CN111401914A - Risk assessment model training and risk assessment method and device - Google Patents
Risk assessment model training and risk assessment method and device Download PDFInfo
- Publication number
- CN111401914A CN111401914A CN202010256586.3A CN202010256586A CN111401914A CN 111401914 A CN111401914 A CN 111401914A CN 202010256586 A CN202010256586 A CN 202010256586A CN 111401914 A CN111401914 A CN 111401914A
- Authority
- CN
- China
- Prior art keywords
- risk
- user
- layer
- risk assessment
- feature coding
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q20/00—Payment architectures, schemes or protocols
- G06Q20/38—Payment protocols; Details thereof
- G06Q20/40—Authorisation, e.g. identification of payer or payee, verification of customer or shop credentials; Review and approval of payers, e.g. check credit lines or negative lists
- G06Q20/401—Transaction verification
- G06Q20/4016—Transaction verification involving fraud or risk level assessment in transaction processing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/24323—Tree-organised classifiers
Abstract
The embodiment of the specification provides a training method and a risk assessment method and device for a risk assessment model. Training the feature coding layer based on the user features of the user samples and the total risk scores in the risk labels to obtain a trained feature coding layer, and acquiring feature coding results of the user samples based on the trained feature coding layer. And training the comprehensive risk assessment layer based on the feature coding result of each user sample and the total risk score in the risk label. For each of the single risk assessment tiers, training the single risk assessment tier based on the feature coding results of the respective user sample and the individual risk scores of the corresponding risk types in the risk labels.
Description
Technical Field
One or more embodiments of the present disclosure relate to the field of machine learning technologies, and in particular, to a method and an apparatus for training a risk assessment model.
Background
With the increasing popularity of network technologies, the risks in network transactions are increasing. In order to effectively reduce high-risk transactions and enhance market monitoring, it is very important to evaluate risks for users.
Conventionally, risk assessments are typically made for only a single risk type (e.g., a risk type for a suspected gambling). The evaluation process may specifically be: and training a risk assessment model based on the sample user and the normal sample user of the risk type. And then, based on the risk assessment model, assessing whether the target user is the user with the corresponding risk type. Therefore, conventional risk assessment methods are inefficient.
It is therefore desirable to provide improved solutions that can improve the efficiency of risk assessment for users.
Disclosure of Invention
One or more embodiments of the present disclosure describe a risk assessment model training method, a risk assessment method, and a risk assessment model training device, which can reduce the cost and overhead of model training.
In a first aspect, a method for training a risk assessment model is provided, including:
collecting a collection of user samples, wherein each user sample comprises a user characteristic and a risk label; the risk label is for indicating a total risk score for the user and a plurality of individual risk scores corresponding to the respective predetermined risk types;
training the feature coding layer based on the user features of the user samples in the batch of user samples and the total risk scores in the risk labels of the user samples to obtain a trained feature coding layer;
acquiring a feature coding result of each user sample based on the trained feature coding layer;
training the comprehensive risk assessment layer based on the feature coding result of each user sample and the total risk score in the risk label of each user sample;
for each of the single risk assessment layers, training the single risk assessment layer based on the feature coding results of the user samples and the individual risk scores in the risk labels of the user samples corresponding to the risk type of the single risk assessment layer.
In a second aspect, a method for risk assessment is provided, comprising:
acquiring target characteristics of a user to be assessed for risk;
inputting the target characteristics into a characteristic coding layer of a risk assessment model to obtain a target characteristic coding result; the risk assessment model is obtained by training through the method of the first aspect;
respectively inputting the target feature coding results into a comprehensive risk evaluation layer and each single risk evaluation layer of the risk evaluation model;
obtaining the total risk score of the risk user to be evaluated through the output of the comprehensive risk evaluation layer; and obtaining the individual risk score of the risk user to be evaluated corresponding to the risk type of each single risk evaluation layer through the output of each single risk evaluation layer in each single risk evaluation layer.
In a third aspect, a training apparatus for a risk assessment model is provided, including:
the system comprises a collecting unit, a processing unit and a processing unit, wherein the collecting unit is used for collecting a batch of user samples, and each user sample comprises a user characteristic and a risk label; the risk label is for indicating a total risk score for the user and a plurality of individual risk scores corresponding to the respective predetermined risk types;
the training unit is used for training the feature coding layer based on the user features of the user samples in the batch of user samples collected by the collecting unit and the total risk scores in the risk labels of the user samples to obtain a trained feature coding layer;
the acquisition unit is used for acquiring the characteristic coding result of each user sample based on the characteristic coding layer trained by the training unit;
the training unit is further configured to train the comprehensive risk assessment layer based on the feature coding result of each user sample acquired by the acquisition unit and the total risk score in the risk label of each user sample;
the training sample is further configured to, for each of the single risk assessment layers, train the single risk assessment layer based on the feature coding result of the user sample acquired by the acquisition unit and the individual risk score in the risk label of the user sample corresponding to the risk type of the single risk assessment layer.
In a fourth aspect, a risk assessment device is provided, comprising:
the system comprises an acquisition unit, a processing unit and a processing unit, wherein the acquisition unit is used for acquiring target characteristics of a risk user to be evaluated;
the input unit is used for inputting the target characteristics acquired by the acquisition unit into a characteristic coding layer of a risk assessment model to obtain a target characteristic coding result; the risk assessment model is obtained by training through the method of the first aspect;
the input unit is further configured to input the target feature coding results into a comprehensive risk assessment layer and each single risk assessment layer of the risk assessment model respectively;
the acquisition unit is further used for obtaining the total risk score of the risk user to be evaluated through the output of the comprehensive risk evaluation layer; and obtaining the individual risk score of the risk user to be evaluated corresponding to the risk type of each single risk evaluation layer through the output of each single risk evaluation layer in each single risk evaluation layer.
In a fifth aspect, there is provided a computer storage medium having stored thereon a computer program which, when executed in a computer, causes the computer to perform the method of the first or second aspect.
In a sixth aspect, there is provided a computing device comprising a memory having stored therein executable code, and a processor that when executing the executable code, implements the method of the first or second aspect.
