CN113962799A

CN113962799A - Training method of wind control model, risk determination method, device and equipment

Info

Publication number: CN113962799A
Application number: CN202111235762.6A
Authority: CN
Inventors: 陈奇石; 刘昊骋; 许海洋
Original assignee: Beijing Baidu Netcom Science and Technology Co Ltd
Current assignee: Beijing Baidu Netcom Science and Technology Co Ltd
Priority date: 2021-10-22
Filing date: 2021-10-22
Publication date: 2022-01-21

Abstract

The disclosure provides a training method, a risk determination device, equipment and a storage medium of a wind control model, and relates to the technical field of computers, in particular to the fields of artificial intelligence, deep learning and the like. The specific implementation scheme is as follows: determining a search sample with a specified intention in search histories of a plurality of users; determining a target user among the plurality of users according to the search sample; obtaining a risk level prediction result of the target user by utilizing the first data model; and updating and training the first data model by using the accuracy of the determined risk level prediction result of the target user to obtain the wind control model. The training effect of the model can be improved.

Description

Training method of wind control model, risk determination method, device and equipment

Technical Field

The present disclosure relates to the field of computer technologies, and in particular, to the fields of artificial intelligence, deep learning, and the like, and in particular, to a training method, a risk determination method, an apparatus, a device, and a storage medium for a wind control model.

Background

In recent years, the field of application of artificial intelligence technology has become more and more widespread. Such as the financial field, etc. One branch of artificial intelligence is a neural network, a positive sample and a negative sample are used in the training process of the neural network, and the problem of difficulty in obtaining the samples exists in the special field. For example, the number of people who lose credit is small, the process of confirming a sample of losing credit is complicated, and the like, and sample mining is required for the above cases.

The traditional manual sample digging mode has poor efficiency, and the accuracy cannot be guaranteed.

Disclosure of Invention

The disclosure provides a training method, a risk determination device, equipment and a storage medium of a wind control model.

According to an aspect of the present disclosure, there is provided a training method of a wind control model, which may include the steps of:

determining a search sample with a specified intention in search histories of a plurality of users;

determining a target user among the plurality of users according to the search sample;

obtaining a risk level prediction result of the target user by utilizing the first data model;

and updating and training the first data model by using the determined risk level prediction result of the target user to obtain the wind control model.

According to another aspect of the present disclosure, there is provided a risk determination method, which may include the steps of:

acquiring related information of a target object;

inputting relevant information of the target object into a pre-trained wind control model to obtain a risk prediction result of the target object;

wherein, the wind control model is obtained by the training of the method.

According to another aspect of the present disclosure, there is provided a training apparatus of a wind control model, which may include:

the search sample determining module is used for determining search samples with specified intentions in the search histories of a plurality of users;

the target user determining module is used for determining a target user in the plurality of users according to the search sample;

the risk level prediction result determining module is used for obtaining a risk level prediction result of the target user by utilizing the first data model;

and the model training module is used for updating and training the first data model by using the accuracy of the determined risk level prediction result of the target user to obtain the wind control model.

According to another aspect of the present disclosure, there is provided a risk determination apparatus, which may include:

acquiring related information of a target object;

wherein, the wind control model is obtained through the training of the device.

According to another aspect of the present disclosure, there is provided an electronic device including:

at least one processor; and

a memory communicatively coupled to the at least one processor; wherein the content of the first and second substances,

the memory stores instructions executable by the at least one processor to enable the at least one processor to perform a method according to any one of the embodiments of the present disclosure.

According to another aspect of the present disclosure, there is provided a non-transitory computer readable storage medium having stored thereon computer instructions for causing a computer to perform a method in any of the embodiments of the present disclosure.

According to another aspect of the present disclosure, there is provided a computer program product comprising computer programs/instructions which, when executed by a processor, implement the method in any of the embodiments of the present disclosure.

The techniques according to the present disclosure may employ an automatic search for sample mining. In the case of mining a sample, automatic labeling of a target user can be realized. Model training is performed on the model by using the labeling result, so that the number of samples can be enriched to solve the problem of few samples. And moreover, model training is carried out on the model by using the labeling result, so that the training effect of the model can be improved.

