TWI726341B - Sample attribute evaluation model training method, device, server and storage medium - Google Patents

Abstract

The embodiment of this specification provides a sample attribute evaluation method. First, a training sample is determined. The training sample includes only a small number of black samples with confirmed attributes and most unknown samples with unconfirmed attributes. Based on the relationship graph corresponding to the training samples, the black sample concentration of each community is determined, combined with the community black sample concentration and the semi-supervised machine learning algorithm, even if the number of black samples is small, the method in this embodiment can also mine unknown samples Potential black samples, and then determine the white samples required for model training, to meet the model training requirements, so that the trained model can accurately evaluate whether the samples belong to the attributes of the black samples.

Description

Sample attribute evaluation model training method, device, server and storage medium

The embodiments of this specification relate to the field of Internet technology, and in particular to a training method, device, and server for a sample attribute evaluation model.

With the rapid development of the Internet, more and more services can be realized through the Internet, such as online payment, online shopping, online insurance claims and other Internet services. While the Internet provides convenience to people's lives, it also brings risks. Unscrupulous persons may engage in electronic business fraud and cause losses to other users. For a large business sample set, the number of risky black samples with a clear attribute of black is small, and most of them are samples with unknown attributes. Due to the hidden nature of business fraud data samples, there is an urgent need to improve the overall risk control capability. Design a scheme that can be trained based on a small number of known black samples to accurately evaluate the attributes of unknown samples.

The embodiments of this specification provide a sample attribute evaluation method, device and server.

In the first aspect, an embodiment of this specification provides a sample attribute evaluation method, including: determining the blackness of each community in the relationship graph corresponding to the training sample. Sample concentration, wherein the training sample includes a black sample and an unknown sample; based on the black sample concentration of each community, the white sample sampling probability of each unknown sample is determined, and the white sample of each unknown sample Sampling with sampling probability to obtain white samples; training the black samples and the white samples based on a semi-supervised machine learning algorithm to obtain a target sample attribute evaluation model.

In the second aspect, an embodiment of this specification provides a sample attribute evaluation model training device, including: a first determining unit for determining the black sample concentration of each community in the relationship graph corresponding to the training sample, wherein the training sample Including black samples and unknown samples; the second determining unit is used to determine the white sample sampling probability of each unknown sample based on the black sample concentration of each community, and use the white sample sampling probability of each unknown sample Sampling is performed to obtain white samples; a training unit is used to train the black samples and the white samples based on a semi-supervised machine learning algorithm to obtain a target sample attribute evaluation model.

In the third aspect, an embodiment of this specification provides a server, including a memory, a processor, and a computer program stored on the memory and capable of running on the processor. When the processor executes the program, any one of the above is implemented. The steps of the sample attribute evaluation method described in item.

In a fourth aspect, an embodiment of this specification provides a computer-readable storage medium on which a computer program is stored, and when the program is executed by a processor, the steps of any one of the above-mentioned sample attribute evaluation methods are implemented.

The beneficial effects of the embodiments of this specification are as follows: In the embodiments of this specification, by determining the training samples, the training samples only include a small number of black samples with confirmed attributes, and most of the unknown samples with unconfirmed attributes. Based on the relationship graph corresponding to the training samples, the black sample concentration of each community is determined, combined with the community black sample concentration and the semi-supervised machine learning algorithm, even if the number of known black samples is small, the method in this embodiment can also be used from unknown samples. Mining potential black samples in, and then determine the white samples required for model training, to meet the model training requirements, so that the trained model can accurately evaluate whether the sample belongs to the attribute of the black sample.

