Method, device and equipment for waybill anomaly evaluation and readable medium

Abstract

The embodiment of the invention provides a waybill anomaly evaluation method, device and equipment and a readable medium. The method comprises the following steps: determining a triple comprising a target waybill, a target initiator of the target waybill and a target operator of delivery of the target waybill, wherein an object transported by the target waybill is marked as a loss behavior; determining a target sub-graph matched with the triple from the multi-dimensional heterogeneous graph; determining a first abnormal probability of the triple based on the determined target sub-graph; using a trained first machine learning model to determine a second abnormal probability of the triad based on related information of a target waybill, a target initiator and a target operator in the triad; and by using a trained second machine learning model, based on the related information of the triad, the first anomaly probability and the second anomaly probability, determining an anomaly evaluation result for the triad, the anomaly evaluation result indicating whether the loss behavior of the target waybill belongs to a risk behavior. In this way, the evaluation accuracy can be improved.

Description

Technical Field

[0001] Example embodiments of the present disclosure generally relate to the field of computer technology, and more particularly, to a method, apparatus, device, and computer-readable storage medium for waybill anomaly evaluation. Background Art

[0002] In the delivery scenario of the objects transported by the waybill, after the order for the waybill is completed, there may be a risk of abnormal loss during the delivery of the objects transported by the waybill. However, the risk control effect is poor only through manual reporting and random on-site inspections by risk control personnel to identify and control risks. Summary of the invention

[0003] In a first aspect of the present disclosure, a method for evaluating abnormality of a waybill is provided. The method includes: determining a triple including a target waybill, a target initiator of the target waybill, and a target operator of the target waybill, wherein the object transported by the target waybill is marked as having a loss behavior; determining a target subgraph matching the triple from a multidimensional heterogeneous graph, wherein the multidimensional heterogeneous graph includes a plurality of nodes and relationships between the nodes, wherein each node represents a historical waybill, an initiator of the historical waybill, and an operator of the historical waybill respectively marked as having a loss behavior; determining a first abnormal probability of the triple based on the determined target subgraph; using a trained first machine learning model, based on the relevant information of the target waybill, the target initiator, and the target operator in the triple, determining a second abnormal probability of the triple, wherein the first abnormal probability and the second abnormal probability respectively indicate the probability that the loss behavior of the target waybill belongs to a risk behavior; using a trained second machine learning model, based on the relevant information of the triple, the first abnormal probability, and the second abnormal probability, determining an abnormal evaluation result for the triple, wherein the abnormal evaluation result indicates whether the loss behavior of the target waybill belongs to a risk behavior.

[0004] In a second aspect of the present disclosure, a device for evaluating waybill anomalies is provided. The device includes: a determination module, configured to determine a triple including a target waybill, a target initiator of the target waybill, and a target operator of the target waybill, wherein the object transported by the target waybill is marked as having a loss behavior; a matching module, configured to determine a target subgraph matching the triple from a multidimensional heterogeneous graph, wherein the multidimensional heterogeneous graph includes a plurality of nodes and relationships between the nodes, each node representing a historical waybill, an initiator of the historical waybill, and an operator of the historical waybill, which are marked as having the loss behavior; an execution module, configured to determine a first anomaly probability of the triple based on the determined target subgraph; and a utilization module, It is configured to use a trained first machine learning model to determine a second abnormality probability of the triplet based on the relevant information of the target waybill, the target initiator and the target operator in the triplet, wherein the first abnormality probability and the second abnormality probability respectively indicate the probability that the loss behavior of the target waybill is a risky behavior; and a judgment module is configured to use a trained second machine learning model to determine an abnormality evaluation result for the triplet based on the relevant information of the triplet, the first abnormality probability and the second abnormality probability, wherein the abnormality evaluation result indicates whether the loss behavior of the target waybill is a risky behavior.

[0005] In a third aspect of the present disclosure, an electronic device is provided. The electronic device includes at least one processing unit; and at least one memory, the at least one memory is coupled to the at least one processing unit and stores instructions for execution by the at least one processing unit, and when the instructions are executed by the at least one processing unit, the electronic device executes the method of the first aspect of the present disclosure.

[0006] In a fourth aspect of the present disclosure, a computer-readable storage medium is provided, wherein a computer program is stored on the computer-readable storage medium, and the computer program can be executed by a processor to perform the method according to the first aspect of the present disclosure.

[0007] In a fifth aspect of the present disclosure, a computer program product is provided, comprising computer executable instructions, wherein the computer executable instructions can be executed by a processor to perform the method according to the first aspect of the present disclosure.

[0008] It should be understood that the contents described in the summary of the invention are not intended to limit the key features or important features of the embodiments of the present disclosure, nor are they intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In the following, in conjunction with the accompanying drawings and with reference to the following detailed description, the above and other features, advantages and aspects of various implementations of the present disclosure will become more apparent. In the accompanying drawings, the same or similar reference numerals represent the same or similar elements, wherein:

[0010] Figure 1 A schematic diagram showing an example environment in which embodiments of the present disclosure can be implemented;

[0011] Figure 2 A schematic diagram showing a process of abnormality evaluation of waybills according to some embodiments of the present disclosure;

[0012] Figure 3 A schematic diagram showing an example of a space corresponding to a multi-dimensional heterogeneous graph according to some embodiments of the present disclosure;

[0013] Figure 4 A schematic diagram showing an example of node iteration according to some embodiments of the present disclosure;

[0014] Figure 5A A flowchart showing a process of training a first machine learning model according to some embodiments of the present disclosure;

[0015] Figure 5B A flowchart showing a process of training a second machine learning model according to some embodiments of the present disclosure;

[0016] Figure 6 A flowchart of a process for abnormality evaluation of waybills according to some embodiments of the present disclosure is shown;

[0017] Figure 7 A schematic structural block diagram of an apparatus for evaluating abnormality of waybills according to certain embodiments of the present disclosure is shown; and

[0018] Figure 8 A block diagram of an electronic device is shown in which one or more embodiments of the present disclosure may be implemented. DETAILED DESCRIPTION

[0019] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although some embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments described herein. On the contrary, these embodiments are provided to provide a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not intended to limit the scope of protection of the present disclosure.

[0020] In the description of the embodiments of the present disclosure, the term "including" and similar terms should be understood as open inclusion, that is, "including but not limited to". The term "based on" should be understood as "based at least in part on". The term "one embodiment" or "the embodiment" should be understood as "at least one embodiment". The term "some embodiments" should be understood as "at least some embodiments". Other explicit and implicit definitions may also be included below.

[0021] It should be noted that the acquisition, storage and application of user personal information involved in the technical solution of the present disclosure are in compliance with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0022] It is understandable that before using the technical solutions disclosed in the embodiments of the present disclosure, the types, scope of use, usage scenarios, etc. of the personal information involved in the present disclosure should be informed to the user and the user's authorization should be obtained in an appropriate manner in accordance with relevant laws and regulations.