It should be understood that the statements in this section do not necessarily identify key or critical features of the embodiments of the present disclosure, nor do they limit the scope of the present disclosure. Other features of the present disclosure will become apparent from the following description.

Drawings

The drawings are included to provide a better understanding of the present solution and are not to be construed as limiting the present disclosure. Wherein:

FIG. 1 is a flow chart of a training method of a wind control model according to the present disclosure;

FIG. 2 is a flow diagram of a search sample determining the presence of a specified intent in accordance with the present disclosure;

FIG. 3 is a flow diagram of filtering candidate results according to predetermined rules in accordance with the present disclosure;

FIG. 4 is a flow chart of determining a target user according to the present disclosure;

FIG. 5 is a flow chart of an update training of a first data model, resulting in a wind control model according to the present disclosure;

FIG. 6 is a flow chart of a risk determination method according to the present disclosure;

FIG. 7 is a schematic view of a training apparatus for a wind control model according to the present disclosure;

FIG. 8 is a schematic view of a risk determination device according to the present disclosure;

FIG. 9 is a block diagram of an electronic device for implementing a training method of a wind control model according to an embodiment of the present disclosure.

Detailed Description

Exemplary embodiments of the present disclosure are described below with reference to the accompanying drawings, in which various details of the embodiments of the disclosure are included to assist understanding, and which are to be considered as merely exemplary. Accordingly, those of ordinary skill in the art will recognize that various changes and modifications of the embodiments described herein can be made without departing from the scope and spirit of the present disclosure. Also, descriptions of well-known functions and constructions are omitted in the following description for clarity and conciseness.

As shown in fig. 1, the present disclosure provides a training method of a wind control model, which may include the following steps:

s101: determining a search sample with a specified intention in search histories of a plurality of users;

s102: determining a target user among the plurality of users according to the search sample;

s103: obtaining a risk level prediction result of the target user by utilizing the first data model;

s104: and updating and training the first data model by using the determined risk level prediction result of the target user to obtain the wind control model.

The scheme can be applied to the scene of model training, and can be adopted for the conditions of complex sample obtaining process and the like. In particular, the method can be applied to wind control models in the fields of financial credit, intelligent transportation and the like.

The search history may be selected from the search data of the last month or the last half year, and the specific time is not limited herein. The acquisition of the search history may be acquired with the user's authorization. The specified intent may differ for different domains. Taking the financial credit field as an example, in the financial credit field, the intention of losing credit may be taken as the designated intention. The search history of multiple users may include "how to query this monthly bill", "how to modify repayment date", "how to do not repayment due after term", and the like. Wherein a search sample for which a specified intent exists can be determined from the search histories for a plurality of users. As explained in the foregoing example, "how to query this month's bill", "how to modify the repayment date" may all be determined as a normal query, i.e., there is no specified intent. The term "how to do overdue yet" indicates that the user has an intention of overdue yet not to pay, and based on this, it can be determined that the designated intention exists.

Taking the intelligent transportation field as an example, the search histories of multiple users can include "query the nearest route", "query the violation probe in the navigation route", "do the traffic light in front of" and the like. With respect to the above search history, "it is determined that there is a prescribed intention that this traffic light ahead has been photographed". That is, in the field of intelligent transportation, a dangerous driving intention or an illegal driving intention may be taken as a designated intention.

In the case where the search sample is determined, the target user may be determined among the plurality of users according to a relationship between the search history of the user and the search sample. For example, if the first user does not include a search sample in the search history, the first user may be excluded. The second user may be determined to be the target user if the second user searches multiple times (exceeding a threshold number of times) in the search history of the second user using the search sample. As can be seen from the above examples, the target user may be considered as a user at risk, such as a user with a credit crisis or a potentially dangerous driver. That is, the target user may be a positive sample to achieve calibration.

Among the plurality of users, a part of the users may be randomly selected. Randomly picked parts of the users may be used as negative examples. The aforementioned target user may be taken as a positive sample. Illustratively, the number of randomly picked partial users may be the same as the number of target users; alternatively, the difference between the two numbers is within the margin threshold.