In order to better understand the above technical solutions, the technical solutions of the embodiments of this specification are described in detail below through the drawings and specific embodiments. It should be understood that the embodiments of this specification and the specific features in the embodiments are of the technical solutions of the embodiments of this specification. The detailed description is not a limitation on the technical solution of this specification. The embodiments of this specification and the technical features in the embodiments can be combined with each other if there is no conflict. Please refer to FIG. 1, which is a schematic diagram of an application scenario of sample attribute evaluation according to an embodiment of this specification. The terminal 100 is located on the user side and communicates with the server 200 on the network side. The user can generate real-time events and some business data through the APP or website in the terminal 100. The server 200 collects real-time events generated by each terminal, and then selects training samples. The embodiments of this specification can be applied to risk control scenarios such as identifying risk samples or identifying fraud samples in insurance claims, and can also be applied to two-class classification scenarios. In the first aspect, the embodiment of this specification provides a sample attribute evaluation method. Please refer to FIG. 2, which includes steps S201-S203. S201: Determine the black sample concentration of each community in the relationship graph corresponding to the training sample, where the training sample includes the black sample and the unknown sample; S202: Determine the white sample sampling probability of each unknown sample based on the black sample concentration of each community, and perform sampling with the white sample sampling probability of each unknown sample to obtain a white sample; S203: Training the black sample and the white sample based on a semi-supervised machine learning algorithm to obtain an attribute evaluation model of the target sample. Specifically, in this embodiment, the training samples are first determined in step S201. The training samples may be business data generated by each terminal as shown above. The training samples include black samples with marked attributes and unknown attributes. Of unknown samples. For example: in the insurance claim scenario, the training samples are the relevant data of the user applying for the claim. Among them, the sample corresponding to the user who is determined to be a fraudulent user is a black sample. In the insurance claim scenario, there are fewer black samples that have deceived insurance facts. A large number of black samples are marked, and the accuracy of the sample attribute evaluation model is greatly reduced. How to solve the model training problem in this scenario is a very important work. The method in this embodiment can combine the community attributes of the samples with the semi-supervised machine learning algorithm to mine potential black samples from a large number of unknown samples to reach the number of black samples required for model training, and filter to obtain high trustworthiness For white samples, the purity of black and white samples is ensured during training, thereby completing model training and obtaining a sample attribute evaluation model with higher accuracy. Further, step S201 is used to determine the black sample concentration of each community in the relationship graph corresponding to the training sample. Specifically, in this embodiment, it is necessary to construct a relationship graph including training samples in advance. Specifically, each training sample corresponds to a node, and the constructed relationship graph may only include nodes corresponding to the training samples, or it may be a relationship graph corresponding to nodes in the entire network. The construction process of the graph can be to obtain the historical events of each node in a predetermined time period, based on the historical events, determine the relationship graph according to the preset composition method, and use the preset community discovery algorithm to divide the nodes in the relationship graph into communities, where, Each node corresponds to the community label to which the node belongs. Among them, the preset time period can be specified in advance, and the preset composition method needs to define the following contents: the definition of the node, the definition of the edge, and the definition of the weight value of the edge. This embodiment also does not limit specific composition rules. Different composition rules can be used in different scenarios and different implementations. For example, in an insurance claims scenario, the default composition method can be: take the user as the point, if two users have had financial transactions (such as transfers) within half a year, then connect the two users. The weight of the edge can be the number of transfers between two users. Specifically, in this embodiment, one or more preset community discovery algorithms are run on the relationship graph constructed above, so that each point gets a community label of the community to which the node belongs. The preset community exploration method can be Label Propagation Algorithm (LPA), Fast Unfolding Algorithm (FU), etc., and this application does not limit it here. Among them, the process of tag propagation algorithm is briefly described as follows: Step1: Each point on the graph uses its own point id as its own label; Step2: Each point gets its neighbor label from its neighbor; Step3: After each point receives the labels from all neighbors, it will receive the label with the most occurrences as its own label (if there is a right map, it will be the one with the highest weight and the highest). If there are tags with the same number of tags, choose one of the tags with the most occurrences as your own tag; Step4: Output the label on each point as its own community label. Step3: Repeat Step2 until all points do not change; Step4: Regard each community obtained in Step3 as a point, repeat Step2 until all communities do not change; Step5: Output the label on each point as its own community label. After dividing the communities in the relationship diagram, the black sample concentration of each community can be calculated. The black sample concentration of each community can be determined in the following three ways: The first method is to determine the first proportion of the nodes corresponding to all black samples in each community in the total nodes of the community, and use the first proportion as the black sample concentration of the community. The second method is to determine the second proportion of the nodes corresponding to all black samples in each community in the total nodes in the relationship graph, and use the first proportion as the black sample concentration of the community. The third type: determine the third proportion of the nodes corresponding to all black samples in each community in the total nodes of the community, and the fourth proportion of the total nodes in the community in the total nodes in the relationship graph, and obtain the A weighted average of the third proportion and the fourth proportion, and the weighted average is used as the black sample concentration of the