[0023] For example, in response to receiving an active request from a user, a prompt message is sent to the user to clearly prompt the user that the operation requested to be performed will require obtaining and using the user's personal information, so that the user can independently choose whether to provide personal information to software or hardware such as electronic devices, applications, servers or storage media that execute operations of the technical solution of the present disclosure based on the prompt message.

[0024] As an optional but non-limiting implementation, in response to receiving an active request from the user, the prompt information is sent to the user in a manner such as a pop-up window, in which the prompt information can be presented in text form. In addition, the pop-up window can also carry a selection control for the user to choose "agree" or "disagree" to provide personal information to the electronic device.

[0025] As used herein, the term "model" can learn the association between the corresponding input and output from the training data, so that after the training is completed, the corresponding output can be generated for a given input. The generation of the model can be based on machine learning technology. Deep learning is a machine learning algorithm that processes inputs and provides corresponding outputs by using multi-layer processing units. A neural network model is an example of a model based on deep learning. In this article, "model" may also be referred to as "machine learning model", "learning model", "machine learning network" or "learning network", and these terms are used interchangeably in this article.

[0026] Generally, machine learning can be roughly divided into three stages, namely the training stage, the testing stage, and the application stage (also called the inference stage). In the training stage, a given model can be trained using a large amount of training data, and the parameter values ​​are continuously updated iteratively until the model can obtain consistent inferences that meet the expected goals from the training data. Through training, the model can be considered to be able to learn the association between input and output (also called input-to-output mapping) from the training data. The parameter values ​​of the trained model are determined. In the testing stage, the test input is applied to the trained model to test whether the model can provide the correct output, thereby determining the performance of the model. The testing stage can sometimes be integrated into the training stage. In the application or inference stage, the trained model can be used to process the actual model input based on the parameter values ​​obtained through training to determine the corresponding model output.

[0027] In the scenario of delivery of the object transported by the waybill, after the order for the waybill is completed, the waybill will be delivered by the logistics company. If the waybill is lost in the delivery link of the operator who delivers the waybill, the logistics company will generally bear the corresponding losses. However, in addition to the normal loss due to force majeure, there is also a potential risk. This potential risk is that some initiators of the waybill cooperate with the operator who delivers the waybill, deliberately claim the waybill as their own in the name of loss, and use the compensation system to take improper benefits. This behavior brings risks and financial losses to the logistics company. In response to this risk, the traditional waybill abnormality assessment solution mainly identifies and controls risks through manual reports from the delivery personnel of the operator who delivers the waybill and random inspections of the site by risk control personnel.

[0028] However, there are many problems with the above scheme. The first is that the coverage is not comprehensive, that is, the above scheme relies heavily on random inspections and cannot ensure comprehensiveness. Furthermore, manual reporting or manual verification may often cover a small number of orders, which may result in a large number of risky waybills not being discovered. In this case, not only may the losses that have already occurred be irrecoverable, but the logistics company may also face potential economic losses. Secondly, this scheme is costly and inefficient, that is, a lot of manpower costs are required in the identification process of the scheme, and the identification scale is limited, which cannot cope with the batch identification needs of large-scale waybills, so this method has no advantages in efficiency and cost. In addition, the accuracy of the scheme is low and the subjectivity is strong, that is, the scheme mainly relies on manual judgment, and the subjective factors in these manual judgments may introduce errors. There is also a lack of objective evaluation criteria and model iteration, which may lead to insufficient objectivity, low accuracy and poor stability of the identification results.

[0029] In view of this, an embodiment of the present disclosure provides a scheme for abnormal evaluation of waybills. The scheme includes: determining a triple including a target waybill, a target initiator of the target waybill, and a target operator of the target waybill, and the object transported by the target waybill is marked as having a loss behavior; determining a target subgraph matching the triple from a multidimensional heterogeneous graph; determining a first abnormal probability of the triple based on the determined target subgraph; using a trained first machine learning model, based on the relevant information of the target waybill, the target initiator, and the target operator in the triple, determining a second abnormal probability of the triple; using a trained second machine learning model, based on the relevant information of the triple, the first abnormal probability, and the second abnormal probability, determining an abnormal evaluation result for the triple, the abnormal evaluation result indicating whether the loss behavior of the target waybill is a risky behavior. In this way, target waybills with potential risks can be effectively identified, the efficiency of risk management can be improved, and unreasonable losses caused by risky behaviors can be reduced.

[0030] Figure 1 A schematic diagram of an example environment 100 is shown in which embodiments of the present disclosure can be implemented.

[0031] like Figure 1 As shown, the evaluation system 110 can determine an abnormal evaluation result 125 for the triple based on the triple including the target waybill 101, the target initiator 102, and the target operator 103. In some embodiments, the target initiator 102 can be associated with the target waybill 101. In some examples, the target initiator 102 can indicate an ordering account associated with the target waybill 101. It can be understood that this is merely exemplary and not restrictive.

[0032] In other embodiments, the target operator 103 may be associated with the delivery of the target waybill 101. In some embodiments, the target operator 103 may indicate the terminal operator who delivers the object transported by the target order 101. It will be understood that this is merely exemplary and not restrictive. In more embodiments, the object transported by the target waybill 101 may be marked as having been lost. In some examples, the target waybill 101 may indicate a logistics transport order. In other examples, the object transported by the target waybill 101 may indicate a logistics package. In more examples, the object transported by the target waybill 101 may indicate a commodity purchased using a network platform. It will be understood that this is merely exemplary and not restrictive.

[0033] The evaluation system 110 may include any computing system with computing capabilities, such as various computing devices / systems, terminal devices, server devices, etc. The terminal device may be any type of mobile terminal, fixed terminal or portable terminal, including a mobile phone, a desktop computer, a laptop computer, a notebook computer, a netbook computer, a tablet computer, a media computer, a multimedia tablet, a handheld computer, a portable game terminal, a VR / AR device, a personal communication system (PCS) device, a personal navigation device, a personal digital assistant (PDA), an audio / video player, a digital camera / camcorder, a positioning device, a television receiver, a radio broadcast receiver, an e-book device, a game device, or any combination of the foregoing, including accessories and peripherals of these devices or any combination thereof.

[0034] It should be understood that the structure and function of the various elements in the environment 100 are described for exemplary purposes only and do not imply any limitation on the scope of the present disclosure.

[0035] Some example embodiments of the present disclosure will be described below with continued reference to the accompanying drawings.

[0036] Figure 2 FIG. 2 is a schematic diagram of a process 200 for evaluating a waybill abnormality according to some embodiments of the present disclosure. The process 200 may be implemented at the evaluation system 110. For ease of discussion, reference will be made to FIG. Figure 1 The process 200 is described with reference to the environment 100 of FIG.

[0037] In block 210 , the evaluation system 110 determines a triplet including a target waybill 101 , a target initiator 102 , and a target operator 103 .