The relevant information of the positive sample and the relevant information of the negative sample are input into the first data model, and the risk level prediction result of each sample can be obtained. The information related to the positive sample and the information related to the negative sample may include information such as an identifier of the user and a search history of the user. The risk level prediction result can be displayed in a probability form, and can also be displayed in a high, medium and low grade. For example, users with a probability of 0 to 33% may correspond to a low level, users with a probability of 34 to 66% may correspond to a medium level, users with a probability of 67 to 100% may correspond to a high level, and so on.

The first data model may be a neural network model to be trained, or may be a neural network model that has been trained but has not yet achieved a predetermined effect. In the case where the first data model results in a risk level prediction, the risk level prediction for the target user may be used as a new input feature.

And inputting the new input features and the original input features into the first data model, wherein the first data model can obtain a new risk level prediction result of the target user. And adjusting parameters in the first data model by using the error between the labeled risk level prediction result (the true value of the recognition result) and the prediction value of the recognition result until the error is within an allowable range.

By the method, sample mining can be performed in an automatic searching mode. In the case of mining a sample, automatic labeling of a target user can be realized. Model training is performed on the model by using the labeling result, so that the number of samples can be enriched to solve the problem of few samples. And moreover, model training is carried out on the model by using the labeling result, so that the training effect of the model can be improved.

As shown in fig. 2, in one embodiment, step S101 may include the following steps:

s201: acquiring historical behaviors of a plurality of users;

s202: screening the search history according to the history behavior to obtain a candidate result;

s203: and filtering the candidate results according to a preset rule, and taking the filtered results as a search sample with a specified intention.

Still taking the financial credit field as an example, the historical behavior may include the user's inquiry of a monthly bill operation, modification of a repayment date operation, a past due unpaid operation, and the like.

The search history may be filtered based on the user's historical behavior. For any Query information (Query) in the search history, calculating the ratio of the number of positive samples for searching the Query information to the total number of samples for searching the Query information, and in the case that the ratio is greater than a ratio threshold, retaining the Query information. Where a positive sample may correspond to the presence of historical behavior. For example, if there are 100 persons who have made "overdue unpaid money" in the historical search, and 91 persons who have made overdue unpaid money, the ratio of the number of positive samples for searching the query information to the total number of samples for searching the query information is represented as 91/100.

For the query information retained after screening, a predetermined rule may be used for further screening. The predetermined rule may include at least one of using a manual filter, using a keyword auto filter, and using a neural network model auto filter.

Taking the automatic screening using keywords as an example, the keyword set may be predetermined. If the query information includes the keywords in the keyword set, the query information may be retained. Otherwise, the query information is deleted. And taking the result remained after filtering as a search sample with the specified intention. In addition, the manner of screening using the neural network model will be described in detail later.

Through the process, automatic mining of the search samples can be achieved.

As shown in fig. 3, in one embodiment, filtering the candidate results according to a predetermined rule includes:

s301: inputting the candidate result into a pre-trained intention recognition model to obtain an intention prediction result of each search of the user;

s302: and determining the search corresponding to the intention prediction result as a search sample under the condition that the intention prediction result represents the designated intention.

The intention recognition model may be a model based on Natural Language Processing (NLP). For example, intent recognition of queries such as "query monthly bill," "modify repayment date," "how outstanding," etc. in the foregoing example may result in an intent to check the bill, modify repayment date, and potentially being overdue.

Correspondingly, a confidence losing intention set can be established in advance, and when the intention prediction result belongs to the confidence losing intention, the search corresponding to the intention prediction result is determined as the search sample.

Through the process, the sample can be determined automatically and efficiently.

As shown in fig. 4, in one embodiment, step S102 may include the following steps:

s401: determining users with search samples in the search history as candidate users;

s402: and determining the target user from the candidate users by using the number of search samples appearing in the search history of the candidate users.

Specifically, step S402 may include the following steps: counting search samples appearing in the search history of the candidate user and the occurrence frequency of each appearing search sample;

and determining the candidate user as the target user under the condition that the number of the search samples appearing in the search history exceeds a number threshold value or the number of the appearing search samples exceeds a number threshold value.