community. Specifically, in this embodiment, using the first method, the black sample concentration can be defined as the number of black samples in the community divided by the total number of community nodes. For example: Community A includes a total of 5 nodes, of which one node is the node corresponding to the black sample. In this way, the black sample concentration of community A can be calculated to be 1/5, and the black sample concentration of all nodes in the community is 1/5. Of course, you can also use the second method above to define the size of the entire relationship graph. For example: Community A includes a total of 5 nodes, of which one node is the node corresponding to the black sample, and the relationship graph includes 10 nodes. In this way, the black sample concentration of community A can be calculated to be 1/10. The black sample concentration of all nodes is 1/10. Of course, it can be set in combination with the size of the community and the proportion of black samples in the community, using the third method described above. For example: Community A includes a total of 5 nodes, of which one node is the node corresponding to the black sample, and the relationship graph includes 10 nodes. In this way, the black sample concentration of Community A can be calculated as K1*1/5+K2 *5/10, where K1 and K2 represent weighting coefficients, which can be set according to actual needs, and the black sample concentration of all nodes in the community is 0.2K1+0.5k2. In the specific implementation process, the definition of the black sample concentration can be set according to actual needs, and this application is not limited here. After the black sample concentration of each sample is determined, through step S202, based on the black sample concentration of each community, the white sample sampling probability of each unknown sample is determined, and sampling is performed based on the white sample sampling probability of each unknown sample. Obtain a white sample. In the method of this embodiment, when mining potential black samples from unknown samples, combined with community attributes, even if a small part of the unknown samples do not reflect the true characteristics of the black samples at the current moment, in-depth mining can be carried out in combination with community characteristics, which can truly expand the black samples. The sample ratio meets the requirements of model training. Specifically, in this embodiment, for each unknown sample in the training sample, the white sample sampling probability of the sample can be determined according to the black sample concentration of the sample. For example: if unknown sample 1 is located in community A, the black sample concentration of this community A is 1/5, so the black sample concentration of unknown sample 1 is 1/5, and the white sample sampling probability of unknown sample 1 is P1=1-1/ 5=4/5. Further, in the initial first training, the white sample sampling probability of each unknown sample can be set to a fixed value. For example, if there are 100 unknown samples, the white sample probability of each unknown sample can be set to 1/100 during the first training. In the subsequent rounds of training, the black sample concentration of the unknown sample is combined to determine the white sample sampling probability of the unknown sample. In this way, after the white sample sampling probabilities of each unknown sample are determined, the unknown samples can be sampled with white samples according to their respective white sample sampling probabilities, the white samples drawn are determined, and then the black samples with marked attributes are combined through step S203 , Based on the semi-supervised machine learning algorithm, the black samples and white samples are trained to obtain the sample attribute evaluation model. The specific implementation may include the following steps: Train black samples and white samples based on semi-supervised machine learning algorithms to obtain sample attribute evaluation models; Determine whether the sample attribute evaluation model meets the preset convergence conditions; If not, update the black sample concentration of each community, and continue training based on the updated black sample concentration of each community and the semi-supervised machine learning algorithm until the trained sample attribute evaluation model meets the preset convergence conditions, which will meet the pre-set Set the sample attribute evaluation model of the convergence condition as the target sample attribute evaluation model. In this embodiment, the semi-supervised machine learning algorithm includes semi-supervised (Positive and Unlabeled Learning, positive sample and unlabeled learning) machine learning algorithm, which is a kind of semi-supervised learning machine learning algorithm, which refers to training Among the training samples of the machine learning model, only part of the training samples are labeled samples, while the rest of the training samples are unlabeled samples. Unlabeled samples are used to assist the learning process of labeled samples. Among the training samples collected by the modeling party, there are only a small number of labeled black samples, and the remaining samples are unlabeled unknown samples. The machine learning process is aimed at labeled positive samples and unlabeled samples. After constructing the black samples and white samples, these training samples can be trained based on a semi-supervised machine learning algorithm to build a sample attribute evaluation model. For semi-supervised machine learning algorithms, multiple machine learning strategies can usually be included. For example, semi-supervised machine learning algorithms include typical machine learning strategies, including two-stage strategy and cost-sensitive strategy. The so-called two-stage method, the algorithm is based on the known positive samples and unlabeled samples, and discovers the potential reliable negative samples in the unlabeled samples, and then based on the known positive samples and the mined reliable negative samples, the problem is transformed Train the classification model for the traditional supervised machine learning process. For cost-sensitive strategies, the algorithm assumes that the proportion of positive samples in unlabeled samples is extremely low, and by directly treating unlabeled samples as negative samples, a higher cost-sensitive weight is set for positive samples relative to negative samples. For example, usually in the target equation based on cost-sensitive semi-supervised machine learning algorithms, a higher cost-sensitive weight is set for the loss function corresponding to the positive sample. By setting a higher cost-sensitive weight for the positive samples, the cost of the final trained classification model for dividing a positive sample is far greater than the cost of dividing a negative sample. In this way, you can directly use positive samples and unlabeled The samples (as negative samples) learn a cost-sensitive classifier to classify unknown samples. In this embodiment, the above training samples can be trained based on a cost-sensitive semi-supervised machine learning algorithm, or a two-stage method can be used to train the above training samples. In the specific implementation process, it can be set according to needs, and this application does