Object state recognition method, apparatus and device

Abstract

This application provides a method, apparatus, and device for object state recognition, relating to the field of artificial intelligence technology, and particularly to object state recognition in mobile payment, to improve the accuracy of object state recognition. The method includes: acquiring various feedback data of the object to be identified in the target business; identifying whether each feedback data contains target information; determining the recognition result corresponding to each feedback data; and based on the recognition results corresponding to each feedback data, determining whether the object state of the object to be identified in the target business is abnormal. The feedback data is triggered by business operations performed on the object to be identified in the target business using the target business's account, and the target information represents that the business operation does not meet the operation agreement information associated with the target business. This method can improve the accuracy of identifying whether the object state in the target business is abnormal.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a method, apparatus and device for object state recognition. Background Technology

[0002] With the development of internet technology, mobile payment has become an integral part of daily life. Currently, in the process of identifying objects with abnormal states in mobile payments, the judgment is often made based on user complaints about these objects in the target business. However, user complaints are highly subjective, and the veracity of the information in the complaints is unreliable. Directly relying on user complaint data to identify whether an object's state is abnormal has low accuracy. Therefore, how to improve the accuracy of identifying object states is a problem worth considering. Summary of the Invention

[0003] This application provides an object state recognition method, apparatus, and device to improve the accuracy of identifying whether an object's state is abnormal.

[0004] In a first aspect, this application provides an object state recognition method, comprising:

[0005] Obtain various feedback data of the object to be identified in the target service; wherein, the feedback data is triggered by the business operations performed by the object to be identified in the target service using the account of the target service;

[0006] Each feedback data is identified as containing target information, and the identification result corresponding to each feedback data is determined. The target information represents the operation that does not meet the operation convention information associated with the target business.

[0007] Based on the identification results corresponding to each of the feedback data, it is determined whether the object status of the object to be identified in the target service is abnormal.

[0008] In one possible implementation, the object attribute information includes at least one of the following:

[0009] Transfer information of electronic resources of the object to be identified;

[0010] Fraud anomaly information corresponding to the object to be identified, wherein the fraud anomaly information indicates the degree of suspicion that the object to be identified may be involved in online fraudulent activities;

[0011] Operation information associated with abnormal business operations, wherein the abnormal business operations include business operations associated with feedback data containing the target information.

[0012] In one possible implementation, the object attribute information includes operation information associated with the abnormal business operation, and the operation information includes at least one of the following:

[0013] The electronic resource value associated with the abnormal business operation;

[0014] The account associated with the abnormal business operation;

[0015] The business object associated with the abnormal business operation;

[0016] The triggering time information of the abnormal business operation.

[0017] A second aspect of this application provides an object state recognition device, comprising:

[0018] The data acquisition unit is used to acquire various feedback data of the object to be identified in the target business; wherein, the feedback data is triggered by the business operation performed by the object to be identified in the target business using the account of the target business;

[0019] The first identification unit is used to identify whether each feedback data contains target information, and to determine the identification result corresponding to each feedback data. The target information represents that the business operation does not meet the operation agreement information associated with the target business.

[0020] The second identification unit is used to determine whether the object status of the object to be identified in the target service is abnormal based on the identification results corresponding to each of the feedback data.

[0021] In one possible implementation, the first identification unit is specifically configured to perform the following operations for each of the feedback data:

[0022] Based on the contextual information of each word contained in one of the feedback data, extract the text features to be processed corresponding to the feedback data; based on the text features to be processed, identify whether the feedback data contains the target information, and determine the identification result corresponding to the feedback data.

[0023] In one possible implementation, the first identification unit is specifically used for:

[0024] Input the feedback data into the trained data classification model;

[0025] Based on the language learning sub-model in the data classification model, feature extraction is performed on the contextual information of each word contained in the feedback data to obtain the text features to be processed corresponding to the feedback data; wherein, the language learning sub-model is trained by using historical feedback data in the target business as training samples and based on the contextual information of each word contained in the training samples.

[0026] The second identification unit is specifically used to: predict the second correlation between the text features to be processed and the first identification information based on the first correlation learned by the data prediction sub-model in the data classification model; determine the identification result corresponding to the feedback data based on the second correlation; the first identification information represents that the feedback data contains the target information; and the first correlation is determined based on the degree of correlation between the historical text features corresponding to the historical feedback data and the first identification information.

[0027] In one possible implementation, the first identification unit is specifically used to: if the second correlation degree is greater than the first preset threshold, determine that the identification result corresponding to the feedback data is that the feedback data contains the target information; if the second correlation degree is not greater than the first preset threshold, determine that the identification result corresponding to the feedback data is that the feedback data does not contain the target information.

[0028] In one possible implementation, the second identification unit is specifically used for:

[0029] Based on the first identification information in the identification results corresponding to each of the feedback data, the valid feedback information corresponding to the object to be identified is determined, and the first identification information is used to characterize that the feedback data contains the target information.

[0030] Obtain the object attribute information associated with the object to be identified;

[0031] Based on the valid feedback information and the object attribute information, it is determined whether the object status of the object to be identified in the target service is abnormal.

[0032] In one possible implementation, the valid feedback information includes one or any combination of the following: the number of first identification information in the identification results corresponding to each of the feedback data; and the ratio of the number of first identification information to the total number of the feedback data in the identification results corresponding to each of the feedback data.

[0033] In one possible implementation, the second identification unit is specifically used to: process the feedback valid information and the object attribute information based on the target scorecard model to obtain the target operation evaluation value of the object to be identified for the target service; and determine whether the object status of the object to be identified in the target service is abnormal based on the target operation evaluation value.

[0034] In one possible implementation, the second identification unit is further configured to: determine the object type of the object to be identified before processing the feedback valid information and the object attribute information based on the target scorecard model to obtain the target operation evaluation value of the object to be identified for the target business; and obtain the target scorecard model corresponding to the object type from the target scorecard model set associated with the target business.

[0035] In one possible implementation, the second identification unit is specifically used for:

[0036] If the target operation evaluation value is determined to be lower than the evaluation value threshold, then the object to be identified is determined to be an abnormal object; wherein, the abnormal object includes objects whose object state is abnormal in the target service, and the operation evaluation value is negatively correlated with the degree of suspicion that the object's object state is abnormal in the target service; or

[0037] Based on the operation evaluation value range mapped to at least one object anomaly level, the operation evaluation value range to which the target operation evaluation value belongs is determined, and the object anomaly level mapped to the determined operation evaluation value range is determined as the object anomaly level corresponding to the object to be identified; wherein, the object anomaly level is determined based on the degree of suspicion that the object's object status in the target business is an abnormal state.