According to the risk assessment model training and risk assessment method and device provided by one or more embodiments of the specification, firstly, a task of assessing comprehensive risks of a user and a task of assessing each single risk can be jointly trained, so that the cost and the expense of model training can be reduced. Secondly, in the multitask joint training process, the user samples can be shared, and therefore the utilization rate of the user samples can be improved. Finally, the comprehensive risk assessment and the single risk assessment of the user can be simultaneously realized based on the model obtained by training, so that the risk assessment efficiency of the user can be greatly improved.
Drawings
In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the drawings needed to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present disclosure, and it is obvious for those skilled in the art to obtain other drawings based on the drawings without creative efforts.
FIG. 1 is one of the architectural diagrams of a risk assessment model provided herein;
FIG. 2 is a flowchart of a method for training a risk assessment model according to an embodiment of the present disclosure;
FIG. 3 is a second schematic diagram of a risk assessment model provided in the specification;
FIG. 4 is a flow chart of a risk assessment method provided by one embodiment of the present description;
FIG. 5 is a schematic diagram of a training apparatus for a risk assessment model according to an embodiment of the present disclosure;
fig. 6 is a schematic view of a risk assessment device according to an embodiment of the present disclosure.
Detailed Description
The scheme provided by the specification is described below with reference to the accompanying drawings.
Before describing the solution provided in the present specification, the inventive concept of the present solution will be explained below.
As mentioned above, in the conventional technology, the trained risk assessment model can only achieve assessment of a single risk type, that is, the risk assessment model trained by the conventional method is generally a single task model, which wastes training resources greatly. The applicant of the present application considers that there is often similarity between tasks that evaluate different risk types. For example, 48% of fraudulent users are high-risk users who violate the rule. Furthermore, there is a great correlation between the task of assessing the composite risk and the task of assessing each individual risk. For example, for a user with a higher individual score for a part of the risk types, the score for the overall risk is also higher. The combined risk is understood here as a combined consideration of the different risk types.
In view of this, the applicant of the present application considers establishing a risk assessment model that can simultaneously achieve multitask assessment. The plurality of tasks includes a task of evaluating the composite risk and a task of evaluating each first risk. Specifically, in the risk assessment model, the tasks share a feature coding layer, and each task corresponds to a different upper task layer. For example, the task of evaluating the integrated risk corresponds to the integrated risk evaluation layer, and the task of evaluating each single risk corresponds to the single risk evaluation layer. Furthermore, the risk assessment model is trained based on the same set of training samples.
It should be noted that, because a plurality of tasks can share the training sample and share the underlying network part (i.e., the feature coding layer), not only can the training resources be saved, but also the training efficiency can be improved.
The present invention has been made in view of the above-mentioned problems, and it is to be understood that the present invention is not limited to the above-mentioned embodiments.
Fig. 1 is a schematic diagram of a risk assessment model provided in the present specification. In fig. 1, the risk assessment model includes a feature coding layer, a comprehensive risk assessment layer, and several single risk assessment layers. The feature coding layer is an underlying network part shared by multiple tasks, and may also be referred to as a shared layer. The comprehensive risk assessment layer is an upper network part corresponding to a task for assessing comprehensive risks, and each single risk assessment layer is an upper network part corresponding to a task for assessing a risk type (also called a task for assessing a single risk). Each upper network portion may also be referred to herein as an upper task layer.
Specifically, target characteristics of a user to be assessed for risk may be obtained. And inputting the target features into a feature coding layer of the risk assessment model to obtain a target feature coding result. And respectively inputting the target feature coding results into a comprehensive risk evaluation layer and each single risk evaluation layer of the risk evaluation model. And obtaining the total risk score of the risk user to be evaluated through the output of the comprehensive risk evaluation layer. At this point, the task of assessing the composite risk is completed. And obtaining the individual risk score of the risk user to be evaluated corresponding to the risk type of each single risk evaluation layer through the output of each single risk evaluation layer in each single risk evaluation layer. The task of assessing each individual risk is now also completed.
The following describes a training method of a risk assessment model provided in the present specification, with reference to specific examples. Specifically, fig. 2 shows a flowchart of a training method of a risk assessment model provided in an embodiment of the present specification, where the architecture of the risk assessment model may refer to fig. 1, and an execution subject of the method may be a device with processing capability: a server or system or device, etc. As shown in fig. 2, the method may include the steps of:
at step 202, a collection of user samples is collected.
For the batch of user samples, user samples attributed to each of a plurality of risk types may be counted, such user samples collectively being referred to as black samples. In addition, for each risk type, a corresponding white sample may also be counted. The black and white samples for each risk type are collectively referred to below as user samples. The multiple risk types may include, but are not limited to, violation types, anti-cheating types, investment financing types, and fraud types, among others.
Further, each user sample in the collection of user samples may include a user characteristic and a risk label. The user characteristics may include, but are not limited to, several of user attributes, historical high-risk audit records, historical behavior records, and daily policy penalty results. For the user attributes, it may include, but is not limited to, user age, account balance, age, and the like. For the above-mentioned historical high-risk audit record, it may refer to whether the user was audited as a high-risk user in the past. For the above-mentioned daily policy penalty result, it may be an evaluation result of the user based on the corresponding risk policy on the day. Risk policies herein may include, but are not limited to, fraud policies, violation violations policies, and the like. For the above historical behavior record, it may refer to browsing record of the user, transaction record, and the like.
The risk label is used to indicate a total risk score for the user and a plurality of individual risk scores corresponding to each of the predetermined risk types. The value ranges of the total risk score and the individual risk scores may be: [0,1].
In one example, the risk labels may be represented in the form of a vector, and the dimensions of the vector may correspond to the number of tasks. For example, assume that the number of tasks is 5, and the 5 tasks are: the task of evaluating comprehensive risks, the task of evaluating violation and banning types, the task of evaluating anti-cheating types, the task of evaluating investment financing types and the task of evaluating cheating types. Furthermore, the method is simple. Further assuming that a user is a fraudulent user, the risk label of the user sample corresponding to the user may be: [1,0,0,0,1]. Wherein, the 1 st '1' represents the total risk score of the user, the 1 st '0' represents the individual risk score that the user belongs to the violation type, the 2 nd '0' represents the individual risk score that the user belongs to the anti-cheating type, the 3 rd '0' represents the individual risk score that the user belongs to the investment and financing type, and the 2 nd '1' represents the individual risk score that the user belongs to the cheating type.