[0038] In some embodiments, the target initiator 102 may be associated with the target waybill 101. In some examples, the target initiator 102 may indicate the order account associated with the target waybill 101. It is understood that this is merely exemplary and not restrictive. In other embodiments, the target operator 103 is associated with the delivery of the target waybill 101. In some embodiments, the target operator 103 may indicate the terminal operator of the object transported by the target order 101. It is understood that this is merely exemplary and not restrictive. In more embodiments, the object transported by the target waybill 101 is marked as having lost behavior. In some examples, the target waybill 101 may indicate a logistics transport order. In other examples, the object transported by the target waybill 101 may indicate a commodity purchased using a network platform. It is understood that this is merely exemplary and not restrictive. In an embodiment of the present disclosure, it is expected to determine whether the object loss behavior (e.g., commodity loss) associated with the target waybill 101, the target initiator 102, and the target operator 103 is abnormally lost or normally lost by means of unsupervised and supervised learning techniques.

[0039] Furthermore, the evaluation system 110 may perform anomaly detection 220 on the triples. Furthermore, the evaluation system 110 may pre-construct a multi-dimensional heterogeneous graph 221 and determine whether the triples of interest have abnormal loss behavior in an unsupervised learning manner based on the multi-dimensional heterogeneous graph 221.

[0040] Specifically, the evaluation system 110 may determine the target subgraph 235 that matches the triplet from the multidimensional heterogeneous graph 221. In some embodiments, the multidimensional heterogeneous graph 221 may include multiple nodes and relationships between the nodes, and each node may represent a historical waybill marked as having a loss behavior, an initiator of the historical waybill, and an operator of the delivery of the historical waybill.

[0041] In some examples, the multidimensional heterogeneous graph 221 can be constructed based on a large number of historical waybills, the initiators of the historical waybills, and the operators of the delivery of the historical waybills in which loss behavior has occurred. The evaluation system 110 can use a high-density subgraph mining algorithm to perform subgraph mining on the multidimensional heterogeneous graph 221 to discover one or more target subgraphs 235 with potential abnormal relationships. For example, assuming that it is determined that there is an abnormal relationship between the initiator and the operator of the waybill, it can be explained that the initiator and the operator may have a behavior of jointly occupying the target object. It can be understood that this is merely exemplary and not restrictive. Furthermore, since the high-density subgraph mining algorithm is good at identifying gang fraud, such an algorithm can find the subgraph to which the densest community belongs, and the densest community usually indicates a synchronous and large number of behavior patterns, which is consistent with the abnormal behavior characteristics. In addition, since the high-density subgraph mining algorithm has various node types and there is no direct connection between nodes of the same type, it is also well suited for subgraph mining of heterogeneous graphs. Specifically, the M-Zoom model in the high-density subgraph mining algorithm can be selected. The M-Zoom model is a flexible framework that can be used to find dense blocks in tensors and can also be applied to a wide range of density measurements.

[0042] In some embodiments, the evaluation system 110 may obtain first relevant information of each node in the multidimensional heterogeneous graph 221 and second relevant information of each node in the first candidate subgraph where the node is removed. Further, the evaluation system 110 may determine the predicted abnormal probability of the first candidate subgraph where the node is removed relative to the multidimensional heterogeneous graph 221 based on the first relevant information and the second relevant information. Specifically, the evaluation system 110 may determine the first attribute information of the multidimensional heterogeneous graph 221 based on the first relevant information, and the first attribute information indicates at least one of the first spatial quality attribute, the first spatial size attribute, and the first spatial volume attribute of the plurality of multidimensional heterogeneous graphs 221. Further, the evaluation system 110 may also determine the second attribute information of the first candidate subgraph where the node is removed based on the second relevant information, and the second attribute information indicates at least one of the second spatial quality attribute, the second spatial size attribute, and the second spatial volume attribute. Further, the evaluation system 110 may also determine the predicted abnormal probability of the first candidate subgraph where the node is removed relative to the multidimensional heterogeneous graph 221 based on the first attribute information and the second attribute information.

[0043] Further, the evaluation system 110 may iteratively perform the following steps through the greedy loop search algorithm 230. In some examples, the greedy loop search algorithm 230 may indicate an algorithm that always takes the best choice in the current state at each decision, thereby hoping to lead to the best or optimal result. In other examples, the greedy loop search algorithm 230 may perform operations such as defining suspiciousness, greedy search, and loop search. More specifically, the evaluation system 110 may determine the node to be removed from the first candidate subgraph based on the predicted abnormal probability of the first candidate subgraph with respect to the multidimensional heterogeneous graph 221 when the node is removed, so as to find the target node that can maximize the predicted abnormal probability. In addition, the first candidate subgraph may be initialized as the multidimensional heterogeneous graph 221. Further, the evaluation system 110 may also remove the determined node from the first candidate subgraph and perform the next iteration. Further, the evaluation system 110 may determine the first candidate subgraph corresponding to the maximum predicted abnormal probability as the first target subgraph 235.

[0044] Further, after determining the first target subgraph 235, the evaluation system 110 can delete the first target subgraph 235 from the multidimensional heterogeneous graph 221 to obtain an updated multidimensional heterogeneous graph 221. Further, the evaluation system 110 can also iteratively perform the following steps through the greedy loop search algorithm 230: the evaluation system 110 can also determine the node to be removed from the second candidate subgraph based on the predicted abnormal probability of the second candidate subgraph with the node removed relative to the updated multidimensional heterogeneous graph 221, and then remove the determined node from the second candidate subgraph and perform the next iteration. At this time, the second candidate subgraph is initialized to the updated multidimensional heterogeneous graph 221 obtained by removing the first target subgraph. Then, the evaluation system 110 can also determine the second candidate subgraph corresponding to the maximum predicted abnormal probability as the second target subgraph.

[0045] Further, the evaluation system 110 can determine the first abnormal probability 236 of the triple based on the determined target subgraph 235. The first abnormal probability 236 can respectively indicate the probability that the loss behavior of the target waybill 101 is a risky behavior. Specifically, if the evaluation system 110 determines that the first target subgraph or the second target subgraph includes the target waybill 101, the target initiator 102, and the target operator 103, the maximum predicted abnormal probability corresponding to the target subgraph 235 including the target waybill 101, the target initiator 102, and the target operator 103 can be determined as the first abnormal probability 236.

[0046] Figure 3 A schematic diagram showing an example of a space corresponding to a multi-dimensional heterogeneous graph according to some embodiments of the present disclosure is shown. Figure 3As shown, a multidimensional heterogeneous graph 221 consisting of waybills, initiators of waybills, and operators of delivery of waybills is constructed. Assume that U represents the set of initiators of waybills (order accounts), G represents the set of lost objects of waybills, O represents the set of operators of delivery, R represents the space corresponding to the entire multidimensional heterogeneous graph 221, and X represents the number of lost objects. Therefore, the data set R can be expressed as {U, G, O, X}. An example data set R is shown in Table 1 below:

[0047] Table 1 Data table corresponding to the sample data set

[0048] Waybill initiator Waybill Lost Object Delivery operator Number of lost <![CDATA[U 1 =Xiao Ming]]> <![CDATA[G 1 =Mobile phone]]> <![CDATA[O 1 =A]]> <![CDATA[X 1 =3]]> <![CDATA[U 2 =Xiao Wang]]> <![CDATA[G 2 =Mobile phone]]> <![CDATA[O 2 =B]]> <![CDATA[X 2 =5]]>

[0049] In addition, the two data entries in Table 1 above can constitute a sample space β. Specifically, the sample space β can be a subspace belonging to the space R. It can be understood that the above is only a simple example of a multi-dimensional heterogeneous space. The multi-dimensional heterogeneous graph 221 corresponding to Table 1 above is as follows: Figure 3 As shown, each node in the multi-dimensional heterogeneous space can represent a set of triples, that is, each node can represent the waybill, the initiator and the delivery operator.