[0038] In one possible implementation, the second recognition unit is specifically used to: input the feedback valid information and the object attribute information into a trained state recognition model, and perform the following processing through the state recognition model:

[0039] The valid feedback information and the object attribute information are determined as the features of the object to be processed corresponding to the object to be identified;

[0040] Based on the third correlation degree learned by the state recognition model, a fourth correlation degree is predicted between the features of the object to be processed and the target object recognition result. If the fourth correlation degree is greater than a second set threshold, the object to be identified is determined to be an abnormal object. The third correlation degree is determined based on the degree of correlation between the historical object features corresponding to the historical object and the target recognition result. The abnormal object includes objects whose object state is abnormal in the target business. The target object recognition result is used to characterize that the object's object state is abnormal in the target business.

[0041] In one possible implementation, the second identification unit is further configured to:

[0042] After determining that the object to be identified is an abnormal object, at least one of the first permission and the second permission of the object to be identified is cancelled; the first permission includes the permission to perform electronic resource transfer operations, and the second permission includes electronic resource transfer operations for a target electronic resource value, the target electronic resource value being determined based on the feedback data.

[0043] In one possible implementation, the object attribute information includes at least one of the following:

[0044] Transfer information of electronic resources of the object to be identified;

[0045] Fraud anomaly information corresponding to the object to be identified, wherein the fraud anomaly information indicates the degree of suspicion that the object to be identified may be involved in online fraudulent activities;

[0046] Operation information associated with abnormal business operations, wherein the abnormal business operations include business operations associated with feedback data containing the target information.

[0047] In one possible implementation, the object attribute information includes operation information associated with the abnormal business operation, and the operation information includes at least one of the following: the electronic resource value associated with the abnormal business operation; the account associated with the abnormal business operation; the business object associated with the abnormal business operation; and the trigger time information of the abnormal business operation.

[0048] A third aspect of this application provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method described in the first aspect.

[0049] In a fourth aspect, this application provides a computer program product including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the method described in the first aspect above.

[0050] In a fifth aspect, this application provides a computer-readable storage medium storing computer instructions that, when executed on a computer, cause the computer to perform the method described in the first aspect.

[0051] Since the embodiments of this application adopt the above technical solution, they have at least the following technical effects:

[0052] In this embodiment, the feedback data of the object in the target business is first identified to determine whether the business operation corresponding to each feedback data meets the operation agreement information associated with the target business. Then, based on the identification results corresponding to each feedback data, it is determined whether the object status of the object to be identified in the target business is abnormal, rather than simply identifying whether the object status of the object in the target business is abnormal based on the content of the user's complaint, thereby improving the accuracy of identifying whether the object status of the object in the target business is abnormal. Attached Figure Description

[0053] Figure 1 A schematic diagram illustrating an application scenario for object state recognition provided in an embodiment of this application;

[0054] Figure 2 This application provides a schematic diagram of a process for obtaining feedback data in an embodiment of the present application.

[0055] Figure 3 A flowchart illustrating an object state recognition method provided in this application embodiment;

[0056] Figure 4 This is a schematic diagram of the structure of a data classification model provided in an embodiment of this application;

[0057] Figure 5 A schematic diagram illustrating the training process of a language learning sub-model provided in an embodiment of this application;

[0058] Figure 6 A flowchart illustrating the training process of a data classification model provided in an embodiment of this application;

[0059] Figure 7 A flowchart for creating a target scorecard model is provided as an embodiment of this application;

[0060] Figure 8A flowchart for identifying merchants whose object status is abnormal in a cashback business, provided as an embodiment of this application;

[0061] Figure 9 A flowchart illustrating the training of a data classification model provided in an embodiment of this application;

[0062] Figure 10 A system flowchart for object state recognition provided in this application embodiment;

[0063] Figure 11 This is a schematic diagram of the structure of an object state recognition device provided in an embodiment of this application;

[0064] Figure 12 This is a structural diagram of a computer device provided in an embodiment of this application. Detailed Implementation

[0065] To make the objectives, technical solutions, and beneficial effects of this application clearer, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this application, and not all of the embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0066] To facilitate a better understanding of the technical solutions of this application by those skilled in the art, some of the concepts involved in this application are explained below.

[0067] 1) Object

[0068] In this application embodiment, the object may be, but is not limited to, a target capable of making mobile payments. The object may be a user's identity identifier on the Internet, such as a merchant account or individual account on the Internet in this application embodiment.

[0069] 2) Target business, business operations, and feedback data

[0070] Business generally refers to the transactions that need to be processed in various industries. The target business in this application embodiment may include, but is not limited to, the transactions that need to be processed in mobile payment, such as transaction business, dating business, work information related business, rebate business (i.e., business related to returning benefits), counterfeit goods related business, account theft related business, product discount related business, financial credit related business, etc. Among them, business related to returning benefits may include, but is not limited to: financial rebate type (such as rebate red envelope rain, word games, social activity rebates, charitable donations, real and fake payment codes, etc.) and shopping rebate type (such as rock-paper-scissors to win prizes, 100% winning rate, charity shopping, fan reward benefits, etc.).

[0071] Business operations: Operations performed by the object in the target business, such as, but not limited to, electronic resource transfer operations (e.g., receiving electronic resources or transferring electronic resources), and goods delivery operations. Those skilled in the art can set these up according to actual needs.

[0072] Feedback data can be data triggered by the business operations performed by the account using the target business in relation to the aforementioned target object. For example, feedback data can include, but is not limited to, account complaints and praise.

[0073] The electronic resource transfer operation in this application embodiment may include the operation of transferring electronic resources, which can also be called mobile payment. Mobile payment is an account making a payment operation through a mobile network. Among them, the electronic resources involved in this application embodiment can be at least one of funds and information resources. The funds mentioned above can include at least fiat currency, electronic money, etc. Fiat currency refers to a currency that is legally mandated to circulate and be used. Electronic money refers to currency stored in electronic form in the electronic wallet held by the account (such as wallets in applications that support mobile payment, but not limited to them). The information resources mentioned above can include, but are not limited to, game resources (such as game equipment), multimedia resources (such as video, audio, etc.), and electronic coupons (such as electronic group buying coupons, electronic discount coupons, etc.).

[0074] 3) Operational agreement information and target information

[0075] The operational agreement information in this application embodiment may be, but is not limited to, information used to constrain business operations in the target business. For example, the operational agreement information may be, but is not limited to, protocols for business operations in the target business. The specific content of the above-mentioned operational agreement information is not limited in this application embodiment. Those skilled in the art can set the above-mentioned operational agreement information according to actual needs. Here, we take merchants as the object and rebate business as the target business for illustrative purposes. The operational agreement information associated with the rebate business may include, but is not limited to, sending electronic resource value to the user within the rebate period agreed upon by the merchant and the target user (i.e., the account using the rebate business), sending the agreed electronic resource value to the target user, and sending electronic resources to the target user according to the agreed electronic resource transfer method.

[0076] In this embodiment, the target information is information that characterizes the fact that the business operation of the object in the target business does not meet the operation agreement information associated with the target business. That is, the target information can be understood as complaint information about the business operation in the target business. For example, if the target business is a rebate business, the target information may be, but is not limited to, "Object X has not returned the agreed benefits" or "Object X returned less electronic resource value than agreed". Those skilled in the art can set it according to actual needs.