And step 204, training the feature coding layer based on the user features of the user samples in the collected batch of user samples and the total risk scores in the risk labels of the user samples to obtain the trained feature coding layer.
In one example, the feature coding layer may be implemented using an iterative decision tree model. The iterative decision tree model herein may include, but is not limited to, a Gradient Boosting Decision Tree (GBDT) model, an adaboost decision tree model, and an XGBoost decision tree model.
For example, the XGBoost decision tree model is used in the feature coding layer, and the training process is a process of constructing a decision tree. The construction process is described in detail below:
first, a batch of user samples collected in step 202 above is taken as an initial training set and is expressed as: d1 ═ x(i),y(i)}NAnd N is the number of user samples, namely the number of users. x is the number of(i)For the ith user sample, it may be, for example, an S-dimensional vector, i.e., x ═ x (x1, x2, …, xs), where each xi characterizes one user feature of the user sample, y(i)The total risk score of the risk label for the ith user sample. Then, the N user samples are segmented by the first decision tree. Splitting features and feature thresholds are set at each parent node of the decision tree (e.g., based on the principle that the sum of information gains of feature splitting is the largest, a target feature and feature value are selected from a plurality of user features of a user sample as the splitting features and feature thresholds of the parent node), and the user sample is split into corresponding child nodes by comparing the corresponding features of the user sample with the feature thresholds at the parent node. Through such a process, the N user samples are finally segmented into respective leaf nodes. And the score of each leaf node is the mean value of the total risk scores of the user samples in the leaf node.
After the first decision tree is obtained, subtracting the total risk score of the user sample from the score of the leaf node of the user sample in the first decision tree to obtain the residual error r of each user sample(i)D2 ═ x(i),r(i)}NWhich corresponds to the same user sample as D1 for the new training set. In the same way as above, a second decision tree may be obtained, in which the N user samples are divided into leaf nodes, and the score of each leaf node is the mean of the residual values of the user samples. Similarly, multiple decision trees may be taken sequentially, each decision tree being taken based on the residuals of the previous decision tree. An XGBoost model comprising multiple decision trees may thus be obtained. That is, the trained feature coding layer may include a plurality of decision trees.
It should be noted that, as can be seen from the above process of constructing the decision tree, each parent node (i.e., non-leaf node) of each decision tree in each constructed decision tree corresponds to one user feature (i.e., split feature). Furthermore, at each parent node of the decision tree, the user samples are segmented into respective child nodes by comparing the corresponding features of the user samples to be segmented with feature thresholds. For example, when the corresponding feature is greater than the feature threshold, the corresponding feature is segmented into left child nodes; and when the corresponding characteristic is not larger than the characteristic threshold value, the corresponding characteristic is divided into right child nodes. Thus, each segmentation of the user sample can also be understood as a process of normalizing the corresponding feature, for example, when the corresponding feature is greater than the feature threshold, normalizing the corresponding feature to 0 (or 1); and when the corresponding feature is not greater than the feature threshold, it is normalized to 1 (or 0).
It can be understood that, in a decision tree, after a user sample is divided into leaf nodes, a normalization processing result of a partial feature of the user sample (i.e., a user feature corresponding to each node covered by a path from the leaf node to a root node of the decision tree where the leaf node is located) can be obtained generally. In all decision trees, after the user sample is divided into leaf nodes, the normalization processing results of all the characteristics of the user sample can be obtained.
In one example, in the decision tree Y, the user sample X corresponds to the normalized processing result of the decision tree Y and can be characterized by the node identification of the leaf node in which it falls. For example, when the node identifier of the leaf node where the user sample X falls in the decision tree Y is "01", the representation is normalized in the decision tree Y first and then for two features of the user sample X, and the result of the normalization process for one feature is: "0", and the normalized result for another feature is: "1". And the two characteristics can be determined based on the splitting characteristics of each parent node covered by the path from the leaf node to the root node of the decision tree.
And step 206, acquiring the feature coding result of each user sample based on the trained feature coding layer.
Here, the feature coding result obtained from each user sample is a normalization processing result of each user feature obtained from each user sample.
In an example, in a case that the node identifier of the leaf node where the user sample falls in one decision tree may characterize a normalization processing result of the user sample corresponding to the one decision tree, the step of obtaining the feature coding result of each user sample based on the trained feature coding layer may include: and determining the node identification of the leaf node of each user sample in the plurality of decision trees, wherein each decision falls into. And determining the feature coding result of each user sample based on the node identification of the leaf node of each user sample in the multiple decision trees.
For example, assume that the trained feature coding layer includes three decision trees, which are in the order of a1, a2, and A3, each decision tree includes 4 leaf nodes, and the node identifiers of the 4 leaf nodes are: "00", "01", "10", and "11". Assume that the node identification of the leaf node into which any first user sample falls in A1, A2, and A3, respectively, is: "10", "01", and "10", the feature encoding result of the first user sample may be: 100110.
it should be understood that the above is only an exemplary illustration, and in practical applications, after the node identifiers of the leaf nodes are determined, the node identifiers of the leaf nodes may be further processed, for example, expanded to a predetermined number of bits, and the like, which is not limited in this specification.
Regarding the node identifier, since it is represented by binary data, the number of bits (represented by N) included in the node identifier can be adjusted according to the number of leaf nodes included in each decision tree. The regulation principle is that the number of the node identifications (namely 2) can be expressedN) Greater than or equal to the number of leaf nodes. For example, when a decision tree includes 5 leaf nodes, the node identifier of each leaf node may include 3 bits, i.e., 2 bits3>5。
And step 208, training the comprehensive risk assessment layer based on the feature coding result of each user sample and the total risk score in the risk label of each user sample.
In one example, the integrated risk assessment layer described above may also be implemented using an iterative decision tree model. The training process is similar to that of the feature coding layer, and is not repeated herein.
The feature encoding result of each user sample is a normalization processing result of each user feature of each user sample. Therefore, the training comprehensive risk assessment layer is only different from the training feature coding layer in that the user features of the user samples are subjected to normalization processing. Since each user characteristic after normalization is within a predetermined range (e.g., [0,1]), the complexity of training samples can be reduced, and the training efficiency can be improved.