not limit it here. In this embodiment, a two-stage semi-supervised machine learning algorithm is mainly used for detailed introduction. Taking an insurance claim scenario as an example, suppose that the black sample in the above training sample set is marked as 1, which means that the sample is known fraudulent insurance data, and the white sample is marked as 0, which means that the insurance data corresponding to the training sample is normal. After training the black sample and the white sample sampled based on the sampling probability of the white sample, the binary classification model is obtained to obtain the sample attribute evaluation model, and then the sample attribute evaluation model is used to evaluate the unknown samples, and each unknown sample is marked as The black sample score of the black sample. The black sample score is a value ranging from 0 to 1, indicating the probability that the unknown sample belongs to the black sample. Of course, the black sample, white sample and the corresponding black sample score can also be defined in other ways, and this application is not limited here. In this way, the training samples are trained in multiple rounds. After each round of training, the corresponding sample attribute evaluation model is obtained. It is necessary to determine whether the sample attribute evaluation model meets the preset convergence conditions. If the model converges, the sample obtained from this round of training The attribute evaluation model is used as the final target sample attribute evaluation model. If the model has not converged, update the black sample concentration of each unknown sample and continue training in the aforementioned manner until the trained model reaches the convergence condition. In this embodiment, judging whether the model converges can be achieved through the following steps: Evaluate each unknown sample based on the sample attribute evaluation model, obtain the current round of attribute evaluation results for each unknown sample, and obtain a total of M current round of attribute evaluation results, where M is the number of unknown samples; Based on M current-round attribute evaluation results and M last-round attribute evaluation results, it is judged whether the sample attribute evaluation model meets the preset convergence condition. Among them, each unknown sample is evaluated based on the sample attribute evaluation model, and the current round of attribute evaluation results for each unknown sample are obtained, including: Each unknown sample is evaluated based on the sample attribute evaluation model, and the black sample score of each unknown sample is obtained. If the black sample score value is greater than the preset score, the attribute information of the unknown sample is marked as a black sample. The attribute information of the unknown sample is included in the current round of attribute evaluation results of the unknown sample. Specifically, in this embodiment, the sample attribute evaluation model corresponding to the round of training is obtained in each round of training, and the black sample of each unknown sample is scored by using the model, and the unknown sample can be marked according to the score. Specifically, if the score value of the black sample is greater than the preset score, the attribute information of the unknown sample is marked as a black sample. For example, the default score is set to 0.8, the black sample score of the unknown sample 1 is 0.9, and the attribute information of the unknown sample 1 is marked as a black sample. The black sample score of the unknown sample 2 is 0.4, and the attribute information of the unknown sample 2 remains unchanged, and it is still an unknown attribute sample. In this way, the attribute evaluation result of each unknown sample in this round of training can be determined, and the evaluation result can include the black sample score and attribute information of the unknown sample in this round of training. If the number of unknown samples is M, then M results of this round of attribute evaluation are obtained. Furthermore, the attribute evaluation results corresponding to the unknown sample in the previous round of training can also be obtained, that is, M attribute evaluation results of the previous round. According to the M results of this round of attribute evaluation and the M results of the previous round of attribute evaluation, it can be determined whether the sample attribute evaluation model meets the preset convergence conditions, which can be achieved through the following steps: Determine whether the attribute information in the current round of attribute evaluation results of each unknown sample is consistent with the attribute information in the previous round of attribute evaluation results of the unknown sample. If so, it indicates that the current round of sample attribute evaluation model meets the preset convergence condition. Specifically, in this embodiment, through the attribute information in each evaluation result of the M last round of attribute evaluation results, it is determined which unknown samples are included in the black samples marked as black samples in the previous round of training, and the M current round attributes are passed. The attribute information in each evaluation result in the evaluation result determines which unknown samples are included as black samples in this round of training. If the unknown samples marked as black samples in the previous round are consistent with the unknown samples marked as black samples in this round, it indicates that the label of each unknown sample has not changed, and the model has reached convergence. For example, the black samples in the unknown samples in the previous round of training include unknown sample 1, unknown sample 2, unknown sample 5, and unknown sample 10. The black samples in this round of training also include unknown sample 1, unknown sample 2, unknown sample 5, and unknown sample 10, indicating that the unknown sample has not changed, and the sample attribute evaluation model trained in this round has reached convergence. The sample attribute evaluation model obtained in this round of training is used as the target sample attribute evaluation model. Further, the preset convergence condition for determining whether the model reaches convergence can be set according to actual needs. The foregoing example is only an example of specific implementation, and does not constitute a limitation to the application. For example: it can also be set as if the unknown samples marked as black samples in the previous round are consistent with the unknown samples marked as black samples in this round, the proportion of unknown samples reached the preset proportion, indicating that each unknown sample has no mark Change, the model reaches convergence. Further, if it is determined that the sample attribute evaluation model obtained in the current round of training does not meet the preset convergence condition, it indicates that the model has not yet converged, and the next round of training is required. Before the next round of training, since the marked black samples have changed from the previous round of training, it is necessary to update the black sample concentration of the community based on the marked black samples, and then update the black sample concentration of each unknown