[0077] 4) Object state

[0078] In this application's embodiments, the object state refers to the object's state in the target business, which may include, but is not limited to, a normal state or an abnormal state. A normal state can be understood, but is not limited to, that the object's business operations in the target business meet the operational conventions associated with the target business. An abnormal state can be understood, but is not limited to, that the object's business operations in the target business do not meet the operational conventions associated with the target business. For example, if an object engages in fraudulent operations in the target business, it can be considered that the object's business operations in the target business do not meet the operational conventions associated with the target business, i.e., the object's state in that target business can be considered an abnormal state. If the target business is related to electronic resource transfer, and an object is currently prohibited from performing electronic resource transfer, it can be considered, but is not limited to, that the object's state in that target business is an abnormal state. Those skilled in the art can set the abnormal state of the object in different target businesses according to actual needs.

[0079] 5) Fraudulent operations

[0080] The fraudulent operations involved in this application mainly include online fraud, where a fraudulent account (also known as a fraudster) uses false information (such as, but not limited to, false product information or job information) through social networks and information exchange platforms to induce a victim account (also known as a victim) to voluntarily transfer electronic resources. If the victim believes the false information sent by the fraudster, they may voluntarily transfer their electronic resources to the fraudster. Currently, fraudsters may conduct fraudulent operations in transaction services, dating services, job information-related services, rebate services, counterfeit goods-related services, account theft-related services, product discount-related services, and financial credit-related services, especially those involving the return of benefits. Therefore, the fraudulent operations in this application can include, but are not limited to, transaction operations in transaction services, dating operations in dating services, part-time job fraud in job information-related services, rebate fraud in rebate services, counterfeit goods, account theft, and product discount fraud.

[0081] This application relates to Artificial Intelligence (AI) and machine learning technologies, and is designed based on computer vision and machine learning (ML) technologies within AI. Artificial intelligence utilizes digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceiving the environment, acquiring knowledge, and using that knowledge to obtain optimal results—theories, methods, technologies, and application systems. In other words, artificial intelligence is a comprehensive technology within computer science that attempts to understand the essence of intelligence and produce a new kind of intelligent machine capable of reacting in a manner similar to human intelligence.

[0082] Artificial intelligence (AI) studies the design principles and implementation methods of various intelligent machines, enabling them to perceive, reason, and make decisions. AI technology mainly includes computer vision, natural language processing, and machine learning or deep learning. With the research and advancement of AI technology, it is being researched and applied in many fields, such as smart homes, intelligent customer service, virtual assistants, smart speakers, intelligent marketing, autonomous driving, robotics, and smart healthcare. It is believed that with further technological development, AI will be applied in even more fields and play an increasingly important role.

[0083] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments in this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0084] The design concept of this application will be explained below.

[0085] With the development of internet technology, mobile payment has become an integral part of daily life, greatly improving users' consumption and payment methods. Currently, in the process of identifying objects with abnormal status in mobile payments, the method usually relies on the user's complaint regarding the business operation of the aforementioned object. The user selects the target business associated with the aforementioned business operation and the content of the complaint, directly judging whether the object's status in the aforementioned target business is abnormal. However, users may not be able to accurately select the target business associated with the aforementioned business operation, and the content of the user's complaint is highly subjective, making the authenticity of the information in the complaint unreliable. Therefore, directly identifying whether the object's status in the target business is abnormal based on the user's complaint data and the selected target business has low accuracy.

[0086] In view of this, this application embodiment designs an object state recognition method to improve the accuracy of identifying whether the object state of an object is abnormal. In this application embodiment, the feedback data of the object to be identified in the target business is first obtained, and whether the business operation that triggers each feedback data meets the operation agreement information associated with the target business is identified. Then, based on the recognition results corresponding to each feedback data, it is determined whether the object state of the object to be identified in the target business is abnormal.

[0087] Furthermore, to improve the accuracy of identifying whether the object status of the target object is abnormal in the target business, this embodiment of the application can also combine the identification results corresponding to each feedback data with the object attribute information associated with the object to be identified to determine whether the object status of the target object is abnormal in the target business. In this embodiment of the application, the obtained feedback data, object attribute information, etc. are used for positive channels, such as anti-money laundering and anti-fraud.

[0088] Furthermore, in order to improve the efficiency of determining whether the object state of the object to be identified is abnormal in the target business, the embodiments of this application can also determine whether the object state of the object to be identified is abnormal in the target business based on the target scorecard model or a trained state recognition model; the above state recognition model is not limited in many ways, and those skilled in the art can set it according to actual needs, such as creating the above state recognition model based on a neural network model. The above neural network model can include, but is not limited to, convolutional neural networks (CNN), recurrent neural networks (RNN), etc.

[0089] To better understand the design concept of this application, the following examples illustrate the application scenarios in the embodiments of this application.

[0090] Please see Figure 1 This provides an application scenario for object state recognition, which may include a terminal device 110 and a server 120; the terminal device 110 and the server 120 can communicate via a network, wherein:

[0091] The terminal device 110 may have a target application installed. This target application supports mobile payment functionality. The target application may include, but is not limited to, at least one of the following: payment applications, social applications, multimedia applications, and assistance tool applications. This embodiment does not limit the specific type of application. In this embodiment, the object to be identified and the account using the target service can log in to the target application and perform corresponding operations. For example, after logging in to the target application, the object to be identified can perform electronic resource transfer operations or other business operations. After logging in to the target application, the object to be identified can also send messages to other objects or accounts, or receive messages sent by other objects or accounts. These other objects are objects other than the object to be identified. After logging in to the target application, the account using the target service can perform electronic resource transfer operations or other operations. The account can also trigger feedback data based on the business operations performed by the object to be identified in the target service.

[0092] As one example, please refer to Figure 2 In (a), after the object to be identified performs a business operation in the target business, the account using the target business can trigger a feedback operation for the business operation. Then, the terminal device 110 collects the feedback data indicated by the above feedback operation and sends the collected feedback data to the server 120. Then, the server 120 can receive the feedback data and record the received feedback data.

[0093] As one example, please refer to Figure 2 In (b), after the target service account triggers the service operation feedback operation for the service operation, the terminal device 110 can also send a data collection instruction to the server 120, and then the server 120 responds to the above data collection instruction, obtains the feedback data of the service operation feedback operation instruction and records it.

[0094] Server 120 acquires various feedback data of the object to be identified in the target business, identifies whether each feedback data contains target information, determines the identification result corresponding to each feedback data, and determines whether the object status of the object to be identified in the target business is abnormal based on the identification result corresponding to each feedback data; wherein, the target information is used to characterize the operation of the business that does not meet the operation convention information associated with the target business.

[0095] The terminal device 110 in this application embodiment may be a mobile terminal, a fixed terminal, or a portable terminal, such as a mobile phone, site, unit, device, multimedia computer, multimedia tablet, Internet node, communicator, desktop computer, laptop computer, notebook computer, netbook computer, tablet computer, personal communication system (PCS) device, personal navigation device, personal digital assistant (PDA), audio / video player, digital camera or camcorder, in-vehicle device, positioning device, television receiver, radio broadcast receiver, e-book device, gaming device, or any combination thereof, including accessories and peripherals of these devices or any combination thereof.