Step 210, for each of the single risk assessment layers, training the single risk assessment layer based on the feature coding result of the user sample and the individual risk score corresponding to the risk type of the single risk assessment layer in the risk label of the user sample.
In one example, each of the above-described single risk assessment layers may also be implemented using an iterative decision tree model. The training process is similar to that of the feature coding layer as a whole, and only the following two differences exist:
first, the feature encoding result of each user sample is a normalization processing result of each user feature of each user sample. Thus, the user characteristics of the user samples on which each of the single risk assessment layers is trained are normalized. Since each user characteristic after normalization is within a predetermined range (e.g., [0,1]), the complexity of training samples can be reduced, and the training efficiency can be improved. Second, when each of the individual risk assessment layers is trained, a separate risk score for the corresponding risk type in the risk label for each user sample is used.
It should be understood that in practical applications, the above step 208 and step 210 may be executed in parallel. In addition, the above steps 202 to 210 may be performed iteratively until an iteration stop condition is satisfied, for example, the number of iterations reaches a predetermined number, and the like.
In summary, the feature coding layer, the comprehensive risk assessment layer, and each single risk assessment layer in the risk assessment model described in this specification may be implemented by using an iterative decision tree model, that is, the risk assessment model trained in the embodiments of this specification has a tree non-transformation capability, so that it can handle more complex tasks. In addition, under the condition that the feature coding layer adopts the XGboost model, the feature coding result obtained based on the model is input into the comprehensive risk evaluation layer and each single risk evaluation layer, so that some work of feature engineering can be saved, and the user features can be accurately expressed. Finally, under the condition that each layer of the wind control evaluation model adopts the XGboost model, the risk evaluation model can have certain interpretability. For example, based on nodes covered by paths from leaf nodes to root nodes in a decision tree, a corresponding feature combination can be determined, and the feature combination can be used for interpreting the final output total risk score and each individual risk score.
In addition, the scheme can realize the joint training of the task for evaluating the comprehensive risk of the user and the task for evaluating each single risk, namely, the scheme makes full use of the thought and the method of multi-task learning to combine the evaluations of different risk types together in a multi-task mode, so that the cost and the expense of model training can be reduced. Secondly, in the multi-task joint training process, user samples can be shared, so that the utilization rate of the user samples can be improved, and the problem that a corresponding risk evaluation model cannot be effectively trained due to the fact that a certain risk type of user samples is few can be solved.
The following describes a training process of the risk assessment model provided in the embodiment of the present disclosure with reference to fig. 3.
Fig. 3 is a second schematic diagram of the architecture of the risk assessment model provided in the present specification, and in fig. 3, each of the feature coding layer, the comprehensive risk assessment layer, and each of the single risk assessment layers employs an iterative decision tree model, such as an XGBoost decision tree model. The specific training process may be: a collection of user samples is collected, each of which includes a user characteristic and a risk label, wherein the risk label is indicative of a total risk score for the user and a plurality of individual risk scores corresponding to respective predetermined risk types. And then, training the feature coding layer based on the user features of all the collected user samples in the batch of user samples and the total risk scores in the risk labels of all the user samples, namely constructing all decision trees in an XGboost decision tree model corresponding to the feature coding layer. After the construction of each decision tree is completed, determining a feature coding result of each user sample based on the node identification of the leaf node of each user sample in which each constructed decision falls.
Then, the comprehensive risk evaluation layer can be trained based on the feature coding result of each user sample and the total risk score in the risk label of each user sample, that is, each decision tree in the XGBoost decision tree model corresponding to the comprehensive risk evaluation layer is constructed. In addition, for each of the individual risk assessment layers, the risk assessment layer may be trained based on the feature-coded results of the individual user samples and the individual risk scores in the risk labels of the individual user samples that correspond to the risk type of the risk assessment layer. And constructing each decision tree in the XGboost decision tree model corresponding to the single risk assessment layer. At this point, one training process of the risk assessment model is finished.
Similarly, the risk assessment model may be iteratively trained multiple times until an iteration stop condition is satisfied.
After the risk assessment model is obtained through training, risk assessment can be performed on the user based on the risk assessment model, and a specific assessment method of the risk assessment model can be as shown in fig. 4. In fig. 4, the method may include the steps of:
step 402, obtaining target characteristics of a user to be assessed with risk.
The objective characteristics may refer to the user characteristics, and may include, but are not limited to, user attributes, historical high-risk audit records, historical behavior records, and daily policy penalty results.
And step 404, inputting the target features into a feature coding layer of the risk assessment model to obtain a target feature coding result.
Taking fig. 3 as an example, the target feature coding result of the risk user to be evaluated may be determined based on the node identifier of the leaf node where each decision of the risk user to be evaluated falls in the multiple decision trees.
And 406, inputting the target feature coding results into the comprehensive risk evaluation layer and each single risk evaluation layer of the risk evaluation model respectively.
Taking fig. 3 as an example, the target feature coding result may be input into the decision input model corresponding to the comprehensive risk assessment layer, and the target feature coding result may be input into the decision tree model corresponding to each single risk assessment layer.
And step 408, obtaining the total risk score of the risk user to be evaluated through the output of the comprehensive risk evaluation layer. And obtaining the individual risk score of the risk user to be evaluated corresponding to the risk type of each single risk evaluation layer through the output of each single risk evaluation layer in each single risk evaluation layer.
That is, the risk assessment method provided by the present description may obtain both the total risk score of the user and the individual risk score corresponding to each of the plurality of risk types. The individual risk scores for each risk type herein can be referenced to the overall risk score to allow a more in-depth dimension to be observed for the user. On the other hand, the risk assessment method and the risk assessment system can assess various risk types of the user at the same time, and can greatly improve the risk assessment efficiency of the user.
Corresponding to the training method of the risk assessment model, an embodiment of the present specification further provides a training device of the risk assessment model. The risk assessment model comprises a feature coding layer, a comprehensive risk assessment layer and a plurality of single risk assessment layers. Each of the individual risk assessment layers corresponds to one of the predetermined risk types. As shown in fig. 5, the apparatus may include:
a collecting unit 502 for collecting a collection of user samples, wherein each user sample comprises a user characteristic and a risk label. The risk label is used to indicate a total risk score for the user and a plurality of individual risk scores corresponding to each of the predetermined risk types.