sample , The specific implementation can include the following steps: Based on the M results of this round of attribute evaluation and the M results of the last round of attribute evaluation, determine the unknown sample whose attribute information has changed; recalculate the black sample concentration of the community corresponding to the unknown sample whose attribute information has changed. Specifically, in this embodiment, based on the attribute evaluation result of each unknown sample corresponding to the current round of training and the attribute evaluation result of each unknown sample corresponding to the previous round of training, it is possible to locate which unknown samples’ attribute information has changed. The change can be from a black sample attribute to an unknown attribute, or from an unknown attribute to a black sample attribute. It then locates the community where the unknown sample that caused the attribute change is located, recalculates the black sample concentration of the community, and updates the white sample probability of the node corresponding to the community based on the updated black sample concentration of the community. For example, community A includes nodes corresponding to unknown sample 1, unknown sample 2, and black sample 1. In the last round of training, the attributes of all nodes in community A have changed, and the black sample concentration corresponding to each node in community A is 1/3. In this round of training, the unknown sample 1 is marked as a black sample, and the attributes of the remaining nodes have not changed, and the black sample concentration corresponding to each node in the community A is updated to 2/3. In this way, the sampling probability of the white samples corresponding to the unknown sample 1 and the unknown sample 2 are both 1/3. In this way, the white sample sampling probability of the unknown node can be updated, and then the white sample sampling according to the white sample sampling probability of each node can be repeated to obtain the white sample. The white sample obtained by the sampling and the known black sample are combined, based on semi-supervised The machine learning algorithm performs the next round of training of the sample attribute evaluation model until the sample attribute evaluation model obtained by training reaches the above-mentioned preset convergence condition. Furthermore, the target sample attribute evaluation model is obtained by training in the foregoing manner, and the model can be used to evaluate the sample attribute of the newly-injected sample to determine the evaluation result of the newly-injected sample, wherein the evaluation result includes the black sample score and/or attribute information of the newly-injected sample . Specifically, in this embodiment, the known black samples and the remaining white samples with higher trustworthiness after screening out potential black samples are used for model training, and the obtained target sample attribute evaluation model has a higher evaluation accuracy, which can be used for new samples. Perform sample attribute evaluation, and the evaluation result may include the aforementioned black sample score, indicating the probability that the new sample belongs to the black sample. The evaluation result may also include the attribute information of the new sample. For example, the black sample score of the new sample is 0.9, which is greater than a preset score of 0.8, and it is determined that the attribute information of the new sample is a black sample. Through the evaluation results, relevant personnel can learn the attributes of the new sample in time and conduct risk control in a timely manner. Further, in this embodiment, since the relationship between nodes will change over time, the relationship graph in the foregoing embodiment can be updated at a preset time interval, and the updated relationship graph can be re-divided into communities. Obtain the black sample concentration of the corresponding community, and retrain the sample attribute evaluation model so that the model can be updated at a preset time interval. The method in this embodiment can be applied to insurance claims scenarios. The training sample is the insurance information of claimants, and the black sample is the insurance information of known fraudsters. The fraud evaluation model is obtained by the aforementioned method, and the new claims applicants After inputting the relevant insurance data into the insurance fraud evaluation model, the newly recruited claim adjuster can obtain the evaluation score of the fraudulent insurance personnel or the attribute of whether it is fraudulent insurance. In this way, the relevant personnel of the insurance company can conduct follow-up related examinations on the suspected fraudulent insurance personnel based on such assessment results, avoiding unnecessary property losses. In the second aspect, based on the same inventive concept, an embodiment of this specification provides a sample attribute evaluation model training device. Please refer to FIG. 3, which includes: The first determining unit 301 is configured to determine the black sample concentration of each community in the relationship graph corresponding to the training sample, where the training sample includes the black sample and the unknown sample; The second determining unit 302 is configured to determine the white sample sampling probability of each unknown sample based on the black sample concentration of each community, and perform sampling with the white sample sampling probability of each unknown sample to obtain a white sample ； The training unit 303 is configured to train the black sample and the white sample based on a semi-supervised machine learning algorithm to obtain an attribute evaluation model of the target sample. In an optional implementation manner, the training unit 303 is specifically configured to: Training the black sample and the white sample based on a semi-supervised machine learning algorithm to obtain a sample attribute evaluation model; Judging whether the sample attribute evaluation model satisfies a preset convergence condition; If not, update the black sample concentration of each community, and continue training based on the updated black sample concentration of each community and the semi-supervised machine learning algorithm until the sample attribute evaluation model obtained by training satisfies the preset The convergence condition is to use the sample attribute evaluation model that satisfies the preset convergence condition as the target sample attribute evaluation model. In an optional implementation manner, the training unit 303 is specifically configured to: Evaluate each of the unknown samples based on the sample attribute evaluation model, obtain the current round of attribute evaluation results of each of the unknown samples, and obtain a total of M current round of attribute evaluation results, where M is the number of unknown samples; Based on the M current-round attribute evaluation results and the M last-round attribute evaluation results, it is determined whether the sample attribute evaluation model satisfies a preset convergence condition. In an optional implementation manner, the training unit 303 is specifically configured to: Evaluate each of the unknown samples based on the sample attribute evaluation model to obtain the black sample score of each unknown sample. If the black sample score value is greater than the preset score value, mark the attribute information of the unknown sample as black Samples, wherein the current round of attribute evaluation results of each of the unknown samples include attribute information of the unknown sample. In an optional implementation manner, the training unit 303 is specifically configured to: Determine whether the attribute information in the current round of attribute evaluation results of each unknown sample is consistent with the attribute information in the previous round of attribute evaluation results of the unknown sample. If so, it indicates that the current round of sample attribute evaluation model satisfies the preset convergence condition. In an optional implementation manner, the training unit 303 is specifically configured to: Based on the M current-round attribute evaluation results and the M last-round attribute evaluation results, determine unknown samples whose attribute information has changed; Recalculate the black sample concentration of the community corresponding to the unknown sample whose attribute information has changed. In an optional implementation manner, the device further includes an evaluation unit, and the evaluation unit is specifically configured to: After the sample attribute evaluation model that satisfies the preset convergence condition is used as the target sample attribute evaluation model, the new sample is evaluated according to the target sample attribute evaluation model, and the evaluation result of the new sample is determined, wherein the The evaluation result includes the score and/or attribute information of the newly injected black sample. In an optional implementation manner, the training samples are insurance data corresponding to claimants, and the black samples are insurance data corresponding to insurance fraudsters. In an optional implementation manner, the first determining unit is specifically configured to: Determine the first proportion of nodes corresponding to all black samples in each community in the total nodes of the community, and use the first proportion as the black sample concentration of the community; or Determine the second proportion of the nodes corresponding to all black samples in each community in the total nodes in the relationship graph, and use the first proportion as the black sample concentration of the community; or Determine the third proportion of the nodes corresponding to all black samples in each community among the total nodes of the community, and the fourth proportion of the total nodes of the community among the total nodes in the relationship graph, to obtain the third proportion And the weighted average of the fourth proportion, and the weighted average is used as the black sample concentration of the community. In the third aspect, based on the same inventive concept as the sample attribute evaluation method in the previous embodiment, the present invention also provides a server, as shown in FIG. 4, which includes a memory 404, a processor 402, and a device stored on the memory 404. A computer program that can run on the processor 402, and when the processor 402 executes the program, the steps of any one of the aforementioned sample attribute evaluation methods are implemented. Among them, in FIG. 4, the bus bar architecture (represented by the bus bar 400), the bus bar 400 may include any number of interconnected bus bars and bridges, and the bus bar 400 will include one or more processes represented by the processor 402 The memory and various circuits of the memory represented by the memory 404 are linked together. The bus bar 400 can also link various other circuits such as peripheral devices, voltage regulators, and power management circuits, which are all known in the art, and therefore, no further description will be given herein. The bus interface 406 provides an interface between the bus 400 and the receiver 401 and transmitter 403. The receiver 401 and the transmitter 403 may be the same element, that is, a transceiver, which provides a unit for communicating with various other devices on the transmission medium. The processor 402 is responsible for managing the bus 400 and general processing, and the memory 404 can be used to store data used by the processor 402 when performing operations. In a fourth aspect, based on the inventive concept of the sample attribute evaluation in the foregoing embodiment, the present invention also provides a computer-readable storage medium on which a computer program is stored, and when the program is executed by a processor, the above-mentioned sample attribute evaluation is realized The steps of any method of the method. This specification is described with reference to flowcharts and/or block diagrams of methods, equipment (systems), and computer program products according to the embodiments of this specification. It should be understood that each process and/or block in the flowchart and/or block diagram, and the combination of processes and/or blocks in the flowchart and/or block diagram can be realized by computer program instructions. These computer program instructions can be provided to the processors of general-purpose computers, dedicated computers, embedded processors, or other programmable data processing equipment to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing equipment can be used to generate It is a device that realizes the functions specified in one process or multiple processes in the flowchart and/or one block or multiple blocks in the block diagram. These computer program instructions can also be stored in a computer-readable memory that can guide a computer or other programmable data processing equipment to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product that includes the instruction device, The instruction device implements the functions specified in one process or multiple processes in the flowchart and/or one block or multiple blocks in the block diagram. These computer program instructions can also be loaded on a computer or other programmable data processing equipment, so that a series of operation steps are executed on the computer or other programmable equipment to generate computer-implemented processing, which can be executed on the computer or other programmable equipment. The instructions provide steps for implementing functions specified in one flow or multiple flows in the flowchart and/or one block or multiple blocks in the block diagram. Although the preferred embodiments of this specification have been described, those skilled in the art can make additional changes and modifications to these embodiments once they learn the basic progressive concepts. Therefore, the scope of the attached patent application is intended to be interpreted as including the preferred embodiments and all changes and modifications falling within the scope of this specification. Obviously, those skilled in the art can make various changes and modifications to this specification without departing from the spirit and scope of the specification. In this way, if these modifications and variations of this specification fall within the scope of the patent application for this specification and its equivalent technology, this specification is also intended to include these modifications and variations.

301: The first determination unit 302: The second determination unit 303: Training Unit 400: bus 401: Receiver 402: processor 403: Transmitter 404: Memory 406: bus interface

Fig. 1 is a schematic diagram of an application scenario of sample attribute evaluation in an embodiment of the specification; FIG. 2 is a flowchart of the method for evaluating the attributes of a sample in the first aspect of the embodiment of this specification; 3 is a schematic diagram of the structure of the training device for the sample attribute evaluation model of the second aspect of the embodiment of this specification; 4 is a schematic diagram of the structure of the sample attribute evaluation server in the third aspect of the embodiment of this specification.

Claims (15)

A sample attribute model training method includes: determining the black sample concentration of each community in a relationship graph corresponding to the training sample, wherein the training sample includes a black sample and an unknown sample; based on the black sample concentration of each community, The white sample sampling probability of each unknown sample is determined, and the white sample sampling probability of each unknown sample is sampled to obtain a white sample; based on a semi-supervised machine learning algorithm, the black sample and the white sample The sample is trained to obtain a sample attribute evaluation model; it is judged whether the sample attribute evaluation model meets the preset convergence condition; if not, the black sample concentration of each community is updated based on the updated black sample concentration of each community and The semi-supervised machine learning algorithm continues training until the sample attribute evaluation model obtained by training satisfies the preset convergence condition, and the sample attribute assessment model that satisfies the preset convergence condition is used as the target sample attribute assessment model, wherein The judging whether the sample attribute evaluation model satisfies a preset convergence condition includes: evaluating each of the unknown samples based on the sample attribute evaluation model, and obtaining the current round of attribute evaluation results of each of the unknown samples, in total Obtain M current-round attribute evaluation results, where M is the number of the unknown samples; based on the M current-round attribute evaluation results and M last-round attribute evaluation results, determine whether the sample attribute evaluation model meets preset convergence condition. According to the method described in item 1 of the scope of patent application, the determination of the black sample concentration of each community in the relationship graph corresponding to the training sample includes: determining the number of nodes corresponding to all black samples in each community among the total nodes of the community Take the first proportion as the black sample concentration of the community; or determine the second proportion of the nodes corresponding to all black samples in each community in the total nodes in the relationship graph, and set the first proportion The proportion is used as the black sample concentration of the community; or determine the third proportion of the nodes corresponding to all black samples in each community in the total nodes of the community, and the third proportion of the total nodes of the community in the total nodes in the relationship graph Four proportions, a weighted average of the third proportion and the fourth proportion is obtained, and the weighted average is used as the black sample concentration of the community. According to the method described in item 3 of the scope of patent application, the evaluation of each of the unknown samples based on the sample attribute evaluation model to obtain the current round of attribute evaluation results of each of the unknown samples includes: The sample attribute evaluation model evaluates each of the unknown samples to obtain a black sample score of each of the unknown samples, and if the black sample score value is greater than a preset score value, the attribute information of the unknown sample is marked as a black sample , Wherein the current round of attribute evaluation results of each of the unknown samples include the attribute information of the unknown sample. According to the method described in item 3 of the scope of patent application, the M current round of attribute evaluation results and M last round of attribute evaluation results, judging whether the sample attribute evaluation model satisfies the preset convergence condition, including: judging all of the current round of attribute evaluation results of each of the unknown samples Whether the attribute information is consistent with the attribute information in the previous round of attribute evaluation results of the unknown sample, if so, it indicates that the current round of sample attribute evaluation model satisfies the preset convergence condition. According to the method described in item 3 of the scope of patent application, said updating the concentration of black samples in each community includes: determining all the results based on the M current-round attribute evaluation results and the M last-round attribute evaluation results. The unknown sample whose attribute information has changed; and the black sample concentration of the community corresponding to the unknown sample whose attribute information has changed is recalculated. According to the method described in any one of items 1 to 5 in the scope of the patent application, the training sample is insurance data corresponding to the claim settlement personnel, and the black sample is the