[0096] The server 120 in this embodiment can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or multiple cloud servers (such as, but not limited to, servers 120-1, 120-2 or 120-3 shown in the figure) that provide basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms. The functions of the server 120 can be implemented by one or more cloud servers, or by one or more cloud server clusters, etc.

[0097] The object state recognition method provided in the embodiments of this application will be described in detail below. It should be noted that the above application scenarios are only shown for the purpose of understanding the spirit and principles of this application, and the implementation of this application is not limited in any way.

[0098] based on Figure 1 The following describes an example of an object state recognition method involved in this application scenario. Please refer to [link / reference]. Figure 3 This diagram illustrates a flowchart of an object state recognition method designed according to an embodiment of this application, specifically including the following steps:

[0099] Step S301: Obtain various feedback data of the object to be identified in the target service; wherein, the feedback data is triggered by the business operations performed by the account using the target service for the object to be identified in the target service.

[0100] As one embodiment, in step S301, feedback data of the object to be identified in the target business within a preset time period can also be obtained, so as to determine whether the object status of the object to be identified in the target business within the preset time period is abnormal based on the feedback data within the preset time period, thereby improving the flexibility of determining whether the object status of the object to be identified is abnormal. The preset time period is not limited, and those skilled in the art can set it according to actual needs. For example, it can be, but is not limited to, a period of time before the current moment, such as 1 day, 3 days, 5 days, 7 days, or 1 month before the current moment; or it can be set to the period from the second time before the current moment to the second time before the current moment. For example, if the current moment is February 1, 2021, the period between December 1, 2020 (i.e., the first time mentioned above) and January 1, 2020 (i.e., the second time mentioned above) can be set as the preset time period, etc.

[0101] Step S302: Identify whether each of the above feedback data contains target information, and determine the identification result corresponding to each of the above feedback data. The target information represents the operation agreement information associated with the target business that the above business operation does not meet the above target business association information.

[0102] As one embodiment, the above-mentioned identification result may be, but is not limited to, the feedback data containing the target information. The identification result may also be that the feedback data does not contain the target information. For ease of description, in the following embodiments of this application, the identification result containing the target information in the feedback data is described as first identification information, and the identification result not containing the target information in the feedback data is described as second identification information. For example, taking merchant A as the object to be identified, the rebate business as the target business, and the target information as the business operation not meeting the operation agreement information associated with the rebate business, let's illustrate this. If merchant A's feedback data 1 in the rebate business is "Merchant A has not performed any rebate-related operations," then merchant A's feedback in the rebate business... Data 2 indicates "Merchant A did not ship product X". Feedback data 3 from Merchant A's rebate business indicates "Merchant A's rebate amount is much less than the agreed amount". Based on the above-mentioned operational agreement information associated with the rebate business, it can be determined that the identification results of feedback data 1 and 3 are the first identification information, and the identification result of feedback data 2 is the second identification information. In this embodiment, the specific form of the above-mentioned first identification information and second identification information is not limited. Those skilled in the art can set it according to actual needs. For example, the first identification information can be represented by a first label, and the second identification information can be represented by a second label. The first label can be, but is not limited to, "1", and the second label can be, but is not limited to, "0".

[0103] As one embodiment, to improve the accuracy of identifying whether each feedback data contains target information, this embodiment can identify each feedback data based on the context information of each word contained in each feedback data. Specifically, for each of the above feedback data, the following operations can be performed respectively: extract the text features to be processed corresponding to the above feedback data according to the context information of each word contained in one of the above feedback data; based on the text features to be processed, identify whether the above feedback data contains the target information, and determine the identification result corresponding to the above feedback data; wherein, in this embodiment, the context information of a word may include, but is not limited to, the position information of the word in the feedback data, the semantic features of the word, etc.

[0104] As one embodiment, to improve processing efficiency and the accuracy of feature extraction for each feedback data, in step S302 above, a trained data classification model can be used to extract the text features corresponding to each feedback data, and based on the text features corresponding to each feedback data, the recognition result corresponding to each feedback data can be determined. The data classification model can include a language learning sub-model and a data prediction sub-model. The language learning sub-model can be used to extract features from each feedback data. There are no strict limitations on the language learning sub-model; in this embodiment, it can include, but is not limited to, pre-trained models, such as, but not limited to, BERT models, Fast-BERT models, Tiny-BERT models, or Word2vec models. The data prediction sub-model can be used to determine the recognition result corresponding to each feedback data. In this embodiment, it can include, but is not limited to, networks used for binary classification. Detailed information about the data classification model will be further explained below.

[0105] The Word2vec model described above uses a single-layer neural network (CBOW network) to map sparse word vectors in one-hot form into dense vectors of n dimensions (n ​​can be, but is not limited to, several hundred). To accelerate model training, the tricks in the model in this embodiment can include, but are not limited to, Hierarchical softmax, negative sampling, or Huffman Tree. The Bert (Bidirectional Encoder Representations from Transformer) model described above is a bidirectional Transformer encoder network. The goal of the Bert model is to train on large-scale unlabeled corpora to obtain semantic representations of text containing rich semantic information, and then fine-tune the semantic representations of text in a specific Natural Language Processing (NLP) task before finally applying them to that specific NLP task. More detailed information about the data classification model described above will be provided below.

[0106] Step S303: Based on the identification results corresponding to each of the above feedback data, determine whether the object status of the object to be identified in the above target business is abnormal.

[0107] As one embodiment, in order to improve the accuracy of determining whether the object status of the object to be identified in the target service is abnormal, this embodiment can determine whether the object status of the object to be identified in the target service is abnormal based on the identification results corresponding to each of the feedback data and the object attribute information of the object to be identified. Specifically, it can be, but is not limited to, determining the feedback valid information corresponding to the object to be identified based on the first identification information in the identification results corresponding to each of the feedback data, wherein the first identification information is used to characterize that the feedback data contains the target information; obtaining the object attribute information associated with the object to be identified; and determining whether the object status of the object to be identified in the target service is abnormal based on the feedback valid information and the object attribute information.

[0108] As one embodiment, the above-mentioned valid feedback information includes one or any combination of the following first type of valid feedback information and second type of valid feedback information: wherein, the first type of valid feedback information may be the number of first identification information in the identification results corresponding to each of the above-mentioned feedback data; the second type of valid feedback information may be the ratio of the number of the first identification information in the identification results corresponding to each of the above-mentioned feedback data to the total number of the above-mentioned feedback data.

[0109] For ease of understanding, an example of the two types of valid feedback information is given here. If the number of feedback data of the target object in the target business is N1, and N2 (N2 is less than or equal to N1) of the above N1 feedback data contain target information, then the number of first identification information in the identification results corresponding to the N1 feedback data is N2. That is, in this case, N2 can be determined as the first type of valid feedback information, and (N2 / N1) can be determined as the second type of valid feedback information.