Each predetermined risk type herein may include several of a violation type, an anti-cheating type, an investment financing type, and a fraud type. Further, the user characteristics may include several of user attributes, historical high-risk audit records, historical behavior records, and daily policy penalty results.
The training unit 504 is configured to train the feature coding layer based on the user features of each user sample in the batch of user samples collected by the collection unit 502 and the total risk score in the risk label of each user sample, so as to obtain a trained feature coding layer.
An obtaining unit 506, configured to obtain a feature coding result of each user sample based on the feature coding layer trained by the training unit 504.
Here, the trained feature coding layer may include a plurality of decision trees, and the obtaining unit 506 may be specifically configured to:
and determining the node identification of the leaf node of each user sample in the plurality of decision trees, wherein each decision falls into.
And determining the feature coding result of each user sample based on the node identification of the leaf node of each user sample in the multiple decision trees.
Each non-leaf node of each of the decision trees corresponds to a user characteristic. The node identification of the leaf node of each decision tree represents the normalization processing result of the user characteristics corresponding to each node covered by the path from the leaf node to the root node of the decision tree where the leaf node is located.
The training unit 504 is further configured to train the comprehensive risk assessment layer based on the feature coding result of each user sample acquired by the acquiring unit 506 and the total risk score in the risk label of each user sample.
The training sample 504 is further configured to, for each of the single risk assessment layers, train the single risk assessment layer based on the feature coding result of the user sample obtained by the obtaining unit 506 and the individual risk score corresponding to the risk type of the single risk assessment layer in the risk label of the user sample.
The integrated risk assessment layer and each of the single risk assessment layers include an iterative decision tree model. The iterative decision tree model herein may include any of the following: a gradient boosting decision tree GBDT model, an adaboost decision tree model and an XGboost decision tree model.
The functions of each functional module of the device in the above embodiments of the present description may be implemented through each step of the above method embodiments, and therefore, a specific working process of the device provided in one embodiment of the present description is not repeated herein.
The training device for the risk assessment model provided by one embodiment of the specification can reduce the cost and expense of model training.
Corresponding to the risk assessment method, an embodiment of the present specification further provides a risk assessment apparatus, as shown in fig. 6, the apparatus may include:
an obtaining unit 602, configured to obtain a target feature of a user to be assessed for risk.
An input unit 604, configured to input the target feature acquired by the acquisition unit 602 into a feature coding layer of the risk assessment model, so as to obtain a target feature coding result.
The risk assessment model can be trained by the method shown in fig. 2.
The input unit 604 is further configured to input the target feature coding results into the comprehensive risk assessment layer and each single risk assessment layer of the risk assessment model, respectively.
The obtaining unit 602 is further configured to obtain a total risk score of the risk user to be evaluated through the output of the comprehensive risk assessment layer, and obtain an individual risk score, corresponding to the risk type of each single risk assessment layer, of the risk user to be evaluated through the output of each single risk assessment layer in each single risk assessment layer.
The functions of each functional module of the device in the above embodiments of the present description may be implemented through each step of the above method embodiments, and therefore, a specific working process of the device provided in one embodiment of the present description is not repeated herein.
The risk assessment device provided by one embodiment of the specification can greatly improve the assessment efficiency of the user risk.
In another aspect, embodiments of the present specification provide a computer-readable storage medium having stored thereon a computer program, which, when executed in a computer, causes the computer to perform the method shown in fig. 2 or fig. 4.
In another aspect, embodiments of the present description provide a computing device comprising a memory having stored therein executable code, and a processor that, when executing the executable code, implements the method illustrated in fig. 2 or fig. 4.
The embodiments in the present specification are described in a progressive manner, and the same and similar parts among the embodiments are referred to each other, and each embodiment focuses on the differences from the other embodiments. In particular, for the apparatus embodiment, since it is substantially similar to the method embodiment, the description is relatively simple, and for the relevant points, reference may be made to the partial description of the method embodiment.
The steps of a method or algorithm described in connection with the disclosure herein may be embodied in hardware or may be embodied in software instructions executed by a processor. The software instructions may consist of corresponding software modules that may be stored in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art. An exemplary storage medium is coupled to the processor such the processor can read information from, and write information to, the storage medium. Of course, the storage medium may also be integral to the processor. The processor and the storage medium may reside in an ASIC. Additionally, the ASIC may reside in a server. Of course, the processor and the storage medium may reside as discrete components in a server.
Those skilled in the art will recognize that, in one or more of the examples described above, the functions described in this invention may be implemented in hardware, software, firmware, or any combination thereof. When implemented in software, the functions may be stored on or transmitted over as one or more instructions or code on a computer-readable medium. Computer-readable media includes both computer storage media and communication media including any medium that facilitates transfer of a computer program from one place to another. A storage media may be any available media that can be accessed by a general purpose or special purpose computer.
The foregoing description has been directed to specific embodiments of this disclosure. Other embodiments are within the scope of the following claims. In some cases, the actions or steps recited in the claims may be performed in a different order than in the embodiments and still achieve desirable results. In addition, the processes depicted in the accompanying figures do not necessarily require the particular order shown, or sequential order, to achieve desirable results. In some embodiments, multitasking and parallel processing may also be possible or may be advantageous.
The above-mentioned embodiments, objects, technical solutions and advantages of the present specification are further described in detail, it should be understood that the above-mentioned embodiments are only specific embodiments of the present specification, and are not intended to limit the scope of the present specification, and any modifications, equivalent substitutions, improvements and the like made on the basis of the technical solutions of the present specification should be included in the scope of the present specification.
Claims (18)
1. A training method of a risk assessment model comprises a characteristic coding layer, a comprehensive risk assessment layer and a plurality of single risk assessment layers; each of the single risk assessment tiers corresponding to one of the predetermined risk types; the method comprises the following steps:
collecting a collection of user samples, wherein each user sample comprises a user characteristic and a risk label; the risk label is for indicating a total risk score for the user and a plurality of individual risk scores corresponding to the respective predetermined risk types;
training the feature coding layer based on the user features of the user samples in the batch of user samples and the total risk scores in the risk labels of the user samples to obtain a trained feature coding layer;
acquiring a feature coding result of each user sample based on the trained feature coding layer;
training the comprehensive risk assessment layer based on the feature coding result of each user sample and the total risk score in the risk label of each user sample;
for each of the single risk assessment layers, training the single risk assessment layer based on the feature coding results of the user samples and the individual risk scores in the risk labels of the user samples corresponding to the risk type of the single risk assessment layer.