[0050] After establishing the multi-dimensional heterogeneous graph 221, the quality, size, and volume of the space R corresponding to the heterogeneous graph can be defined. R represents the quality attribute of space R, S R It can represent the size attribute of space R, V R The volume property of space R can be expressed, so the mass, size and volume of space R can be defined by the following formulas:

[0051] M R =Mass(R)=∑ t∈R t[X] (1)

[0052] Where t[X] represents the number of objects lost.

[0053]

[0054] Where N represents the number of nodes in the multidimensional heterogeneous graph 221, R n It can be simply expressed as the number of objects lost, or it can be represented by other attributes of the node.

[0055]

[0056] Where N represents the number of nodes in the multidimensional heterogeneous graph 221, R n Indicates the number of items lost.

[0057] In some embodiments, the predicted abnormal probability of a subgraph (subspace B) in the multi-dimensional heterogeneous graph 221 (global space R) can be obtained based on the M-Zoom model definition. The predicted abnormal probability calculation formula is as follows:

[0058]

[0059] Among them, M R represents the quality attribute of space R, S R It can represent the size attribute of space R, V R It can represent the volume property of space R, M B represents the quality attribute of the subspace B of space R, S B It can represent the size attribute of subspace B, V B It can represent the volume property of subspace B.

[0060] Specifically, the initial value of the sample space β can be set to the space R corresponding to the multi-dimensional heterogeneous graph 221, that is, β n The value is Furthermore, when and Then for any a i , that is, any node in the multidimensional heterogeneous graph 221, that is, a i ∈ So that ρ susp (BB(a i ), R) is the largest. If we find this a i , then delete the node from the subgraph, that is,

[0061] β′=β-β(a i ) (5)

[0062] Among them, β′ represents the deletion of node a i The sample space after that, β means that node a is not deleted i The previous sample space, β(a i ) means it contains node a i The sample space of .

[0063] β i =β n -{a i} (6)

[0064] Among them, β i Indicates deleting node a i The sample space after that, β n represents the initial sample space, {a i} represents node a i .

[0065] Further, the evaluation system 110 can use the greedy loop search algorithm 230 to gradually remove nodes in the sample space and continuously update the predicted abnormal probability until the nth node or the last node. During the entire node iteration process, if the predicted abnormal probability decreases, the system will backtrack to the iterative step that maximizes the predicted abnormal probability. In this step, the corresponding nodes constitute the target (retained) nodes, and the relationship network between these nodes is the most suspicious high-density subgraph in the entire network. In addition, the nodes in this subgraph are closely related to each other, and the connections with external nodes are relatively sparse, indicating that the nodes in this subgraph can be regarded as risky triples. Figure 4 An example of node iteration is shown. It can be seen that when the abnormal probability (or suspicion) reaches the global maximum, continuing to iterate and delete will cause the abnormal probability to decrease. At this time, the remaining nodes are called target retention nodes. These target retention nodes constitute the target subgraph 235 determined in the global iteration. The waybills, waybill initiators, and delivery operators corresponding to the nodes in the target subgraph 235 correspond to a high-risk group, in which the probability of abnormal loss behavior of objects is high.

[0066] Furthermore, one iteration on the high-dimensional heterogeneous graph can obtain a high-risk group (i.e., a target subgraph 235). In some embodiments, the nodes of the target subgraph 235 determined by this global iteration can be deleted from the high-dimensional heterogeneous graph, and then another iteration can be run in the remaining high-dimensional heterogeneous graph to obtain a second high-risk group (i.e., another target subgraph 235). In this way, each high-risk group can correspond to a predicted abnormal probability ρ. n . Specifically, each high-risk group may correspond to a subgraph, and each subgraph may correspond to a predicted abnormality probability. Furthermore, it is determined whether there are nodes containing the target waybill 101, the target initiator 102, and the target operator 103 in these high-risk groups. If there is no node containing the target waybill 101, the target initiator 102, and the target operator 103, then the corresponding predicted abnormality probability is 0. If there is a node containing the target waybill 101, the target initiator 102, and the target operator 103, then the predicted abnormality probability of the node can be used to represent the predicted abnormality probability of the triple.

[0067] In addition to evaluating the first abnormal probability 236 of the target waybill 101, the target initiator 102, and the target operator 103 in the triple based on the multi-dimensional heterogeneous graph 221, the evaluation system 110 can use the trained first machine learning model 240 to determine the second abnormal probability 246 of the triple based on the relevant information of the target waybill 101, the target initiator 102, and the target operator 103 in the triple. The second abnormal probability 246 also indicates the probability that the loss behavior of the target waybill 101 is a risky behavior.

[0068] Furthermore, the evaluation system 110 may also use the trained second machine learning model 250 to determine an abnormal evaluation result 125 for the triple based on the relevant information of the triple, the first abnormal probability 236 and the second abnormal probability 246. Specifically, the abnormal evaluation result 125 may indicate whether the loss behavior of the target waybill 101 is a risky behavior.

[0069] In some embodiments, the first machine learning model 240 and the second machine learning model 250 may be trained based on supervised learning.

[0070] Continue to refer to Figure 5A The example training process of the first machine learning model 240 is further described. It can be understood that this is merely exemplary and not restrictive.

[0071] Figure 5A FIG. 1 is a flowchart of a process 300A for training a first machine learning model according to some embodiments of the present disclosure. The process 300A may be implemented at the evaluation system 110. For ease of discussion, reference will be made to Figure 1 The process 300A is described with reference to the environment 100 of FIG.

[0072] like Figure 5A As shown, the first training data set 310 can be used to train the first machine learning model 240. In some examples, the first machine learning model 240 can be based on a binary classification algorithm such as Random Forest (RF), extreme gradient boosting (XGBoost), etc., which is not limited here. In other embodiments, the first training data set 310 may include relevant information of the first sample triples including the historical waybills where the loss behavior occurred, the initiators of the historical waybills, and the operators of the delivery of the historical waybills, as well as labels for the first sample triples. Specifically, the label may indicate whether the loss behavior of the historical waybills is a risky behavior. In some examples, the first training data set 310 may include at least one of the merchant attributes, shipping information, account attributes, time characteristics, waybill data, commodity value, commodity weight, and commodity category associated with the historical waybills. Specifically, the account attributes may include at least one of the registration duration and the historical active purchase preference. In other examples, the first training data set 310 may also include at least one of order information, billing data, registration time, historical purchase information, purchase frequency, and types of goods of interest associated with the initiator of the historical waybill. In more examples, the first training data set 310 may also include at least one of historical normal operation information, historical abnormal operation information, and abnormal operation rate associated with the operator of the delivery of the historical waybill.