insurance data corresponding to the insurance fraud personnel. A sample attribute evaluation method includes: a target sample attribute evaluation model trained according to any one of the methods described in the scope of patent application 1 to 5, evaluates a new sample, and determines the evaluation result of the new sample, wherein , The evaluation result includes the black sample score and/or attribute information of the new sample. A sample attribute evaluation model training device includes: a first determining unit for determining the concentration of black samples in each community in a relationship graph corresponding to training samples, wherein the training samples include black samples and unknown samples; and second determining A unit for determining the white sample sampling probability of each unknown sample based on the black sample concentration of each community, and sampling with the white sample sampling probability of each unknown sample to obtain a white sample; training A unit for training the black sample and the white sample based on a semi-supervised machine learning algorithm to obtain a sample attribute evaluation model; judging whether the sample attribute evaluation model satisfies a preset convergence condition; if not, updating the The black sample concentration of each community is based on the updated black sample concentration of each community and the semi-supervised machine learning algorithm to continue training until the sample attribute evaluation model obtained by training satisfies the preset convergence condition, and The sample attribute evaluation model that satisfies the preset convergence condition is used as the target sample attribute evaluation model, wherein the training unit is specifically configured to: evaluate each of the unknown samples based on the sample attribute evaluation model to obtain each of the For the current round of attribute evaluation results of unknown samples, a total of M current round of attribute evaluation results are obtained, where M is the number of the unknown samples; based on the M current round of attribute evaluation results and M last round of attribute evaluation results, judge all The sample attribute evaluation model meets the preset convergence conditions. According to the device described in item 8 of the scope of patent application, the first determining unit is specifically configured to: determine the first proportion of all black sample corresponding nodes in each community in the total nodes of the community, and calculate the first proportion As the black sample concentration of the community; or determine the second proportion of all black sample corresponding nodes in each community in the total nodes in the relationship graph, and use the first proportion as the black sample concentration of the community; or Determine the third proportion of the nodes corresponding to all black samples in each community among the total nodes of the community, and the fourth proportion of the total nodes of the community among the total nodes in the relationship graph, to obtain the third proportion And the weighted average of the fourth proportion, and the weighted average is used as the black sample concentration of the community. According to the device described in item 8 of the scope of patent application, the training unit is specifically configured to: evaluate each of the unknown samples based on the sample attribute evaluation model, and obtain the black sample score of each of the unknown samples, if The black sample score value is greater than the preset score value, and the attribute information of the unknown sample is marked as a black sample, wherein the attribute information of the unknown sample is included in the current round of attribute evaluation results of each of the unknown samples. According to the device described in item 10 of the scope of patent application, the training unit is specifically used for: Determine whether the attribute information in the current round of attribute evaluation results of each of the unknown samples is consistent with the attribute information in the previous round of attribute evaluation results of the unknown sample. If so, it indicates that the current round of sample attribute evaluation model satisfies all requirements. The preset convergence conditions are described. According to the device described in item 10 of the scope of patent application, the training unit is specifically configured to: determine the unknown attribute information change based on the M current-round attribute evaluation results and the M last-round attribute evaluation results Sample; recalculate the black sample concentration of the community corresponding to the unknown sample whose attribute information has changed. A sample attribute evaluation device, comprising: an evaluation unit, which is used to evaluate a new sample according to a target sample attribute evaluation model trained by the device described in any one of items 8-12 of the scope of patent application, and determine the new sample The evaluation result of, wherein the evaluation result includes the black sample score and/or attribute information of the new sample. A server includes a memory, a processor, and a computer program stored on the memory and capable of running on the processor, and when the processor executes the program, any one of items 1-7 of the scope of patent application can be realized The steps of the method described in item. A computer-readable storage medium has a computer program stored thereon, and when the program is executed by a processor, the steps of the method described in any one of items 1-7 of the scope of the patent application are realized.

Images