2. The method of claim 1, the trained feature coding layer comprising a plurality of decision trees; the obtaining the feature coding result of each user sample based on the feature coding layer obtained by training includes:
determining node identifiers of leaf nodes into which each decision of the user samples in the plurality of decision trees falls;
and determining a feature coding result of each user sample based on the node identification of the leaf node in which each decision of each user sample in the multiple decision trees falls.
3. The method of claim 2, wherein each non-leaf node of each of the decision trees corresponds to a user feature; and the node identification of the leaf node of each decision tree represents the normalization processing result of the user characteristics corresponding to each node covered by the path from the leaf node to the root node of the decision tree where the leaf node is located.
4. The method of claim 1, the integrated risk assessment layer and each of the individual risk assessment layers comprising an iterative decision tree model.
5. The method of claim 4, the iterative decision tree model comprising any of: a gradient boosting decision tree GBDT model, an adaboost decision tree model and an XGboost decision tree model.
6. The method of claim 1, the predetermined risk types comprising several of a violation type, an anti-cheating type, an investment financing type, and a fraud type.
7. The method of any of claims 1-6, the user characteristics comprising several of user attributes, historical high-risk audit logs, historical behavior logs, and daily policy penalty results.
8. A method of risk assessment, comprising:
acquiring target characteristics of a user to be assessed for risk;
inputting the target characteristics into a characteristic coding layer of a risk assessment model to obtain a target characteristic coding result; the risk assessment model is trained by the method of any one of claims 1-7;
respectively inputting the target feature coding results into a comprehensive risk evaluation layer and each single risk evaluation layer of the risk evaluation model;
obtaining the total risk score of the risk user to be evaluated through the output of the comprehensive risk evaluation layer; and obtaining the individual risk score of the risk user to be evaluated corresponding to the risk type of each single risk evaluation layer through the output of each single risk evaluation layer in each single risk evaluation layer.
9. A training device for a risk assessment model comprises a characteristic coding layer, a comprehensive risk assessment layer and a plurality of single risk assessment layers; each of the single risk assessment tiers corresponding to one of the predetermined risk types; the device comprises:
the system comprises a collecting unit, a processing unit and a processing unit, wherein the collecting unit is used for collecting a batch of user samples, and each user sample comprises a user characteristic and a risk label; the risk label is for indicating a total risk score for the user and a plurality of individual risk scores corresponding to the respective predetermined risk types;
the training unit is used for training the feature coding layer based on the user features of the user samples in the batch of user samples collected by the collecting unit and the total risk scores in the risk labels of the user samples to obtain a trained feature coding layer;
the acquisition unit is used for acquiring the characteristic coding result of each user sample based on the characteristic coding layer trained by the training unit;
the training unit is further configured to train the comprehensive risk assessment layer based on the feature coding result of each user sample acquired by the acquisition unit and the total risk score in the risk label of each user sample;
the training sample is further configured to, for each of the single risk assessment layers, train the single risk assessment layer based on the feature coding result of the user sample acquired by the acquisition unit and the individual risk score in the risk label of the user sample corresponding to the risk type of the single risk assessment layer.
10. The apparatus of claim 9, the trained feature coding layer comprising a plurality of decision trees; the obtaining unit is specifically configured to:
determining node identifiers of leaf nodes into which each decision of the user samples in the plurality of decision trees falls;
and determining a feature coding result of each user sample based on the node identification of the leaf node in which each decision of each user sample in the multiple decision trees falls.
11. The apparatus of claim 10, each non-leaf node of each of the decision trees corresponding to a user feature; and the node identification of the leaf node of each decision tree represents the normalization processing result of the user characteristics corresponding to each node covered by the path from the leaf node to the root node of the decision tree where the leaf node is located.
12. The apparatus of claim 9, the integrated risk assessment layer and each of the individual risk assessment layers comprising an iterative decision tree model.
13. The apparatus of claim 12, the iterative decision tree model comprising any of: a gradient boosting decision tree GBDT model, an adaboost decision tree model and an XGboost decision tree model.
14. The apparatus of claim 9, the predetermined risk types comprising several of a violation type, an anti-cheating type, an investment financing type, and a fraud type.
15. The apparatus of any of claims 9-14, the user characteristics comprising several of user attributes, historical high-risk audit logs, historical behavior logs, and daily policy penalty results.
16. A risk assessment device comprising:
the system comprises an acquisition unit, a processing unit and a processing unit, wherein the acquisition unit is used for acquiring target characteristics of a risk user to be evaluated;
the input unit is used for inputting the target characteristics acquired by the acquisition unit into a characteristic coding layer of a risk assessment model to obtain a target characteristic coding result; the risk assessment model is trained by the method of any one of claims 1-7;
the input unit is further configured to input the target feature coding results into a comprehensive risk assessment layer and each single risk assessment layer of the risk assessment model respectively;
the acquisition unit is further used for obtaining the total risk score of the risk user to be evaluated through the output of the comprehensive risk evaluation layer; and obtaining the individual risk score of the risk user to be evaluated corresponding to the risk type of each single risk evaluation layer through the output of each single risk evaluation layer in each single risk evaluation layer.
17. A computer-readable storage medium, on which a computer program is stored which, when executed in a computer, causes the computer to perform the method of any one of claims 1-7 or the method of claim 8.