In the case that the search sample is determined, a query may be made in the search history based on the search sample to determine a user corresponding to the search sample. That is, a user who searches with a search sample may be determined as a candidate user.

The search history of the candidate user is traversed and the candidate user may be determined as the target user in case the same search sample occurs multiple times (more than a first predetermined number, e.g. 3 times) in the search history of the candidate user. For example, the same search sample may be "how to make outstanding payments". That is, the user searches for "how to make a overdue unpaid money" a plurality of times. Alternatively, the candidate user may be determined as the target user in case that a different search sample occurs multiple times (more than a second predetermined number, e.g. 2 times) in the search history of the candidate user. For example, the different search samples may be search samples that express different content such as "how to do not pay for overdue", "consequence of losing confidence", and the like. The search sample may be a result of normalizing the search content, for example, the search sample may be normalized to a standard form after normalizing the search content with the same meaning but different characters.

By the scheme, the target user can be determined. And the target user can be used as a positive sample for training the wind control model, namely, the target user can be used as a user with possible wrong behaviors. The model is trained by the target user, so that the number of samples can be enriched to solve the problem of less samples.

As shown in fig. 5, in one embodiment, step S103 may include the steps of:

s501: inputting input characteristics including a risk level prediction result of a target user into a first data model to obtain a new risk level prediction result of the target user;

s502: updating and training the first data model by using the difference between the new risk level prediction result and the labeling result of the target user;

s503: and obtaining a wind control model under the condition that the difference is within the threshold value range.

And the risk level prediction result of the target user is used as a new input characteristic, and the new input characteristic and the original input characteristic are input into the first data model together. The first data model can obtain a new risk level prediction result of the target user according to the new input features and the original input features.

In one embodiment, the input features may include at least one of an identification of the target user, risk level prediction results of the target user, and a search history of the target user.

The identification of the target user may be the name, the registered account number, and the like of the target user. The search history of the target user may include search content over a past period of time, or may further utilize the aforementioned search samples to filter the search history of retained content.

The parameters in the first data model are adjusted by using the error between the labeled risk level prediction result (true value of the recognition result) and the prediction value of the recognition result. The above error can be embodied by a loss function, and the effect of the loss function can be understood as: when a predicted value obtained by forward propagation of the first data model is close to the true value, the smaller value of the loss function is selected; conversely, the value of the loss function increases. The loss function is a function having a parameter in the first data model as an argument.

The labeled risk level prediction result may be a result of manual labeling, or may be a labeling result determined by calculating the target user according to a predetermined rule.

The predetermined rules may include: in the case where the target user is determined, the risk level prediction result of the target user may be set to 85% (an illustrative value). And adjusting the set risk level prediction result of the target user according to at least one of the search history content of the target user and the history behavior of the target user. For example, if the number of times of occurrence of a search sample in search history content exceeds a threshold number of times, the risk level prediction result is correspondingly increased; and if the confidence losing behavior occurs in the historical behaviors of the target user, the risk level prediction result is correspondingly increased. It will be appreciated that the adjusted value of the risk level prediction result may be proportional to the number of occurrences of the search pattern in the search history, and the number of occurrences of the loss of confidence behavior.

Through the process, the model is trained by using the labeling result, and the training effect of the model can be improved.

As shown in fig. 6, the present disclosure relates to a risk determination method, which may include the steps of:

s601: acquiring related information of a target object;

s602: inputting relevant information of the target object into a pre-trained wind control model to obtain a risk prediction result of the target object;

the wind control model is obtained through training by the method.

In one embodiment, the related information of the target object includes a search history.

As shown in fig. 7, the present disclosure relates to a training apparatus of a wind control model, which may include:

a search sample determination module 701, configured to determine search samples with specified intentions in search histories of multiple users;

a target user determination module 702, configured to determine a target user among the multiple users according to the search sample;

a risk level prediction result determining module 703, configured to obtain a risk level prediction result of the target user by using the first data model;

and the model training module 704 is configured to perform update training on the first data model by using the accuracy of the determined risk level prediction result of the target user to obtain a wind control model.