[0073] Furthermore, the data contained in the first training data set 310 may be data that has undergone a data cleaning process (such as removing duplicate data, processing missing values, converting data formats, etc.) and a data processing process (such as creating new features, standardizing or normalizing data, etc.).

[0074] Furthermore, during the training process, the first training data set 310 can be provided to the first machine learning model 240 to determine the first predicted label 320. Furthermore, the first label error 325 can be determined based on the first predicted label 320 and the corresponding label in the first training data set 310. In addition, the parameters of the first machine learning model 240 can be adjusted based on the first label error 325. The first machine learning model 240 is continuously iterated and trained in this way until the first label error 325 is less than the first error threshold, and the training process of the first machine learning model 240 can be completed.

[0075] Continue to refer to Figure 5B The example training process of the second machine learning model 250 is further described. It can be understood that this is merely exemplary and not restrictive.

[0076] Figure 5B FIG. 3 is a flowchart of a process 300B for training a second machine learning model according to some embodiments of the present disclosure. The process 300B may be implemented at the evaluation system 110. For ease of discussion, reference will be made to Figure 1 The process 300B is described with reference to the environment 100 of FIG.

[0077] like Figure 5B As shown, the evaluation system 110 can use the second training data set 330 to train the second machine learning model 250. In some embodiments, the second machine learning model 250 is trained based on supervised learning. In other embodiments, the second training data set 330 of the second machine learning model 250 may include relevant information of the second sample triples including the historical waybills where the loss behavior occurred, the initiators of the historical waybills, and the operators of the delivery of the historical waybills. In more embodiments, the second machine learning model 250 can be trained after the training of the first machine learning model 240 is completed. Specifically, the data contained in the second training data set 330 may be data that has undergone a data cleaning process (such as removing duplicate data, processing missing values, converting data formats, etc.) and a data processing process (such as creating new features, standardizing or normalizing data, etc.). In some examples, the samples and labels in the second training data set 330 may be the same or different from the samples and labels in the first training data set 310.

[0078] Further, the evaluation system 110 may perform anomaly detection 220 on the second sample triple in the second training data set 330 to determine the first sample anomaly probability 348. Further, the evaluation system 110 may determine the second sample anomaly probability 346 for the second sample triple by the trained first machine learning model 240. Further, the evaluation system 110 may determine the second prediction label 350 for the second sample triple based on the first sample anomaly probability 348, the second sample anomaly probability 346, and the historical waybills 341, the historical initiators 342, and the historical operators 343 in the second training data set 330. Specifically, the second prediction label 350 may indicate whether the loss behavior of the historical waybills 341 is a risky behavior.

[0079] Furthermore, the second label error 355 can be determined based on the second predicted label 350 and the corresponding label in the second training data set 330. In addition, the parameters of the second machine learning model 250 can be adjusted based on the second label error 355. The second machine learning model 250 is continuously iterated and trained in this way until the second label error 355 is less than the second error threshold, and the training process of the second machine learning model 250 can be completed.

[0080] According to the embodiments of the present disclosure, when judging the abnormal loss behavior of the object of the waybill, the advantages of unsupervised learning and the classification prediction ability of the supervised learning model are combined to improve the interpretability and stability of the model. Through this collaborative training, the system can give full play to the advantages of unsupervised learning in feature discovery and anomaly detection, as well as the advantages of supervised learning in classification and prediction, and provide an accurate assessment of the risk of abnormal loss in the logistics process, so as to more accurately formulate risk prevention and control strategies, thereby improving the efficiency of risk management.

[0081] The model proposed in the embodiment of the present disclosure can accurately identify whether the terminal lost object is normal loss by comprehensively covering all lost waybills, and can accurately determine whether the loss event is caused by the ordering account and the terminal personnel jointly possessing and reporting the loss to obtain the goods. This accurate identification capability enables the effective location of improper behavior.

[0082] In addition, the system proposed in the embodiment of the present disclosure has strong recognition stability, and its accuracy can be continuously iterated and optimized through feedback results, thereby improving the accuracy of recognition. The system is highly objective because it does not rely on human intervention and avoids the influence of human error and subjective bias. This automated system design enhances the objectivity and stability of the recognition process.

[0083] Figure 6FIG. 6 is a flowchart of a process 600 for evaluating a waybill abnormality according to some embodiments of the present disclosure. The process 600 may be implemented at the evaluation system 110. For ease of discussion, reference will be made to FIG. Figure 1 The process 600 is described with reference to the environment 100 of FIG.

[0084] At block 610 , the evaluation system 110 determines a triple including a target waybill, a target initiator of the target waybill, and a target operator of the delivery of the target waybill, wherein the object transported by the target waybill is marked as having a loss behavior.

[0085] In box 620, the evaluation system 110 determines a target subgraph that matches the triplet from a multidimensional heterogeneous graph, where the multidimensional heterogeneous graph includes multiple nodes and relationships between the nodes, each node representing a historical waybill marked as having a loss behavior, an initiator of the historical waybill, and an operator of the delivery of the historical waybill.

[0086] At block 630 , the evaluation system 110 determines a first abnormal probability of the triple based on the determined target subgraph.

[0087] In box 640, the evaluation system 110 uses the trained first machine learning model to determine the second abnormal probability of the triplet based on the relevant information of the target waybill, the target initiator and the target operator in the triplet, wherein the first abnormal probability and the second abnormal probability respectively indicate the probability that the loss behavior of the target waybill is a risky behavior.

[0088] In box 650, the evaluation system 110 uses the trained second machine learning model to determine an abnormal evaluation result for the triplet based on the relevant information of the triplet, the first abnormal probability and the second abnormal probability, and the abnormal evaluation result indicates whether the loss behavior of the target waybill is a risky behavior.

[0089] In some embodiments, the following steps are iteratively performed through a greedy loop search algorithm: determining a node to be removed from a first candidate subgraph based on the following predicted abnormality probability: the predicted abnormality probability of the first candidate subgraph from which the node is removed relative to a multidimensional heterogeneous graph, wherein the first candidate subgraph is initialized as a multidimensional heterogeneous graph, and removing the determined node from the first candidate subgraph and performing the next iteration; and determining the first candidate subgraph corresponding to the maximum predicted abnormality probability as the first target subgraph.

[0090] In some embodiments, after determining the first target subgraph, the first target subgraph is deleted from the multidimensional heterogeneous graph to obtain an updated multidimensional heterogeneous graph; through a greedy loop search algorithm, the following steps are iteratively performed: the node to be removed from the second candidate subgraph is determined based on the following predicted abnormality probability: the predicted abnormality probability of the second candidate subgraph where the node is removed relative to the updated multidimensional heterogeneous graph, wherein the second candidate subgraph is initialized to the updated multidimensional heterogeneous graph; the determined node is removed from the second candidate subgraph and the next iteration is performed; the second candidate subgraph corresponding to the maximum predicted abnormality probability is determined as the second target subgraph.

[0091] In some embodiments, in response to determining that the first target subgraph or the second target subgraph includes a target waybill, a target initiator, and a target operator, the maximum predicted abnormality probability corresponding to the target subgraph including the target waybill, the target initiator, and the target operator is determined as the first abnormality probability.

[0092] In some embodiments, first relevant information for each node in a multidimensional heterogeneous graph is obtained; second relevant information for each node in a first candidate subgraph where the node is removed is obtained; and based on the first relevant information and the second relevant information, a predicted abnormal probability of the first candidate subgraph where the node is removed relative to the multidimensional heterogeneous graph is determined.

[0093] In some embodiments, based on first relevant information, first attribute information of the multidimensional heterogeneous graph is determined, the first attribute information indicating at least one of a first spatial quality attribute, a first spatial size attribute, and a first spatial volume attribute of the plurality of multidimensional heterogeneous graphs; and based on second relevant information, second attribute information of the first candidate subgraph removed at the node is determined, the second attribute information indicating at least one of a second spatial quality attribute, a second spatial size attribute, and a second spatial volume attribute; based on the first attribute information and the second attribute information, a predicted abnormal probability of the first candidate subgraph removed at the node relative to the multidimensional heterogeneous graph is determined.

[0094] In some embodiments, the first machine learning model and the second machine learning model are trained based on supervised learning. In other embodiments, the second machine learning model is trained after the training of the first machine learning model is completed.

[0095] In some embodiments, a first training data set of a first machine learning model includes: relevant information of a first sample triplet including historical waybills where loss occurred, the initiator of the historical waybills, and the operator of the delivery of the historical waybills; and a label for the first sample triplet, the label indicating whether the loss of the historical waybills is a risky behavior.

[0096] In some embodiments, the second machine learning model is trained based on supervised learning. In other embodiments, the second training data set of the second machine learning model includes: relevant information of the second sample triples including the historical waybills where the loss behavior occurred, the initiators of the historical waybills, and the operators of the delivery of the historical waybills; the first sample abnormality probability determined for the second sample triples based on the multidimensional heterogeneous graph; the second sample abnormality probability determined for the second sample triples by the trained first machine learning model; and a label for the second sample triple, the label indicating whether the loss behavior of the historical waybills is a risky behavior.

[0097] In summary, the embodiments of the present disclosure can determine the abnormal evaluation results of the triple including the target waybill, the target initiator of the target waybill, and the target operator of the delivery of the target waybill based on the multi-dimensional heterogeneous graph, the first machine learning model, and the second machine learning model. In this way, target waybills with potential risks can be effectively identified, the efficiency of risk management can be improved, and unreasonable losses caused by risky behaviors can be reduced.

[0098] The embodiments of the present disclosure also provide corresponding devices for implementing the above methods or processes.

[0099] Figure 7 A schematic structural block diagram of an apparatus 700 for evaluating abnormal shipping orders according to certain embodiments of the present disclosure is shown. The apparatus 700 may be implemented as or included in the evaluation system 110. Each module / component in the apparatus 700 may be implemented by hardware, software, firmware, or any combination thereof.

[0100] like Figure 7As shown, the device 700 includes a determination module 710, which is configured to determine a triple including a target waybill, a target initiator of the target waybill, and a target operator of the delivery of the target waybill, and the object transported by the target waybill is marked as having a lost behavior; the device 700 also includes a matching module 720, which is configured to determine a target subgraph matching the triple from a multidimensional heterogeneous graph, the multidimensional heterogeneous graph including a plurality of nodes and relationships between the nodes, each node representing a historical waybill marked as having a lost behavior, an initiator of the historical waybill, and an operator of the delivery of the historical waybill; the device 700 also includes an execution module 730, which is configured to determine the triple based on the determined target subgraph The device 700 also includes a utilization module 740, which is configured to utilize the trained first machine learning model to determine the second abnormal probability of the triplet based on the relevant information of the target waybill, the target initiator and the target operator in the triplet, wherein the first abnormal probability and the second abnormal probability respectively indicate the probability that the loss of the target waybill is a risky behavior; the device 700 also includes a judgment module 750, which is configured to utilize the trained second machine learning model to determine the abnormal evaluation result for the triplet based on the relevant information of the triplet, the first abnormal probability and the second abnormal probability, and the abnormal evaluation result indicates whether the loss of the target waybill is a risky behavior.

[0101] In some embodiments, the matching module 720 is further configured to iteratively perform the following steps through a greedy loop search algorithm: determine the node to be removed from the first candidate subgraph based on the following predicted abnormality probability: the predicted abnormality probability of the first candidate subgraph from which the node is removed relative to the multidimensional heterogeneous graph, wherein the first candidate subgraph is initialized as a multidimensional heterogeneous graph, and remove the determined node from the first candidate subgraph and perform the next iteration; and determine the first candidate subgraph corresponding to the maximum predicted abnormality probability as the first target subgraph.

[0102] In some embodiments, the matching module 720 is also configured to delete the first target subgraph from the multidimensional heterogeneous graph after determining the first target subgraph to obtain an updated multidimensional heterogeneous graph; iteratively perform the following steps through a greedy loop search algorithm: determine the node to be removed from the second candidate subgraph based on the following predicted abnormality probability: the predicted abnormality probability of the second candidate subgraph where the node is removed relative to the updated multidimensional heterogeneous graph, wherein the second candidate subgraph is initialized to the updated multidimensional heterogeneous graph; remove the determined node from the second candidate subgraph and perform the next iteration; determine the second candidate subgraph corresponding to the maximum predicted abnormality probability as the second target subgraph.

[0103] In some embodiments, the matching module 720 is also configured to determine the maximum predicted abnormality probability corresponding to the target subgraph including the target waybill, the target initiator and the target operator as the first abnormality probability in response to determining that the first target subgraph or the second target subgraph includes the target waybill, the target initiator and the target operator.

[0104] In some embodiments, the matching module 720 is also configured to obtain first relevant information for each node in the multidimensional heterogeneous graph; obtain second relevant information for each node in the first candidate subgraph where the node is removed; and determine the predicted abnormality probability of the first candidate subgraph where the node is removed relative to the multidimensional heterogeneous graph based on the first relevant information and the second relevant information.

[0105] In some embodiments, the matching module 720 is also configured to determine first attribute information of the multidimensional heterogeneous graph based on the first relevant information, the first attribute information indicating at least one of the first spatial quality attribute, the first spatial size attribute, and the first spatial volume attribute of the multiple multidimensional heterogeneous graphs; and to determine second attribute information of the first candidate subgraph removed at the node based on the second relevant information, the second attribute information indicating at least one of the second spatial quality attribute, the second spatial size attribute, and the second spatial volume attribute; based on the first attribute information and the second attribute information, determine the predicted abnormal probability of the first candidate subgraph removed at the node relative to the multidimensional heterogeneous graph.

[0106] In some embodiments, the first machine learning model and the second machine learning model are trained based on supervised learning. In other embodiments, the second machine learning model is trained after the training of the first machine learning model is completed.

[0107] In some embodiments, a first training data set of a first machine learning model includes: relevant information of a first sample triplet including historical waybills where loss occurred, the initiator of the historical waybills, and the operator of the delivery of the historical waybills; and a label for the first sample triplet, the label indicating whether the loss of the historical waybills is a risky behavior.

[0108] In some embodiments, the second machine learning model is trained based on supervised learning. In other embodiments, the second training data set of the second machine learning model includes: relevant information of the second sample triples including the historical waybills where the loss behavior occurred, the initiators of the historical waybills, and the operators of the delivery of the historical waybills; the first sample abnormality probability determined for the second sample triples based on the multidimensional heterogeneous graph; the second sample abnormality probability determined for the second sample triples by the trained first machine learning model; and the label for the second sample triples, the label indicating whether the loss behavior of the historical waybills is a risky behavior.

[0109] The units and / or modules included in the device 700 can be implemented in various ways, including software, hardware, firmware, or any combination thereof. In some embodiments, one or more units and / or modules can be implemented using software and / or firmware, such as machine executable instructions stored on a storage medium. In addition to or as an alternative to machine executable instructions, some or all of the units and / or modules in the device 700 can be implemented at least in part by one or more hardware logic components. As an example and not limitation, exemplary types of hardware logic components that can be used include field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), and the like.

[0110] It should be understood that one or more steps in the above method can be performed by a suitable electronic device or a combination of electronic devices. Such an electronic device or a combination of electronic devices may include, for example, Figure 1 The evaluation system 110 in FIG.

[0111] Figure 8 8 is a block diagram of an electronic device 800 in which one or more embodiments of the present disclosure may be implemented. It should be understood that Figure 8 The electronic device 800 shown is merely exemplary and should not constitute any limitation on the functionality and scope of the embodiments described herein. Figure 8 The electronic device 800 shown can be used to implement Figure 1 Evaluation system 110 or Figure 7 The device 700.

[0112] like Figure 8 As shown, the electronic device 800 is in the form of a general electronic device. The components of the electronic device 800 may include, but are not limited to, one or more processors or processing units 810, a memory 820, a storage device 830, one or more communication units 840, one or more input devices 850, and one or more output devices 860. The processing unit 810 may be an actual or virtual processor and is capable of performing various processes according to a program stored in the memory 820. In a multi-processor system, multiple processing units execute computer executable instructions in parallel to improve the parallel processing capability of the electronic device 800.

[0113] The electronic device 800 typically includes a plurality of computer storage media. Such media can be any available media accessible to the electronic device 800, including but not limited to volatile and non-volatile media, removable and non-removable media. The memory 820 can be a volatile memory (e.g., register, cache, random access memory (RAM)), a non-volatile memory (e.g., a read-only memory (ROM), an electrically erasable programmable read-only memory (EEPROM), flash memory) or some combination thereof. The storage device 830 can be a removable or non-removable medium, and can include a machine-readable medium, such as a flash drive, a disk, or any other medium, which can be used to store information and / or data and can be accessed within the electronic device 800.

[0114] The electronic device 800 may further include additional removable / non-removable, volatile / non-volatile storage media. Figure 8 As shown in , a disk drive for reading or writing from a removable, non-volatile disk (e.g., a "floppy disk") and an optical drive for reading or writing from a removable, non-volatile optical disk may be provided. In these cases, each drive may be connected to the bus (not shown) by one or more data media interfaces. The memory 820 may include a computer program product 825 having one or more program modules that are configured to perform various methods or actions of various embodiments of the present disclosure.

[0115] The communication unit 840 implements communication with other electronic devices through a communication medium. Additionally, the functions of the components of the electronic device 800 can be implemented with a single computing cluster or multiple computing machines that can communicate through a communication connection. Therefore, the electronic device 800 can operate in a networked environment using a logical connection with one or more other servers, a network personal computer (PC), or another network node.

[0116] The input device 850 may be one or more input devices, such as a mouse, a keyboard, a tracking ball, etc. The output device 860 may be one or more output devices, such as a display, a speaker, a printer, etc. The electronic device 800 may also communicate with one or more external devices (not shown) through the communication unit 840 as needed, such as a storage device, a display device, etc., communicate with one or more devices that allow a user to interact with the electronic device 800, or communicate with any device that allows the electronic device 800 to communicate with one or more other electronic devices (e.g., a network card, a modem, etc.). Such communication may be performed via an input / output (I / O) interface (not shown).

[0117] According to an exemplary implementation of the present disclosure, a computer-readable storage medium is provided, on which computer-executable instructions are stored, wherein the computer-executable instructions are executed by a processor to implement the method described above. According to an exemplary implementation of the present disclosure, a computer program product is also provided, which is tangibly stored on a non-transitory computer-readable medium and includes computer-executable instructions, and the computer-executable instructions are executed by a processor to implement the method described above.

[0118] Various aspects of the present disclosure are described herein with reference to the flowcharts and / or block diagrams of the methods, devices, equipment, and computer program products implemented according to the present disclosure. It should be understood that each box in the flowchart and / or block diagram and the combination of each box in the flowchart and / or block diagram can be implemented by computer-readable program instructions.

[0119] These computer-readable program instructions can be provided to a processing unit of a general-purpose computer, a special-purpose computer, or other programmable data processing device, thereby producing a machine, so that when these instructions are executed by the processing unit of the computer or other programmable data processing device, a device that implements the functions / actions specified in one or more boxes in the flowchart and / or block diagram is generated. These computer-readable program instructions can also be stored in a computer-readable storage medium, and these instructions cause the computer, programmable data processing device, and / or other equipment to work in a specific manner, so that the computer-readable medium storing the instructions includes a manufactured product, which includes instructions for implementing various aspects of the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0120] Computer-readable program instructions can be loaded onto a computer, other programmable data processing apparatus, or other device so that a series of operational steps are performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, so that the instructions executed on the computer, other programmable data processing apparatus, or other device implement the functions / actions specified in one or more boxes in the flowchart and / or block diagram.

[0121] The flow chart and block diagram in the accompanying drawings show the possible architecture, function and operation of the system, method and computer program product according to multiple implementations of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a part of a module, program segment or instruction, and a part of a module, program segment or instruction includes one or more executable instructions for realizing the logical function of the specification. In some implementations as replacements, the function marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two continuous square boxes can actually be executed substantially in parallel, and they can sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be realized by a special hardware-based system that performs the function or action of the specification, or can be realized by a combination of special hardware and computer instructions.

[0122] The above descriptions of various implementations of the present disclosure are exemplary, non-exhaustive, and not limited to the disclosed implementations. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described implementations. The selection of terms used herein is intended to best explain the principles of the implementations, practical applications, or improvements to the technology in the market, or to enable other persons of ordinary skill in the art to understand the various implementations disclosed herein.

Claims

1. A method for evaluating abnormality of waybills, comprising: Determine a triple including a target waybill, a target initiator of the target waybill, and a target operator of delivery of the target waybill, wherein the object transported by the target waybill is marked as having a lost behavior; Determine a target subgraph matching the triplet from a multidimensional heterogeneous graph, the multidimensional heterogeneous graph comprising a plurality of nodes and relationships between the nodes, each node representing a historical waybill marked as having the loss behavior, an initiator of the historical waybill, and an operator of the delivery of the historical waybill; Determining a first abnormal probability of the triple based on the determined target subgraph; Determine, by using the trained first machine learning model, a second abnormal probability of the triplet based on the relevant information of the target waybill, the target initiator, and the target operator in the triplet, wherein the first abnormal probability and the second abnormal probability respectively indicate the probability that the loss behavior of the target waybill is a risky behavior; as well as Using the trained second machine learning model, based on the relevant information of the triplet, the first abnormal probability and the second abnormal probability, an abnormality assessment result for the triplet is determined, and the abnormality assessment result indicates whether the loss behavior of the target waybill is a risky behavior.

2. The method according to claim 1, wherein determining a target subgraph matching the triple from the multidimensional heterogeneous graph comprises: Through the greedy loop search algorithm, the following steps are iteratively performed: determining a node to be removed from a first candidate subgraph based on the following predicted anomaly probabilities: a predicted anomaly probability of the first candidate subgraph at which the node is removed relative to the multidimensional heterogeneous graph, wherein the first candidate subgraph is initialized as the multidimensional heterogeneous graph, and Removing the determined node from the first candidate subgraph and performing a next iteration; as well as A first candidate subgraph corresponding to the maximum predicted abnormality probability is determined as a first target subgraph.

3. The method according to claim 2, wherein determining a target subgraph matching the triple from the multidimensional heterogeneous graph further comprises: After determining the first target subgraph, deleting the first target subgraph from the multi-dimensional heterogeneous graph to obtain an updated multi-dimensional heterogeneous graph; Through the greedy loop search algorithm, the following steps are iteratively performed: determining a node to be removed from a second candidate subgraph based on the following predicted anomaly probabilities: a predicted anomaly probability of the second candidate subgraph at which the node is removed relative to the updated multidimensional heterogeneous graph, wherein the second candidate subgraph is initialized as the updated multidimensional heterogeneous graph; Removing the determined node from the second candidate subgraph and performing a next iteration; The second candidate subgraph corresponding to the maximum predicted abnormality probability is determined as the second target subgraph.

4. The method according to claim 2 or 3, wherein determining the first abnormal probability of the triple based on the determined target subgraph comprises: In response to determining that the first target subgraph or the second target subgraph includes the target waybill, the target initiator and the target operator, the maximum predicted abnormality probability corresponding to the target subgraph including the target waybill, the target initiator and the target operator is determined as the first abnormality probability.

5. The method according to claim 2, wherein the predicted abnormal probability of the first candidate subgraph at which the node is removed relative to the multi-dimensional heterogeneous graph is determined by: Acquire first relevant information of each node in the multi-dimensional heterogeneous graph; Acquire second relevant information of each node in the first candidate subgraph from which the node is removed; as well as Based on the first relevant information and the second relevant information, a predicted abnormal probability of a first candidate subgraph removed from the node relative to the multi-dimensional heterogeneous graph is determined.

6. The method according to claim 5, wherein determining the predicted abnormal probability of the first candidate subgraph removed at the node relative to the multi-dimensional heterogeneous graph based on the first relevant information and the second relevant information comprises: Determine first attribute information of the multi-dimensional heterogeneous graph based on the first relevant information, where the first attribute information indicates at least one of a first spatial quality attribute, a first spatial size attribute, and a first spatial volume attribute of the plurality of multi-dimensional heterogeneous graphs; Based on the second relevant information, determining second attribute information of the first candidate subgraph to be removed at the node, the second attribute information indicating at least one of a second spatial quality attribute, a second spatial size attribute, and a second spatial volume attribute; as well as Based on the first attribute information and the second attribute information, a predicted abnormal probability of a first candidate subgraph removed at the node relative to the multi-dimensional heterogeneous graph is determined.

7. The method of claim 1, wherein the first machine learning model and the second machine learning model are trained based on supervised learning, and wherein the second machine learning model is trained after the training of the first machine learning model is completed.

8. The method according to claim 7, wherein the first training data set of the first machine learning model comprises: Relevant information of the first sample triplet including the historical waybill where the loss behavior occurred, the initiator of the historical waybill, and the operator of the delivery of the historical waybill; as well as For the label of the first sample triplet, the label indicates whether the loss behavior of the historical waybill is a risky behavior.

9. The method of claim 7, wherein the second machine learning model is trained based on supervised learning, and wherein the second training data set for the second machine learning model comprises: Relevant information of a second sample triplet including the historical waybill where the loss behavior occurred, the initiator of the historical waybill, and the operator of the delivery of the historical waybill; A first sample abnormality probability determined for the second sample triplet based on the multi-dimensional heterogeneous graph; a second sample abnormality probability determined by the trained first machine learning model for the second sample triplet; as well as For the label of the second sample triplet, the label indicates whether the loss behavior of the historical waybill is a risky behavior.

10. A device for evaluating abnormality of waybills, comprising: A determination module is configured to determine a triple including a target waybill, a target initiator of the target waybill, and a target operator of the delivery of the target waybill, wherein the object transported by the target waybill is marked as having a lost behavior; a matching module configured to determine a target subgraph matching the triplet from a multidimensional heterogeneous graph, wherein the multidimensional heterogeneous graph includes a plurality of nodes and relationships between the nodes, each node representing a historical waybill marked as having the loss behavior, an initiator of the historical waybill, and an operator of the delivery of the historical waybill; An execution module, configured to determine a first abnormal probability of the triple based on the determined target subgraph; A utilization module is configured to utilize a trained first machine learning model to determine a second abnormal probability of the triplet based on the relevant information of the target waybill, the target initiator, and the target operator in the triplet, wherein the first abnormal probability and the second abnormal probability respectively indicate a probability that the loss behavior of the target waybill is a risky behavior; as well as The judgment module is configured to use a trained second machine learning model to determine an abnormality assessment result for the triplet based on the relevant information of the triplet, the first abnormality probability and the second abnormality probability, wherein the abnormality assessment result indicates whether the loss behavior of the target waybill is a risky behavior.

11. An electronic device, comprising: at least one processing unit; as well as At least one memory, the at least one memory is coupled to the at least one processing unit and stores instructions for execution by the at least one processing unit, the instructions, when executed by the at least one processing unit, cause the electronic device to perform the method according to any one of claims 1 to 9.

12. A computer-readable storage medium having a computer program stored thereon, wherein the computer program can be executed by a processor to implement the method according to any one of claims 1 to 9.

13. A computer program product comprising computer executable instructions, wherein the computer executable instructions, when executed by a processor, implement the method according to any one of claims 1 to 9.

Images