18. A computing device comprising a memory having executable code stored therein and a processor that, when executing the executable code, implements the method of any of claims 1-7 or the method of claim 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010256586.3A CN111401914B (en) | 2020-04-02 | 2020-04-02 | Risk assessment model training and risk assessment method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010256586.3A CN111401914B (en) | 2020-04-02 | 2020-04-02 | Risk assessment model training and risk assessment method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111401914A true CN111401914A (en) | 2020-07-10 |
| CN111401914B CN111401914B (en) | 2022-07-22 |
Family
ID=71429371
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010256586.3A Active CN111401914B (en) | 2020-04-02 | 2020-04-02 | Risk assessment model training and risk assessment method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111401914B (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111738870A (en) * | 2020-07-28 | 2020-10-02 | 工保科技(浙江)有限公司 | Method and platform for identifying insurance risk of engineering performance guarantee based on characteristic engineering |
| CN111898675A (en) * | 2020-07-30 | 2020-11-06 | 北京云从科技有限公司 | Credit wind control model generation method and device, scoring card generation method, machine readable medium and equipment |
| CN111967807A (en) * | 2020-10-23 | 2020-11-20 | 支付宝(杭州)信息技术有限公司 | Method and device for generating risk event judgment rule executed by computer |
| CN112330432A (en) * | 2020-11-10 | 2021-02-05 | 中国平安人寿保险股份有限公司 | Risk level recognition model training method, recognition method, terminal and storage medium |
| CN112330035A (en) * | 2020-11-10 | 2021-02-05 | 支付宝(杭州)信息技术有限公司 | Training method and device of risk prediction model |
| CN112785157A (en) * | 2021-01-22 | 2021-05-11 | 支付宝(杭州)信息技术有限公司 | Risk identification system updating method and device and risk identification method and device |
| CN113610354A (en) * | 2021-07-15 | 2021-11-05 | 北京淇瑀信息科技有限公司 | Policy distribution method and device for third-party platform user and electronic equipment |
| CN114119037A (en) * | 2022-01-24 | 2022-03-01 | 深圳尚米网络技术有限公司 | Marketing anti-cheating system based on big data |
| CN114422269A (en) * | 2022-03-28 | 2022-04-29 | 北京源堡科技有限公司 | Network security assessment method and system based on machine learning |
| CN115935265A (en) * | 2023-03-03 | 2023-04-07 | 支付宝(杭州)信息技术有限公司 | Method for training risk recognition model, risk recognition method and corresponding device |
| CN116028820A (en) * | 2023-03-20 | 2023-04-28 | 支付宝(杭州)信息技术有限公司 | Model training method and device, storage medium and electronic equipment |
Citations (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150242858A1 (en) * | 2014-02-24 | 2015-08-27 | Bank Of America Corporation | Risk Assessment On A Transaction Level |
| CN105550931A (en) * | 2015-11-20 | 2016-05-04 | 长沙星际泛函网络科技有限公司 | Internet asset securitization risk management system |
| CN107481004A (en) * | 2017-08-11 | 2017-12-15 | 中国工商银行股份有限公司 | Transaction risk crime prevention system and method |
| US20180005235A1 (en) * | 2016-06-29 | 2018-01-04 | Ca, Inc. | Electronic transaction risk assessment based on digital identifier trust evaluation |
| CN107818513A (en) * | 2017-11-24 | 2018-03-20 | 泰康保险集团股份有限公司 | Methods of risk assessment and device, storage medium, electronic equipment |
| CN108062629A (en) * | 2017-12-26 | 2018-05-22 | 平安科技（深圳）有限公司 | Processing method, terminal device and the medium of transaction event |
| CN109242499A (en) * | 2018-09-19 | 2019-01-18 | 中国银行股份有限公司 | A kind of processing method of transaction risk prediction, apparatus and system |
| CN109272323A (en) * | 2018-09-14 | 2019-01-25 | 阿里巴巴集团控股有限公司 | A kind of risk trade recognition methods, device, equipment and medium |
| CN109308616A (en) * | 2018-08-29 | 2019-02-05 | 阿里巴巴集团控股有限公司 | A kind of risk determination method and device of transaction record |
| CN109559214A (en) * | 2017-09-27 | 2019-04-02 | 阿里巴巴集团控股有限公司 | Virtual resource allocation, model foundation, data predication method and device |
| CN109948932A (en) * | 2019-03-20 | 2019-06-28 | 杭州虹晟信息科技有限公司 | Risk monitoring and control analysis system based on big data |
| CN110008349A (en) * | 2019-02-01 | 2019-07-12 | 阿里巴巴集团控股有限公司 | The method and device for the event risk assessment that computer executes |
| CN110245941A (en) * | 2019-04-25 | 2019-09-17 | 阿里巴巴集团控股有限公司 | A kind of transaction risk recognition methods and device |
| CN110534200A (en) * | 2019-09-09 | 2019-12-03 | 吉林大学中日联谊医院 | A kind of efficient physical examination risk management and control system of personnel |
| CN110874743A (en) * | 2019-10-11 | 2020-03-10 | 支付宝(杭州)信息技术有限公司 | Method and device for determining account transaction risk |
-
2020
- 2020-04-02 CN CN202010256586.3A patent/CN111401914B/en active Active
Patent Citations (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150242858A1 (en) * | 2014-02-24 | 2015-08-27 | Bank Of America Corporation | Risk Assessment On A Transaction Level |
| CN105550931A (en) * | 2015-11-20 | 2016-05-04 | 长沙星际泛函网络科技有限公司 | Internet asset securitization risk management system |
| US20180005235A1 (en) * | 2016-06-29 | 2018-01-04 | Ca, Inc. | Electronic transaction risk assessment based on digital identifier trust evaluation |
| CN107481004A (en) * | 2017-08-11 | 2017-12-15 | 中国工商银行股份有限公司 | Transaction risk crime prevention system and method |
| CN109559214A (en) * | 2017-09-27 | 2019-04-02 | 阿里巴巴集团控股有限公司 | Virtual resource allocation, model foundation, data predication method and device |
| CN107818513A (en) * | 2017-11-24 | 2018-03-20 | 泰康保险集团股份有限公司 | Methods of risk assessment and device, storage medium, electronic equipment |
| CN108062629A (en) * | 2017-12-26 | 2018-05-22 | 平安科技（深圳）有限公司 | Processing method, terminal device and the medium of transaction event |
| CN109308616A (en) * | 2018-08-29 | 2019-02-05 | 阿里巴巴集团控股有限公司 | A kind of risk determination method and device of transaction record |
| CN109272323A (en) * | 2018-09-14 | 2019-01-25 | 阿里巴巴集团控股有限公司 | A kind of risk trade recognition methods, device, equipment and medium |