[0110] As an example, in order to improve the efficiency of determining whether the object state of the object to be identified in the target business is abnormal, this embodiment of the application can also determine whether the object state of the object to be identified in the target business is abnormal based on the target scoring card model or the trained state recognition model, based on the above-mentioned effective feedback information and the above-mentioned object attribute information. The details will be further explained below.

[0111] As one embodiment, the object attribute information in this application embodiment may refer to, but is not limited to, data after tagging the relevant information of the object to be identified; the object attribute information may include, but is not limited to, at least one of the following information: transfer information of the electronic resources of the object to be identified, fraud anomaly information corresponding to the object to be identified (the fraud anomaly information indicates the degree of suspicion of network fraud operation of the object to be identified), operation information associated with abnormal business operation (the abnormal business operation includes business operation associated with feedback data containing the above target information), creation time of the object to be identified, asset status of the object to be identified, credit rating of the object to be identified, and information mined from the historical feedback data of the object to be identified, etc.; wherein, when the object to be identified is a merchant, the above object attribute information may also include the merchant's revenue status, the merchant's credit rating, the merchant's business registration information, etc.

[0112] Furthermore, if the aforementioned object attribute information includes the operation information associated with the aforementioned abnormal business operation, the aforementioned operation information includes at least one of the following: the electronic resource value associated with the aforementioned abnormal business operation, the account associated with the aforementioned abnormal business operation, the business object associated with the aforementioned abnormal business operation, the trigger time information of the aforementioned abnormal business operation, etc.

[0113] As one embodiment, the following content further describes the data classification model involved in step S302. Please refer to [link / reference]. Figure 4 This application provides a schematic diagram of the structure of a data classification model, which may include at least a language learning sub-model 401 and a data prediction sub-model 402, wherein:

[0114] The aforementioned language learning sub-model 401 can extract features based on the contextual information of each word contained in each feedback data to obtain the text features to be processed corresponding to each feedback data. Specifically, for one feedback data, the feedback data can be input into a trained data classification model. Based on the language learning sub-model 401 in the data classification model, features are extracted from the contextual information of each word contained in the feedback data to obtain the text features to be processed corresponding to the feedback data. The language learning sub-model is trained using historical feedback data from the target business as training samples, based on the contextual information of each word contained in the training samples. The specific content of the language learning sub-model 401 will be further explained below.

[0115] As one embodiment, in the process of obtaining the text features to be processed corresponding to the above-mentioned feedback data based on the language learning sub-model 401, the above-mentioned feedback data can first be segmented to obtain the words contained in the above-mentioned feedback data. Then, through processing methods such as encoding (Embedding), the context information of each word in the above-mentioned feedback data is processed to obtain the word representation of each word contained in the above-mentioned feedback data (such as, but not limited to, word representation E1, word representation E2, and word representation EN shown in the figure, where N is a positive integer). The word representation can include the position information of the word in the feedback data, the semantic information of the word, etc. Then, after processing through multiple layers of units Trm, a word feature can be extracted for each word representation (as shown in the figure, word features T1 to TN can be extracted for word representations E1 to EN respectively). Then, the word features of the above-mentioned words can be processed to obtain the text features to be processed corresponding to the above-mentioned feedback data. For example, the word features of the above-mentioned words can be concatenated, and the concatenated result is determined as the text features to be processed corresponding to the above-mentioned feedback data.

[0116] The aforementioned data prediction sub-model 402 can identify whether each feedback data contains target information based on the text features corresponding to each feedback data, and determine the recognition result corresponding to each feedback data. Specifically, after obtaining the text features corresponding to a feedback data based on the aforementioned language learning sub-model, the data prediction sub-model 402 can predict the second correlation between the text features and the first recognition information based on the first correlation learned, and determine the recognition result corresponding to the feedback data based on the second correlation. The first recognition information represents that the feedback data contains the target information, and the first correlation is determined based on the degree of correlation between the historical text features corresponding to historical feedback data and the first recognition information.

[0117] As one embodiment, in order to improve the accuracy of identifying each feedback data, in this embodiment, when determining the identification result corresponding to the above-mentioned feedback data based on the above-mentioned second correlation, if the above-mentioned second correlation is greater than the first set threshold, then the identification result corresponding to the above-mentioned feedback data is determined to be that the above-mentioned feedback data contains the above-mentioned target information; if the above-mentioned second correlation is not greater than the first set threshold, then the identification result corresponding to the above-mentioned feedback data is determined to be that the above-mentioned feedback data does not contain the above-mentioned target information; the above-mentioned first set threshold is not limited, and those skilled in the art can set it according to actual needs.

[0118] As an example, in order to improve the accuracy of the data classification model and increase the training efficiency of the data classification model, the language learning sub-model 401 can be trained first to obtain a trained language learning sub-model 401. Then, after the trained language learning sub-model 401, a data prediction sub-model 402 is created to obtain an initial data classification model. The initial data classification model is then trained to obtain a trained data classification model.

[0119] The following is an exemplary introduction to the training process of the data classification model. First, the training process of the language learning sub-model 401 mentioned above will be further explained.

[0120] In this embodiment, historical feedback data from the target business can be used as training samples in the first sample set. Based on the contextual information of each word contained in the training samples in the first sample set, the language learning sub-model is trained by feature extraction. Specifically, the language learning sub-model can be trained at least once based on the first sample set to obtain the trained language learning sub-model. The training samples in the first sample set can include historical feedback data of one historical object in the target business, or historical feedback data of multiple historical objects in the target business. The historical object can include the object to be identified, or it can exclude the object to be identified. Those skilled in the art can obtain a certain amount of historical feedback data as training samples in the first sample set according to actual needs.

[0121] As one embodiment, in order to improve the accuracy of text feature extraction by the language learning model, this embodiment can set a first training termination condition for the training process of the language learning sub-model, and then output the language learning sub-model being trained when the first training termination condition is met during the training process of the language learning sub-model; wherein, the first training termination condition is not limited in much detail, and those skilled in the art can set it according to actual needs.

[0122] As one embodiment, in a training operation of the language learning sub-model 401, text prediction operations can be performed on each historical feedback data in the first sample set to determine the model prediction error of the language learning sub-model. Based on the model prediction error, the model parameters of the language learning sub-model can be adjusted. For details, please refer to [link to relevant documentation]. Figure 5 This document provides a schematic diagram of the process for a single training operation of the language learning sub-model 401. A single training operation may include, but is not limited to, the following steps S501 and S502:

[0123] Step S501: For each historical feedback data obtained from the first sample set, perform a text prediction operation respectively to determine the prediction deviation corresponding to each of the above historical feedback data.

[0124] As an embodiment, the prediction deviation corresponding to a historical feedback data can represent the error information of text prediction for some words in the above historical feedback data by the language learning sub-model; in the text prediction operation for a historical feedback data, it can be but is not limited to predicting some words based on the context information of some words in the above historical feedback data in the above historical feedback data; specifically, the above text prediction operation can be but is not limited to including the following steps S5011 to step S5014.

[0125] Step S5011: Perform word segmentation on a historical feedback data among the above historical feedback data to obtain each word included in the above historical feedback data.

[0126] As an embodiment, the specific method for performing word segmentation on the above historical feedback data is not overly limited, and those skilled in the art can set it according to actual needs. For example, it can be but is not limited to performing Jieba word segmentation on the historical feedback data, and then processing the result of Jieba word segmentation based on the stop word preprocessing technology to obtain each word included in the above historical feedback data; among them, the stop words involved in the above stop word preprocessing can be but is not limited to including English characters, numbers, mathematical characters, punctuation marks, and Chinese characters with particularly high usage frequency and no actual meaning (such as can be but is not limited to including "de", "zai", "he", "yiji", etc.). In the embodiments of the present application, removing stop words in the feedback data can improve the accuracy of identifying the feedback data.

[0127] Step S5012: Mask some words among each word included in the above historical feedback data.

[0128] As an embodiment, in this step, it can be but is not limited to using a preset word mask Mask to randomly block one or more words among each word included in the above historical feedback data, and the specific form of the above word mask Mask is not limited, and those skilled in the art can set it according to actual needs.

[0129] Step S5013: Determine the context information of the above partial words in the above historical feedback data, and select candidate words from the pre-configured candidate word library whose matching degree with the determined context information meets the matching degree condition. The above candidate word library is determined based on the above first sample set.

[0130] As an example, before training the language learning sub-model, word segmentation can be performed on each historical feedback data in the first sample set, but is not limited to this step. Each word obtained from the word segmentation is determined as a candidate word, and the resulting set of candidate words is determined as the aforementioned candidate word library. The specific method of performing word segmentation on each historical feedback data can be referred to step S5011, and will not be repeated here.

[0131] As an example, the matching degree is not limited to the above-mentioned matching degree conditions. Those skilled in the art can set it according to actual needs. For example, the matching degree with the largest value among the matching degrees between candidate words and determined context information can be determined as the matching degree that meets the matching degree conditions. Alternatively, the matching degree that is closest to the matching degree threshold among the matching degrees between candidate words and determined context information can be determined as the matching degree that meets the matching degree conditions.

[0132] Step S5014: The deviation information between the above-mentioned partial words and the selected candidate words is determined as the prediction deviation corresponding to the above-mentioned historical feedback data.

[0133] As one embodiment, the prediction bias described above can be, but is not limited to, characterizing the degree of deviation between the aforementioned partial words and the selected candidate words. This degree of deviation can be negatively correlated with the degree of matching between the partial words and the selected candidate words. In this embodiment, the specific method for determining the prediction bias can be set according to actual needs.

[0134] Step S502: Based on the prediction deviations corresponding to each of the aforementioned historical feedback data, adjust the parameters of the aforementioned language learning sub-model.

[0135] As one example, the prediction error of the language learning sub-model can be determined based on the prediction deviation corresponding to each historical feedback data. The parameters of the language learning sub-model can be adjusted based on the prediction error, such as adjusting the model parameters of the language learning sub-model in a direction that reduces the prediction error.

[0136] To enhance the flexibility of the implementation of the solution, this application embodiment does not impose too many restrictions on the specific method for determining the above-mentioned model prediction error. In this application embodiment, it can be flexibly set according to actual business needs, such as determining the average of the prediction deviations corresponding to each historical feedback data as the above-mentioned model prediction error; or the model prediction error of the language learning sub-model can be determined based on the principle of the following formula (1):

[0137]

[0138] In formula (1), K2 can be, but is not limited to, the total number of historical feedback data obtained from the training sample set, K1 is the number of historical feedback data with correct text prediction, and P2 is the prediction error of the above model; wherein, the historical feedback data with correct text prediction can be, but is not limited to, historical feedback data with prediction deviation greater than the prediction deviation threshold, and the historical feedback data with correct text prediction can also be the historical feedback data of the selected candidate words that are the masked words themselves, etc., which can be set by those skilled in the art according to actual needs.

[0139] As one embodiment, after obtaining the trained language learning sub-model, a certain amount of historical feedback data from the target business can be obtained. The obtained historical rebate feedback data is labeled with data type and used as training samples in the second sample set. Then, the data prediction model in the data classification model is trained based on the second sample set. The data type labeled for the historical feedback data can be the first identification information or the second identification information mentioned above. If a piece of historical feedback data contains the target information mentioned above, the data type of the historical feedback data is labeled as the first identification information. If a piece of historical feedback data contains the target information mentioned above, the data type of the historical feedback data is labeled as the second identification information.

[0140] Please see Figure 6 Furthermore, during the training of the data prediction sub-model based on the second sample set, the text features to be processed corresponding to each historical feedback data can be obtained based on the language learning sub-model. Then, the text features to be processed corresponding to each historical feedback data in the second sample set can be identified based on the data prediction sub-model to determine the predicted data type corresponding to each historical feedback data in the second sample set (the predicted data type is the identification result of whether the above feedback data contains target information). Then, based on the deviation information between the predicted data type and the labeled data type of each historical feedback data in the second sample set, the data prediction sub-model is trained. When the second training termination condition is met, the current language learning sub-model and data prediction sub-model are output as a data classification model. The second training termination condition is not limited in many ways, and those skilled in the art can set it according to actual needs.

[0141] As one example, please continue to see Figure 6 To further improve the accuracy of the data prediction sub-model in the data classification model, after determining that the second training termination condition is met, the above data prediction sub-model can be further adjusted based on the test sample set, which includes a certain amount of historical feedback data after labeling data types.

[0142] As one example, please continue to see Figure 6To further improve the accuracy of the data prediction sub-model in the data classification model and enhance the efficiency of training the data prediction sub-model, data preprocessing can be performed on the acquired historical feedback data before training. This includes data cleaning of historical rebate feedback data to remove abnormal historical feedback data. Those skilled in the art can define abnormal historical feedback data according to actual needs, such as identifying historical feedback data with empty information, or historical feedback data containing numeric or English strings or garbled characters without real semantics. Furthermore, data cleaning can be performed on the data type of the labeled historical feedback data to reduce the impact of abnormal historical feedback data on the data prediction sub-model and improve the accuracy of the trained data prediction sub-model in identifying whether the feedback data contains target information.

[0143] As one embodiment, the following describes the process in step S303 of determining whether the object state of the object to be identified in the target service is abnormal based on the target scoring card model or the trained state recognition model. For details, please refer to the specific content of the first object evaluation method and the second object evaluation method below.

[0144] The first object evaluation method: Based on the target scorecard model, determine whether the object to be identified is in an abnormal state in the target business.

[0145] Specifically, based on the aforementioned target scoring card model, the aforementioned valid feedback information and the aforementioned object attribute information are processed to obtain the target operation evaluation value of the object to be identified for the aforementioned target business; based on the aforementioned target operation evaluation value, it is determined whether the object status of the object to be identified in the aforementioned target business is abnormal, wherein the aforementioned target operation evaluation value is an abbreviation for the operation evaluation value of the object to be identified for the target business.

[0146] In this embodiment of the application, the target scorecard model is used to predict the target operation evaluation value of the object to be identified in the target business when the object state is in an abnormal state; in this embodiment of the application, the above-mentioned feedback effective information and the above-mentioned object attribute information can be used as variables of the target scorecard model to construct a logistic regression model, and the constructed logistic regression model is determined as the above-mentioned target scorecard model; for ease of understanding, please refer to the following formulas (2) to (4) to provide an example of a target scorecard model;

[0147] Score=A1-B1log(Odds) Formula (2)

[0148]

[0149] log(Odds)=a0+a1×x1+a2×x2+a3×x3+…+an×xn Formula (4)

[0150] In formulas (2) to (4), A1 and B1 are constants, Odds is the probability that the object to be identified is in an abnormal state in the target business, p is the probability that the object to be identified will perform a business operation that does not meet the operation agreement information associated with the target business, and Score is the target operation evaluation value of the object to be identified for the target business. That is, the larger the Odds of the object to be identified, the lower the target operation evaluation value of the object to be identified for the target business; x0 to xn are different variables, a0 is a constant, and a1 to an are the parameters of the target variables x1 to xn; if the object state of the object that performs fraudulent operation in the target business is regarded as an abnormal state, then the above Odds is the probability that the object to be identified will perform fraudulent operation in the target business, and the above p is the probability that the object to be identified will perform fraudulent operation in the target business.

[0151] As an example, one of the above object attribute information can be used as a variable from x0 to xn. For example, when the object attribute information includes the electronic resource transfer value of the object to be identified within a set time period, the fraud anomaly level corresponding to the object to be identified, and the asset value of the object to be identified, the above formula (4) can be transformed into formula (4a).

[0152] log(Odds)=a0+a1×x1+a2×x2+a3×x3+a4×x4 Formula (4a)

[0153] In formula (4a), x1 to x4 represent the valid feedback information of the object to be identified, the electronic resource transfer value of the object to be identified within a set time period, the fraud anomaly information corresponding to the object to be identified, and the asset value of the object to be identified, respectively; a0 is a constant, a1 to a4 are the parameters of variables x1 to x4, and Odds is the probability that the object to be identified is in an abnormal state in the target business; if the object state of the object that has committed fraud in the target business is considered to be in an abnormal state, then the above Odds is the probability that the object to be identified has committed fraud in the target business.

[0154] As an example, the operation evaluation value can be set to be negatively correlated with the degree of suspicion that the object's state in the target business is abnormal (e.g., but not limited to referring to the above formula (2)). Therefore, in the process of determining whether the object state of the object to be identified is abnormal in the target business based on the above target operation evaluation value in this embodiment of the application, if it is determined that the above target operation evaluation value is lower than the first evaluation value threshold, then the object to be identified is determined to be an abnormal object; wherein, the abnormal object includes objects whose object state in the target business is abnormal, such as objects whose abnormal state can be, but are not limited to, objects that have performed fraudulent operations in the target business.

[0155] As one embodiment, the operation evaluation value can be set to be positively correlated with the degree of suspicion that the object's state in the target business is abnormal. For example, if the above formula (2) is converted into the form of formula (2a), then in this case, in the process of determining whether the object state of the object to be identified in the target business is abnormal based on the above target operation evaluation value, if it is determined that the above target operation evaluation value is higher than the second evaluation value threshold, then the object to be identified is determined to be an abnormal object. The description of the abnormal object can be found in the above content, and will not be repeated here. At the same time, the above first evaluation value threshold and the second evaluation value threshold are not limited, and those skilled in the art can set them according to actual needs.

[0156] Score = A² + B²log(Odds) (Formula 2a)

[0157] In formula (2a), A2 and B2 are constants, Odds is the probability that the object to be identified is in an abnormal state in the target business, and Score is the target operation evaluation value of the object to be identified for the target business. That is, the larger the Odds of the object to be identified, the higher the target operation evaluation value of the object to be identified for the target business. If the object state of the object that has committed fraud in the target business is regarded as an abnormal state, then the above Odds is the probability that the object to be identified has committed fraud in the target business.

[0158] As one embodiment, to improve the accuracy of evaluating the object to be identified, this embodiment can also classify the object into abnormal object levels. In determining whether the object to be identified is abnormal in the target business, the operation evaluation value range to which the target operation evaluation value belongs can be determined based on the operation evaluation value range mapped to at least one object abnormality level. The object abnormality level mapped to the determined operation evaluation value range is then determined as the object abnormality level corresponding to the object to be identified. The classification of object abnormality levels is not overly limited; those skilled in the art can set it according to actual needs. For example, different object abnormality levels can be determined based on the degree of suspicion that the object's state in the target business is abnormal, such as setting the object abnormality level as normal object, low-abnormal object, medium-abnormal object, and high-abnormal object. If the abnormal object is an object with fraudulent operations, the object abnormality level can be set as normal object, low-abnormal fraudulent object, medium-abnormal fraudulent object, and high-abnormal fraudulent object. The mapping relationship between the object abnormality level and the operation evaluation value range is also not overly limited; those skilled in the art can set it according to actual needs.

[0159] As one embodiment, to improve the accuracy of determining whether the object state of the object to be identified is abnormal based on the target scorecard model, this embodiment of the application can further pre-classify the object types for different objects, create corresponding target scorecard models for different object types, and then, before processing the above-mentioned feedback valid information and the above-mentioned object attribute information based on the above-mentioned target scorecard model to obtain the target operation evaluation value of the object to be identified for the above-mentioned target business, the object type of the object to be identified can be further determined; the target scorecard model corresponding to the above-mentioned object type can be obtained from the target scorecard model set associated with the above-mentioned target business; wherein, in this embodiment of the application, the object type of the object to be identified can be determined based on the identity indication information or type indication information of the object to be identified, etc.

[0160] It should be noted that the object type in this application embodiment is a way of distinguishing objects involved in this application embodiment. In this application embodiment, objects can be divided into different types according to different distinguishing dimensions. For example, when the object is a merchant, the merchant can be divided into direct-connection merchants, sub-merchants, and micro-merchants, but is not limited to using the merchant's electronic resource settlement method as the distinguishing dimension. Direct-connection merchants can be large merchants that directly settle electronic resources with the target application. Sub-merchants can be merchants that settle electronic resources with other applications besides the target application, and the other applications then settle electronic resources with the target application. Micro-merchants can be small merchants that directly settle electronic resources with the target application. In this application embodiment, the main operating time can also be used as the distinguishing dimension to divide merchants into weekday merchants and holiday merchants. Alternatively, the main operating time can be used as the distinguishing dimension to divide merchants into ordinary merchants and late-night merchants. Weekdays are usually used to refer to Monday to Friday of each week, while holidays are usually used to refer to Saturday and Sunday of each week.