In one embodiment, the search sample determination module 701 may include:

the historical behavior acquisition submodule is used for acquiring historical behaviors of a plurality of users;

the candidate result determining submodule is used for screening the search history according to the history behavior to obtain a candidate result;

and the search sample determination execution submodule is used for filtering the candidate results according to a preset rule and taking the filtered results as the search samples with the confidence losing intention.

In one embodiment, the search sample determination performing sub-module may include:

the intention prediction result determining unit is used for inputting the candidate result into a pre-trained intention recognition model to obtain the intention prediction result of each search of the user;

and a search sample determining unit, configured to determine, as a search sample, a search corresponding to the intent prediction result if the intent prediction result belongs to the specified intent.

In one embodiment, the target user determination module 702 may include:

the candidate user determining submodule is used for determining the user with the search sample in the search history as the candidate user;

and the target user determination execution sub-module is used for determining the target user from the candidate users by using the number of the search samples appearing in the search history of the candidate users.

In one embodiment, the target user determination execution sub-module may include:

the statistical unit is used for counting the search samples appearing in the search history of the candidate user and the occurrence frequency of each appearing search sample;

and the execution unit is used for determining the candidate user as the target user under the condition that the number of the search samples appearing in the search history exceeds a number threshold value or the number of the appearing search samples exceeds a number threshold value.

In one embodiment, model training module 704 may include:

the new risk level prediction result determining submodule is used for inputting input characteristics containing the risk level prediction result of the target user into the first data model to obtain a new risk level prediction result of the target user;

the training execution submodule is used for updating and training the first data model by utilizing the difference between the new risk level prediction result and the labeling result of the target user;

and the wind control model generation submodule is used for obtaining the wind control model under the condition that the difference is within the threshold range.

In one embodiment, the input features further comprise: at least one of an identification of the target user and a search history of the target user.

As shown in fig. 8, the present disclosure relates to a risk determination device, which may include:

an information obtaining module 801, configured to obtain relevant information of a target object;

the risk determination module 802 is configured to input relevant information of the target object to a pre-trained wind control model to obtain a risk prediction result of the target object;

wherein, the wind control model is obtained by the training of the device.

In the technical scheme of the disclosure, the acquisition, storage, application and the like of the personal information of the related user all accord with the regulations of related laws and regulations, and do not violate the good customs of the public order.

The present disclosure also provides an electronic device, a readable storage medium, and a computer program product according to embodiments of the present disclosure.

FIG. 9 illustrates a schematic block diagram of an example electronic device 900 that can be used to implement embodiments of the present disclosure. Electronic devices are intended to represent various forms of digital computers, such as laptops, desktops, workstations, personal digital assistants, servers, blade servers, mainframes, and other appropriate computers. The electronic device may also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions, are meant to be examples only, and are not meant to limit implementations of the disclosure described and/or claimed herein.

As shown in fig. 9, the device 900 includes a computing unit 910 that may perform various appropriate actions and processes in accordance with a computer program stored in a Read Only Memory (ROM)920 or a computer program loaded from a storage unit 980 into a Random Access Memory (RAM) 930. In the RAM930, various programs and data required for the operation of the device 900 may also be stored. The calculation unit 910, the ROM 920, and the RAM930 are connected to each other via a bus 940. An input/output (I/O) interface 950 is also connected to bus 940.

Various components in device 900 are connected to I/O interface 950, including: an input unit 960 such as a keyboard, a mouse, etc.; an output unit 970 such as various types of displays, speakers, and the like; a storage unit 980 such as a magnetic disk, optical disk, or the like; and a communication unit 990 such as a network card, a modem, a wireless communication transceiver, or the like. The communication unit 990 allows the device 900 to exchange information/data with other devices via a computer network such as the internet and/or various telecommunication networks.

Computing unit 910 may be a variety of general-purpose and/or special-purpose processing components having processing and computing capabilities. Some examples of the computing unit 910 include, but are not limited to, a Central Processing Unit (CPU), a Graphics Processing Unit (GPU), various specialized Artificial Intelligence (AI) computing chips, various computing units running machine learning model algorithms, a Digital Signal Processor (DSP), and any suitable processor, controller, microcontroller, and so forth. The calculation unit 910 performs various methods and processes described above, such as a training method of a wind control model. For example, in some embodiments, the training method of the wind control model may be implemented as a computer software program tangibly embodied in a machine-readable medium, such as the storage unit 980. In some embodiments, some or all of the computer program may be loaded and/or installed onto device 900 via ROM 920 and/or communication unit 990. When the computer program is loaded into RAM930 and executed by computing unit 910, one or more steps of the training method of the wind control model described above may be performed. Alternatively, in other embodiments, the computing unit 910 may be configured by any other suitable means (e.g., by means of firmware) to perform the training method of the wind control model.

Various implementations of the systems and techniques described here above may be implemented in digital electronic circuitry, integrated circuitry, Field Programmable Gate Arrays (FPGAs), Application Specific Integrated Circuits (ASICs), Application Specific Standard Products (ASSPs), system on a chip (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and/or combinations thereof. These various embodiments may include: implemented in one or more computer programs that are executable and/or interpretable on a programmable system including at least one programmable processor, which may be special or general purpose, receiving data and instructions from, and transmitting data and instructions to, a storage system, at least one input device, and at least one output device.

Program code for implementing the methods of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general purpose computer, special purpose computer, or other programmable data processing apparatus, such that the program codes, when executed by the processor or controller, cause the functions/operations specified in the flowchart and/or block diagram to be performed. The program code may execute entirely on the machine, partly on the machine, as a stand-alone software package partly on the machine and partly on a remote machine or entirely on the remote machine or server.

In the context of this disclosure, a machine-readable medium may be a tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

To provide for interaction with a user, the systems and techniques described here can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to a user; and a keyboard and a pointing device (e.g., a mouse or a trackball) by which a user can provide input to the computer. Other kinds of devices may also be used to provide for interaction with a user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user may be received in any form, including acoustic, speech, or tactile input.

The systems and techniques described here can be implemented in a computing system that includes a back-end component (e.g., as a data server), or that includes a middleware component (e.g., an application server), or that includes a front-end component (e.g., a user computer having a graphical user interface or a web browser through which a user can interact with an implementation of the systems and techniques described here), or any combination of such back-end, middleware, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: local Area Networks (LANs), Wide Area Networks (WANs), and the Internet.

The computer system may include clients and servers. A client and server are generally remote from each other and typically interact through a communication network. The relationship of client and server arises by virtue of computer programs running on the respective computers and having a client-server relationship to each other. The server may be a cloud server, a server of a distributed system, or a server with a combined blockchain.

It should be understood that various forms of the flows shown above may be used, with steps reordered, added, or deleted. For example, the steps described in the present disclosure may be executed in parallel, sequentially, or in different orders, as long as the desired results of the technical solutions disclosed in the present disclosure can be achieved, and the present disclosure is not limited herein.

The above detailed description should not be construed as limiting the scope of the disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may be made in accordance with design requirements and other factors. Any modification, equivalent replacement, and improvement made within the spirit and principle of the present disclosure should be included in the scope of protection of the present disclosure.

Claims

1. A training method of a wind control model comprises the following steps:

obtaining a risk level prediction result of the target user by utilizing a first data model;

and updating and training the first data model by using the determined risk level prediction result of the target user to obtain a wind control model.

2. The method of claim 1, wherein the determining that there is a search sample of specified intentions in the search histories of the plurality of users comprises:

acquiring historical behaviors of the plurality of users;

screening the search history according to the historical behaviors to obtain a candidate result;

and filtering the candidate results according to a preset rule, and taking the filtered results as the search samples with the specified intentions.

3. The method of claim 2, the filtering the candidate results according to a predetermined rule, comprising:

inputting the candidate result into a pre-trained intention recognition model to obtain an intention prediction result of each search of the user;

determining a search corresponding to the intent prediction result as the search sample if the intent prediction result characterizes the specified intent.

4. The method of claim 1, wherein the determining a target user among the plurality of users from the search sample comprises:

determining users with the search samples in the search history as candidate users;

determining the target user among the candidate users using the number of search samples that appear in the search history of the candidate users.

5. The method of claim 4, wherein the determining the target user among the candidate users using the number of search samples that appear in the search history of the candidate users comprises:

counting search samples appearing in the search history of the candidate user and the number of occurrences of each of the appearing search samples;

and determining the candidate user as the target user when the number of the search samples appearing in the search history exceeds a number threshold or the number of the appearing search samples exceeds a number threshold.

6. The method of claim 1, wherein the training of the first data model with the determined risk level prediction of the target user comprises:

inputting input characteristics including the risk level prediction result of the target user into a first data model to obtain a new risk level prediction result of the target user;

updating and training the first data model by using the difference between the new risk level prediction result and the labeling result of the target user;

and obtaining the wind control model under the condition that the difference is within a threshold value range.

7. The method of claim 6, wherein the input features further comprise: at least one of an identification of the target user and a search history of the target user.

8. A method of risk determination, comprising:

acquiring related information of a target object;

inputting the relevant information of the target object into a pre-trained wind control model to obtain a risk prediction result of the target object;

wherein the wind control model is trained by the method of any one of claims 1 to 7.

9. The method of claim 8, wherein the information related to the target object comprises a search history.

10. A training apparatus for a wind control model, comprising:

a target user determination module, configured to determine a target user among the multiple users according to the search sample;

the risk level prediction result determining module is used for obtaining a risk level prediction result of the target user by utilizing a first data model;

and the model training module is used for updating and training the first data model by using the determined risk level prediction result of the target user to obtain a wind control model.

11. The apparatus of claim 10, wherein the search sample determination module comprises:

the historical behavior acquisition submodule is used for acquiring the historical behaviors of the users;

a candidate result determining submodule, configured to screen the search history according to the history behavior to obtain a candidate result;

and the search sample determination execution sub-module is used for filtering the candidate results according to a preset rule and taking the filtered results as the search samples with the specified intentions.

12. The apparatus of claim 11, wherein the search sample determination execution submodule comprises:

the intention prediction result determining unit is used for inputting the candidate result into a pre-trained intention recognition model to obtain an intention prediction result of each search of the user;

and a search sample determining unit, configured to determine, as the search sample, a search corresponding to the intent prediction result if the intent prediction result characterizes the designation.

13. The apparatus of claim 10, wherein the target user determination module comprises:

a candidate user determining submodule for determining a user having the search sample in the search history as a candidate user;

and the target user determination execution sub-module is used for determining the target user in the candidate users by utilizing the number of the search samples appearing in the search history of the candidate users.

14. The apparatus of claim 13, wherein the target user determination execution submodule comprises:

a counting unit, configured to count search samples appearing in the search history of the candidate user and the number of occurrences of each of the appearing search samples;

15. The apparatus of claim 10, wherein the model training module comprises:

a new risk level prediction result determination submodule, configured to input an input feature including a risk level prediction result of the target user to a first data model, to obtain a new risk level prediction result of the target user;

the training execution sub-module is used for updating and training the first data model by utilizing the difference between the new risk level prediction result and the labeling result of the target user;

and the wind control model generation submodule is used for obtaining the wind control model under the condition that the difference is within a threshold range.

16. The apparatus of claim 15, wherein the input features further comprise: at least one of an identification of the target user and a search history of the target user.

17. A risk determination device, comprising:

the information acquisition module is used for acquiring related information of the target object;

the risk determination module is used for inputting the relevant information of the target object into a pre-trained wind control model to obtain a risk prediction result of the target object;

wherein the wind control model is trained by the device of any one of claims 10 to 16.

18. The apparatus of claim 17, wherein the information related to the target object comprises a search history.

19. An electronic device, comprising:

at least one processor; and

the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method of any one of claims 1 to 9.

20. A non-transitory computer readable storage medium having stored thereon computer instructions for causing the computer to perform the method of any one of claims 1 to 9.

21. A computer program product comprising computer programs/instructions, wherein the computer programs/instructions, when executed by a processor, implement the steps of the method of any one of claims 1 to 9.