| CN109242499A (en) * | 2018-09-19 | 2019-01-18 | 中国银行股份有限公司 | A kind of processing method of transaction risk prediction, apparatus and system |
| CN110008349A (en) * | 2019-02-01 | 2019-07-12 | 阿里巴巴集团控股有限公司 | The method and device for the event risk assessment that computer executes |
| CN109948932A (en) * | 2019-03-20 | 2019-06-28 | 杭州虹晟信息科技有限公司 | Risk monitoring and control analysis system based on big data |
| CN110245941A (en) * | 2019-04-25 | 2019-09-17 | 阿里巴巴集团控股有限公司 | A kind of transaction risk recognition methods and device |
| CN110534200A (en) * | 2019-09-09 | 2019-12-03 | 吉林大学中日联谊医院 | A kind of efficient physical examination risk management and control system of personnel |
| CN110874743A (en) * | 2019-10-11 | 2020-03-10 | 支付宝(杭州)信息技术有限公司 | Method and device for determining account transaction risk |
Non-Patent Citations (2)
| Title |
|---|
| JUNCAI ZHANG 等: "Risk Assessment of Mobile Payment System Security Based on Extension Theory", 《2012 INTERNATIONAL CONFERENCE ON COMPUTER SCIENCE AND SERVICE SYSTEM》 * |
| 谢陈昕: "P2P网贷平台借款人信用风险评估模型适应性研究", 《武汉金融》 * |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111738870A (en) * | 2020-07-28 | 2020-10-02 | 工保科技(浙江)有限公司 | Method and platform for identifying insurance risk of engineering performance guarantee based on characteristic engineering |
| CN111898675A (en) * | 2020-07-30 | 2020-11-06 | 北京云从科技有限公司 | Credit wind control model generation method and device, scoring card generation method, machine readable medium and equipment |
| CN111898675B (en) * | 2020-07-30 | 2021-04-23 | 北京云从科技有限公司 | Credit wind control model generation method and device, scoring card generation method, machine readable medium and equipment |
| CN111967807A (en) * | 2020-10-23 | 2020-11-20 | 支付宝(杭州)信息技术有限公司 | Method and device for generating risk event judgment rule executed by computer |
| CN112330432B (en) * | 2020-11-10 | 2024-03-15 | 中国平安人寿保险股份有限公司 | Risk level identification model training method, risk level identification method, terminal and storage medium |
| CN112330432A (en) * | 2020-11-10 | 2021-02-05 | 中国平安人寿保险股份有限公司 | Risk level recognition model training method, recognition method, terminal and storage medium |
| CN112330035A (en) * | 2020-11-10 | 2021-02-05 | 支付宝(杭州)信息技术有限公司 | Training method and device of risk prediction model |
| CN112785157A (en) * | 2021-01-22 | 2021-05-11 | 支付宝(杭州)信息技术有限公司 | Risk identification system updating method and device and risk identification method and device |
| WO2022156701A1 (en) * | 2021-01-22 | 2022-07-28 | 支付宝(杭州)信息技术有限公司 | Risk recognition and update of risk recognition system |
| CN113610354A (en) * | 2021-07-15 | 2021-11-05 | 北京淇瑀信息科技有限公司 | Policy distribution method and device for third-party platform user and electronic equipment |
| CN114119037A (en) * | 2022-01-24 | 2022-03-01 | 深圳尚米网络技术有限公司 | Marketing anti-cheating system based on big data |
| CN114422269A (en) * | 2022-03-28 | 2022-04-29 | 北京源堡科技有限公司 | Network security assessment method and system based on machine learning |
| CN115935265A (en) * | 2023-03-03 | 2023-04-07 | 支付宝(杭州)信息技术有限公司 | Method for training risk recognition model, risk recognition method and corresponding device |
| CN116028820A (en) * | 2023-03-20 | 2023-04-28 | 支付宝(杭州)信息技术有限公司 | Model training method and device, storage medium and electronic equipment |
| CN116028820B (en) * | 2023-03-20 | 2023-07-04 | 支付宝(杭州)信息技术有限公司 | Model training method and device, storage medium and electronic equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111401914B (en) | 2022-07-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111401914B (en) | Risk assessment model training and risk assessment method and device | |
| US11915104B2 (en) | Normalizing text attributes for machine learning models | |
| CN110956273A (en) | Credit scoring method and system integrating multiple machine learning models | |
| CN112711953A (en) | Text multi-label classification method and system based on attention mechanism and GCN | |
| CN111507470A (en) | Abnormal account identification method and device | |
| CN112819604A (en) | Personal credit evaluation method and system based on fusion neural network feature mining | |
| CN110991474A (en) | Machine learning modeling platform | |
| CN112199512B (en) | Scientific and technological service-oriented case map construction method, device, equipment and storage medium | |
| CN111199469A (en) | User payment model generation method and device and electronic equipment | |
| CN112712383A (en) | Potential user prediction method, device, equipment and storage medium of application program | |
| CN110929525A (en) | Network loan risk behavior analysis and detection method, device, equipment and storage medium | |
| CN111242318A (en) | Business model training method and device based on heterogeneous feature library | |
| CN114202243A (en) | Engineering project management risk early warning method and system based on random forest | |
| CN113177700A (en) | Risk assessment method, system, electronic equipment and storage medium | |
| CN115330435A (en) | Method, device, equipment and medium for establishing carbon emission right price index system | |
| CN114968727A (en) | Database through infrastructure fault positioning method based on artificial intelligence operation and maintenance | |
| CN112950347B (en) | Resource data processing optimization method and device, storage medium and terminal | |
| CN113570222A (en) | User equipment identification method and device and computer equipment | |
| CN108647714A (en) | Acquisition methods, terminal device and the medium of negative label weight | |
| CN112949778A (en) | Intelligent contract classification method and system based on locality sensitive hashing and electronic equipment | |
| CN111523604A (en) | User classification method and related device | |
| CN111143533A (en) | Customer service method and system based on user behavior data | |
| CN112732549B (en) | Test program classification method based on cluster analysis | |
| CN116502140B (en) | Encryption algorithm identification method and device based on similarity of control flow graphs | |
| CN116342300B (en) | Method, device and equipment for analyzing characteristics of insurance claim settlement personnel |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |