample abnormal adjustment values, respectively; is the normalized entropy value of the behavior; construct the maximum proportion penalty term , expressed as: ; Construct separation terms , expressed as: ; The final fitness function is expressed as: ; ;in, and are the mean vectors of normal behavior samples and abnormal behavior samples on abnormal scores respectively; F is the individual fitness value; 、 and is the sub-item weight coefficient;

[0013] Step S32: Initialize the particle population; establish a parameter search range based on the cutoff distance, cluster center selection threshold and minimum cluster size used in density peak clustering; and randomly initialize the particle positions;

[0014] Step S33: Particle position update; angle and amplitude control are introduced, and the formula used is: ; ; ; ; ;in, and are the updated positions of the i-th particle after the t+1th iteration and the tth iteration respectively; and is the amplitude coefficient, which controls the search direction; is the global optimal position; is the group mean position; is the search scale; is the search angle; a is the angle control coefficient; rand is a random number uniformly sampled in [0,1]; RC is the radius control coefficient;

[0015] Step S34: inferior angle modulation; perform a secondary update on the 20% individuals with the lowest fitness values ​​and introduce angle modulation; the formula used is: ;in, It is the position after inferior angle modulation; is the modulation angle;

[0016] Step S35: Adaptive update; for , updated using ; For RC, update using ;for , updated using ;in, 、 and They are initial angle coefficient, initial radius coefficient and initial angle respectively; is the maximum number of iterations; is the decay exponent; is the exponential decay rate; and are the maximum modulation angle and the minimum modulation angle respectively;

[0017] Step S36: Optimization judgment; set the maximum number of iterations and the fitness threshold. When there is a particle whose individual fitness value is higher than the fitness threshold, the optimization ends; if the maximum number of iterations is reached, return to step S32; otherwise, continue iterative updating.

[0018] Furthermore, in step S4, the bidding behavior anomaly detection is to perform density peak clustering processing on all bidding samples of the bidding sample subset of the same batch based on the individual particle positions; and to perform early warning processing on the abnormal behavior samples corresponding to the relevant personnel.

[0019] The artificial intelligence-based bidding behavior anomaly detection system provided by the present invention includes a data acquisition module, an anomaly scoring module, a bidding clustering optimization module and a bidding behavior anomaly detection module;

[0020] The data collection module collects historical bidding record data and constructs strategy category and behavior category features; performs standardization processing on each dimension of the features to obtain a bidding sample set, and uses a sliding window to divide the bidding sample set into bidding sample subsets;

[0021] The anomaly scoring module performs density peak clustering on the bidding sample subset and determines whether the bidding behavior sample is normal based on the anomaly score;

[0022] The bidding clustering optimization module tunes the cutoff distance, cluster center selection threshold and minimum cluster size used for density peak clustering based on the optimized particle swarm;

[0023] The bidding behavior anomaly detection module obtains density peak clustering processing results based on individual particle positions and performs bidding behavior anomaly detection.

[0024] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0025] (1) In view of the problem that general bidding behavior anomaly detection methods cannot simultaneously take into account the detection of extreme anomalies of both time and amount, and cannot adapt to complex behavior patterns, which leads to poor accuracy in bidding behavior anomaly detection, this scheme uses dual-category feature construction to conduct complementary detection, comprehensively evaluate bidding behavior, and improve the comprehensiveness of anomaly detection; set anomaly scores to reflect the degree of abnormality of bidding samples, and effectively distinguish between normal and abnormal bidding behaviors.

[0026] (2) In view of the fact that general bidding behavior anomaly detection methods are sensitive to extreme abnormal samples and have a large impact from noise, which causes normal samples to be misjudged as abnormal, and homogeneity masks a small number of anomalies, which in turn leads to poor bidding behavior anomaly detection results, this scheme suppresses the optimal threshold of extreme noise through entropy tuning and extreme correction, reduces false positives and missed negatives; introduces a maximum proportion penalty to ensure that a small number of abnormal samples will not affect normal bidding behavior; based on the separation term and search strategy, it prevents falling into the local optimum and ensures the reliability of the final anomaly detection results. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 A flow chart of the method for detecting abnormal bidding behavior based on artificial intelligence provided by the present invention;

[0028] Figure 2 This is a schematic diagram of the artificial intelligence-based bidding behavior anomaly detection system provided by the present invention.

[0029] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0030] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0031] In the description of the present invention, it should be understood that terms such as "up", "down", "front", "back", "left", "right", "top", "bottom", "inside" and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the system or element referred to must have a specific direction, be constructed and operated in a specific direction. Therefore, they should not be understood as limiting the present invention.

[0032] Example 1, see Figure 1 The present invention provides an artificial intelligence-based bidding behavior anomaly detection method, which includes the following steps:

[0033] Step S1: Data collection: Collect historical bidding record data and construct strategy category and behavior category features; standardize each dimension of the features to obtain a bidding sample set, and use a sliding window to divide the bidding sample set into bidding sample subsets;

[0034] Step S2: Anomaly scoring: Perform density peak clustering on the bidding sample subset and determine whether the bidding behavior sample is normal based on the anomaly score;

[0035] Step S3: bid clustering optimization; based on the optimized particle swarm, the cutoff distance, cluster center selection threshold and minimum cluster size used for density peak clustering are tuned;

[0036] Step S4: Detection of abnormal bidding behavior; based on the density peak clustering processing results obtained from the individual particle positions, abnormal bidding behavior detection is performed.

[0037] Example 2, see Figure 1 This embodiment is based on the above embodiment. In step S1, data collection involves collecting historical bidding record data. Based on the historical bidding record data, strategy category features and behavior category features are constructed. The strategy category features include the bidding time interval, the deviation of the bid amount from the historical average bid amount, the number of repeated bids on the same day, and the bidder's credit rating. The behavior category features include the absolute value of the difference in amount with bidders of the same item, the deviation from the historical best winning bid price, and the reverse indicator of the bid transaction rate. Each dimension of the feature is standardized to obtain a bidding sample set. A sliding window is used to divide the bidding sample set into different batches of bidding sample subsets.

[0038] Example 3, see Figure 1 This embodiment is based on the above embodiment. In step S2, the anomaly scoring is to perform density peak clustering on all bidding samples in the same batch of bidding sample subsets; the special congestion degree within the cluster is calculated for the clustering results, and for the j-th sample, the nearest neighbor h and the farthest neighbor g are found in the cluster to which it belongs, and the strategy category isolation degree is calculated. The formula used is: ;in, is the isolation degree of the jth bidding sample in the strategic category; d is the number of strategic category dimensions, and k is the strategic category dimension index; 、 and are the characteristic values ​​of the bidding sample, the nearest neighbor of the bidding sample, and the farthest neighbor of the bidding sample in the kth dimension of the strategy category; similarly, in the behavior category, we get ;in, is the degree of isolation of the j-th bidding sample in the behavior category; p is the number of behavior category dimensions, l is the behavior category dimension index; 、 and are the characteristic values ​​of the bidding sample, the nearest neighbor of the bidding sample, and the farthest neighbor of the bidding sample in the kth dimension of the behavior category respectively; first calculate the average threshold of the two categories, and the formula used is: ; ;in, and are the average thresholds for the strategy category and the behavior category respectively; B is the number of samples in the current batch; and the anomaly score is calculated using the following formula: ;in, is the anomaly score of the jth sample; The larger the value is, the farther away the nearest neighbor of sample j is from it in its own behavior pattern cluster, indicating that the bidding behavior corresponding to the sample is relatively isolated and suspected to be abnormal. Through the complementarity of the decision-making / behavior dual categories, both time-type and amount-type extremes can be identified. An abnormality score threshold is set to determine that the bidding samples with abnormal scores higher than the abnormality score threshold in the clustering results are abnormal behavior samples, otherwise they are normal behavior samples.

[0039] By performing the above operations, we can address the problem that general bidding behavior anomaly detection methods cannot simultaneously take into account the detection of extreme anomalies in terms of time and amount, and cannot adapt to complex behavior patterns, which in turn leads to poor accuracy in bidding behavior anomaly detection. This solution uses dual-category feature construction to perform complementary detection, comprehensively evaluate bidding behavior, and improve the comprehensiveness of anomaly detection; it sets an anomaly score to reflect the degree of abnormality of the bidding sample, effectively distinguishing between normal and abnormal bidding behavior.

[0040] Example 4, see Figure 1 This embodiment is based on the above embodiment. In step S3, the bidding clustering optimization is to optimize the parameters of the cutoff distance, cluster center selection threshold, and minimum cluster size used for density peak clustering based on the optimized particle swarm. Specifically, the following steps are included:

[0041] Step S31: fitness function design; to prevent the fitness from falling into the state where all scores are squeezed on the same side, avoid excessively extreme thresholds, and prevent excessive dispersion of scores and excessive noise, ensure that there is a clear gap between the normal and abnormal categories in the score range, and it is not easy for the extreme data characteristics of one project to drag down the overall anomaly detection; construct the entropy tuning term ,when Too flat ( , abnormal / normal is difficult to distinguish) or too concentrated ( ), Decrease, automatically avoid these two extremes, reduce the false positives / missing negatives caused by the deviation of single batch data characteristics when the project bidding amounts are very close; and introduce anomaly correction items , reducing the impact of extreme abnormal samples on the identification of normal bidding behavior; entropy tuning term Expressed as: ; ; ; ; Where V is the total number of intervals for the anomaly score, and v is the interval index; is the sample proportion of the vth interval; is the weight coefficient of the extreme anomaly correction term; M is the total number of dimensions, including the strategy category and the behavior category; N is the total number of samples, and n is the sample index; is the eigenvalue of the mth dimension of the nth sample; abs(·) is the absolute value; I(·) is the indicator function; is the extreme threshold; is the normalized entropy; is the sample outlier adjustment value; and are the minimum and maximum values ​​of the sample abnormal adjustment values, respectively; is the normalized entropy value of the behavior; construct the maximum proportion penalty term If too many bidding samples are concentrated in a certain score range (80% of the bid scores are in the same range), it means that the behavior pattern is homogeneous, which is easy to mask a small number of anomalies. By suppressing the threshold and being dragged by normal samples, we can reduce missed reports. The maximum proportion penalty term is expressed as: ; Construct separation terms , directly measures the separability of normal and abnormal bidding behavior under the current parameters, expressed as: ; The final fitness function is expressed as: ; ;in, and are the mean vectors of normal behavior samples and abnormal behavior samples on abnormal scores respectively; F is the individual fitness value; 、 and is the sub-item weight coefficient;

[0042] Step S32: Initialize the particle population; establish a parameter search range based on the cutoff distance, cluster center selection threshold, and minimum cluster size used in density peak clustering; randomly initialize the particle positions; the position of each particle represents a parameter configuration of density peak clustering;

[0043] Step S33: Particle position update; perform a two-way surround search around the current optimal and group mean, introduce angle and amplitude control, and have better coverage; the formula used is: ; ; ; ; ;in, and are the updated positions of the i-th particle after the t+1th iteration and the tth iteration respectively; and is the amplitude coefficient, which controls the search direction; is the global optimal position; is the group mean position; is the search scale; is the search angle; a is the angle control coefficient; rand is a random number uniformly sampled in [0,1]; RC is the radius control coefficient;

[0044] Step S34: inferior angle modulation; perform a secondary update on the 20% individuals with the lowest fitness values ​​and introduce angle modulation to enhance the coverage and convergence of multi-peak solutions; the formula used is: ;in, It is the position after inferior angle modulation; is the modulation angle;

[0045] Step S35: Adaptive update; for , updated using ; For RC, update using ;for , updated using ;in, 、 and They are initial angle coefficient, initial radius coefficient and initial angle respectively; is the maximum number of iterations; is the decay exponent; is the exponential decay rate; and are the maximum modulation angle and the minimum modulation angle respectively;

[0046] Step S36: Optimization judgment; set the maximum number of iterations and the fitness threshold. When there is a particle whose individual fitness value is higher than the fitness threshold, the optimization ends; if the maximum number of iterations is reached, return to step S32; otherwise, continue iterative updating.

[0047] By performing the above operations, we can address the problem that general bidding behavior anomaly detection methods are sensitive to extreme abnormal samples, have a large impact from noise, and thus misjudge normal samples as abnormal. Homogenization masks a small number of anomalies, which in turn leads to poor bidding behavior anomaly detection. This solution uses entropy tuning and extreme correction to suppress the optimal threshold of extreme noise and reduce false positives and missed negatives. It introduces a maximum proportion penalty to ensure that a small number of abnormal samples will not affect normal bidding behavior. Based on the separation term and search strategy, it prevents falling into local optimality and ensures the reliability of the final anomaly detection results.

[0048] Example 5, see Figure 1 This embodiment is based on the above embodiment. In step S4, the bidding behavior anomaly detection is to perform density peak clustering processing on all bidding samples of the bidding sample subset of the same batch based on the individual particle position; and the abnormal behavior samples are corresponding to the relevant personnel for early warning processing.

[0049] Example 6, see Figure 2 ,This embodiment is based on the above embodiment.,The bidding behavior anomaly detection system based on artificial intelligence provided by the present invention, includes a data acquisition module, an anomaly scoring module, a bidding clustering optimization module and a bidding behavior anomaly detection module;

[0050] The data collection module collects historical bidding record data and constructs strategy category and behavior category features; performs standardization processing on each dimension of the features to obtain a bidding sample set, and uses a sliding window to divide the bidding sample set into bidding sample subsets;

[0051] The anomaly scoring module performs density peak clustering on the bidding sample subset and determines whether the bidding behavior sample is normal based on the anomaly score;

[0052] The bidding clustering optimization module tunes the cutoff distance, cluster center selection threshold and minimum cluster size used for density peak clustering based on the optimized particle swarm;

[0053] The bidding behavior anomaly detection module obtains density peak clustering processing results based on individual particle positions and performs bidding behavior anomaly detection.

[0054] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0055] While the embodiments of the present invention have been shown and described, it will be apparent to those skilled in the art that various changes, modifications, substitutions, and alterations can be made to the embodiments without departing from the principles and spirit of the invention.

[0056] The present invention and its embodiments are described above. This description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if a person skilled in the art is inspired by this and, without departing from the purpose of the present invention, designs structures and embodiments similar to this technical solution without inventiveness, they shall fall within the scope of protection of the present invention.

Claims

1. An artificial intelligence-based bidding behavior anomaly detection method, characterized by: The method comprises the following steps: Step S1: Data collection: Collect historical bidding record data and construct strategy category and behavior category features; standardize each dimension of the features to obtain a bidding sample set, and use a sliding window to divide the bidding sample set into bidding sample subsets; Step S2: Anomaly scoring: Perform density peak clustering on the bidding sample subset and determine whether the bidding behavior sample is normal based on the anomaly score; Step S3: bid clustering optimization; based on the optimized particle swarm, the cutoff distance, cluster center selection threshold and minimum cluster size used for density peak clustering are tuned; Step S4: Detecting abnormal bidding behavior: Based on the density peak clustering results obtained from the individual particle positions, abnormal bidding behavior detection is performed; In step S3, the bidding cluster optimization specifically includes the following steps: Step S31: fitness function design; construct entropy tuning term , and introduce anomaly correction terms , expressed as: ; ; ; ; Where V is the total number of intervals for the anomaly score, and v is the interval index; is the sample proportion of the vth interval; is the weight coefficient of the extreme anomaly correction term; M is the total number of dimensions; N is the total number of samples, and n is the sample index; is the eigenvalue of the mth dimension of the nth sample; abs(·) is the absolute value; I(·) is the indicator function; is the extreme threshold; is the normalized entropy; is the sample outlier adjustment value; and are the minimum and maximum values ​​of the sample abnormal adjustment values, respectively; is the normalized entropy value of the behavior; construct the maximum proportion penalty term , expressed as: ; Construct separation terms , expressed as: ; The final fitness function is expressed as: ; ;in, and are the mean vectors of normal behavior samples and abnormal behavior samples on abnormal scores respectively; F is the individual fitness value; 、 and is the sub-item weight coefficient; Step S32: Initialize the particle population; establish a parameter search range based on the cutoff distance, cluster center selection threshold and minimum cluster size used in density peak clustering; and randomly initialize the particle positions; Step S33: Particle position update; angle and amplitude control are introduced, and the formula used is: ; ; ; ; ;in, and are the updated positions of the i-th particle after the t+1th iteration and the tth iteration respectively; and is the amplitude coefficient, which controls the search direction; is the global optimal position; is the group mean position; is the search scale; is the search angle; a is the angle control coefficient; rand is a random number uniformly sampled in [0,1]; RC is the radius control coefficient; Step S34: inferior angle modulation; perform a secondary update on the 20% individuals with the lowest fitness values ​​and introduce angle modulation; the formula used is: ;in, It is the position after inferior angle modulation; is the modulation angle; Step S35: Adaptive update; for , updated using ; For RC, update using ;for , updated using ;in, 、 and They are initial angle coefficient, initial radius coefficient and initial angle respectively; is the maximum number of iterations; is the decay exponent; is the exponential decay rate; and are the maximum modulation angle and the minimum modulation angle respectively; Step S36: Optimization judgment; set the maximum number of iterations and the fitness threshold. When there is a particle whose individual fitness value is higher than the fitness threshold, the optimization ends; if the maximum number of iterations is reached, return to step S32; otherwise, continue iterative updating.

2. The method for detecting abnormal bidding behavior based on artificial intelligence according to claim 1, characterized in that: In step S2, the anomaly scoring is performed by performing density peak clustering on all bidding samples in the same batch of bidding sample subsets; the special congestion degree within the cluster is calculated for the clustering results, and for the j-th sample, the nearest neighbor h and the farthest neighbor g are found in the cluster to which it belongs, and the strategy category isolation degree is calculated. The formula used is: ;in, is the isolation degree of the jth bidding sample in the strategic category; d is the number of strategic category dimensions, and k is the strategic category dimension index; 、 and are the characteristic values ​​of the bidding sample, the nearest neighbor of the bidding sample, and the farthest neighbor of the bidding sample in the kth dimension of the strategy category; similarly, in the behavior category, we get ;in, is the degree of isolation of the j-th bidding sample in the behavior category; p is the number of behavior category dimensions, l is the behavior category dimension index; 、 and are the characteristic values ​​of the bidding sample, the nearest neighbor of the bidding sample, and the farthest neighbor of the bidding sample in the kth dimension of the behavior category respectively; first calculate the average threshold of the two categories, and the formula used is: ; ;in, and are the average thresholds for the strategy category and the behavior category respectively; B is the number of samples in the current batch; and the anomaly score is calculated using the following formula: ;in, is the abnormal score of the jth sample; set the abnormal score threshold, and determine that the bidding samples with abnormal scores higher than the abnormal score threshold in the clustering results are abnormal behavior samples, otherwise they are normal behavior samples.

3. The method for detecting abnormal bidding behavior based on artificial intelligence according to claim 2, characterized in that: In step S1, the data collection is to collect historical bidding record data; construct strategy category features and behavior category features based on the historical bidding record data; Standardize each dimension of the feature to obtain a bidding sample set; A sliding window is used to divide the bidding sample set into bidding sample subsets of different batches.

4. The method for detecting abnormal bidding behavior based on artificial intelligence according to claim 3, characterized in that: In step S4, the bidding behavior anomaly detection is to perform density peak clustering processing on all bidding samples of the bidding sample subset of the same batch based on the individual particle positions; and to perform early warning processing on the abnormal behavior samples corresponding to the relevant personnel.

5. An artificial intelligence-based bidding behavior anomaly detection system, configured to implement the artificial intelligence-based bidding behavior anomaly detection method according to any one of claims 1 to 4, characterized in that: It includes data collection module, anomaly scoring module, bidding clustering optimization module and bidding behavior anomaly detection module; The data collection module collects historical bidding record data and constructs strategic category and behavioral category features; Each dimension of the feature is standardized to obtain a bidding sample set, and a sliding window is used to divide the bidding sample set into bidding sample subsets; The anomaly scoring module performs density peak clustering on the bidding sample subset and determines whether the bidding behavior sample is normal based on the anomaly score; The bidding clustering optimization module tunes the cutoff distance, cluster center selection threshold and minimum cluster size used for density peak clustering based on the optimized particle swarm; The bidding behavior anomaly detection module obtains density peak clustering processing results based on individual particle positions and performs bidding behavior anomaly detection.

Images