[0161] For ease of understanding, the following further explains the process of obtaining the target scorecard model involved in the embodiments of this application:

[0162] Please see Figure 7 This application provides a flowchart for creating a target scorecard model. In this embodiment, the process involves data acquisition, exploratory data analysis (EDA), object type and rating analysis, data preprocessing, variable screening, construction of a logistic regression (LR) model, training of the logistic regression model, evaluation of the trained logistic regression model, and output of the logistic regression model that meets the evaluation conditions as the target scorecard model. In the process of training the logistic regression model, the goal is to find the ideal values ​​of parameters x1 to xn in the above formula (4) to improve the prediction accuracy of the target scorecard model. After obtaining the target scorecard model, it can also be deployed to a real-time system for use.

[0163] As one embodiment, in the above data acquisition process, information such as whether the manually reviewed object is an abnormal object with an abnormal status can be used as a variable in the target scorecard model, valid feedback information can be used as a variable in the target scorecard model, and object type or abnormal rating of the object can be used as a variable in the target scorecard model. For example, taking merchants as instances of objects, return benefits as the target business, and merchants who engage in fraudulent operations in return benefits as abnormal objects with an abnormal status, the information on whether the merchant is an abnormal merchant engaging in fraudulent operations in the target business, the valid feedback information, the merchant type, and the merchant's abnormal rating can each be used as a variable in the target scorecard model. The method for classifying merchant types can be found above and will not be repeated here.

[0164] As one example, in the above EDA processing, information on data related to each variable can be obtained, but is not limited to, such as the missing value situation, outlier situation, mean, median, maximum, minimum, and distribution of each field related to each variable, in order to formulate a reasonable data preprocessing plan; the above object type and rating analysis can be, but is not limited to, further classifying object types and object outlier ratings based on the differences in data after exploratory data analysis.

[0165] As one embodiment, the above data preprocessing process may include, but is not limited to, data cleaning, variable binning, and WOE transformation. Among them, the data cleaning process mainly deals with dirty data, missing values, and outliers in the original data. For example, it may include, but is not limited to, directly deleting variables with a missing rate greater than the missing rate threshold, and treating the missing values ​​and outliers in the remaining variables as a state.

[0166] As one example, in the process of variable binning and WOE transformation, it is possible, but not limited to, segmenting and discretizing continuously numerical variables, and merging multi-state discrete variables to reduce the number of states of discrete variables and improve the processing efficiency of the target scorecard model. Specifically, it is possible, but not limited to, using woebin in the scorecardpy library for binning. woebin supports decision tree binning, chi-square binning, and custom binning. The default WOE value is calculated as the ratio of bad samples or the ratio of good samples. The WOE value can be adjusted by parameters. For example, if a bin contains only good or bad samples, the missing category can be assigned 0.99 to adjust the WOE value.

[0167] As one embodiment, the variable screening process described above may include univariate screening and variable correlation analysis. In univariate screening, variables can be screened based on their Information Value (IV). The higher the IV of a variable, the stronger its ability to evaluate the target. This process filters out individual variables with high IV values. In the variable correlation analysis process, pairwise correlation analysis and multicollinearity analysis can be performed to eliminate features whose correlation with the target variable is below a correlation threshold, thus eliminating linearly correlated variables, avoiding feature redundancy, reducing the burden of later verification, deployment, and monitoring, and ensuring the interpretability of the variables.

[0168] As one embodiment, in the process of constructing the logistic regression model described above, a preliminary logistic regression model can be constructed, variables can be selected based on p-value, and further selection can be performed based on the signs of the coefficients of each variable. The final logistic regression model is then output as the target scorecard model. In the process of model evaluation, the KS value index can be used, where the KS value represents the ability of the target scorecard model to distinguish between normal and abnormal objects. Its essence is the maximum value of the TPR-FPR change with the threshold of normal and abnormal objects. The range of the KS value can be, but is not limited to, between 0.5 and 1. The larger the KS value, the better the accuracy of the target scorecard model in evaluating the object to be identified. In this embodiment, when KS > 0.4, the current target scorecard model can be output as the final target scorecard model.

[0169] The second object evaluation method is based on a state recognition model to determine whether the object to be identified is in an abnormal state in the target business.

[0170] As one embodiment, the state recognition model involved in this application embodiment may include, but is not limited to, a neural network model for classifying objects, such as a binary classification model.

[0171] As one embodiment, in the process of determining whether the object state of the object to be identified in the target service is abnormal based on the above-mentioned valid feedback information and object attribute information, this embodiment of the application can input the above-mentioned valid feedback information and object attribute information into a trained state recognition model, and perform the following processing through the state recognition model:

[0172] The aforementioned valid feedback information and object attribute information are determined as the features of the object to be processed corresponding to the object to be identified. Based on the third correlation degree learned by the aforementioned state recognition model, the fourth correlation degree between the features of the object to be processed and the target object recognition result is predicted. If the fourth correlation degree is greater than the second set threshold, the object to be identified is determined to be an abnormal object. The aforementioned third correlation degree is determined based on the degree of correlation between the historical object features corresponding to the historical object and the aforementioned target recognition result. The abnormal object includes objects whose object state is abnormal in the aforementioned target business. The aforementioned target object recognition result is used to characterize that the object's object state is abnormal in the aforementioned target business. That is, if the abnormal object is an object that has committed fraudulent operation in the aforementioned target business, the aforementioned target object recognition result is used to characterize that the object has committed fraudulent operation in the aforementioned target business.

[0173] As an example, the training process of the state recognition model is explained below using the object that has committed fraud in the target business as the above-mentioned abnormal object. During the training of the state recognition model, the historical object features (including the feedback validity information and object attribute information corresponding to the historical object) corresponding to different historical objects can be labeled to determine whether the object has committed fraud in the target business. These labels are then used as training samples in a third sample set, and the state recognition model is trained based on this third sample set. The labeling of historical object features to determine whether the object has committed fraud in the target business can be for normal or abnormal objects. If a historical object feature corresponds to a historical object that has committed fraud in the target business, then that historical object feature is labeled as an abnormal object; if a historical object feature corresponds to a historical object that has not committed fraud in the target business, then that historical object feature is labeled as a normal object.

[0174] As one embodiment, during the training process of the state recognition model, the state recognition model estimates whether each historical object feature is an abnormal object based on the aforementioned historical object features. Based on the annotation information corresponding to each historical object and the deviation information of the object recognition results, the prediction error of the state recognition model is determined. Then, the parameters of the state recognition model are adjusted in the direction of reducing the prediction error of the state recognition model until the third training termination condition is met. The second training termination condition is not limited in much detail, and those skilled in the art can set it according to actual needs.

[0175] As one embodiment, in step S303 of this application embodiment, after determining that the object to be identified is an abnormal object through the first object evaluation method or the second object evaluation method, further penalty operations can be performed on the abnormal object. Specifically, after determining that the object to be identified is an abnormal object, at least one of the first permission and the second permission of the object to be identified can be cancelled, but is not limited to cancellation. The first permission includes the permission to perform electronic resource transfer operations, and the second permission includes electronic resource transfer operations for the target electronic resource value. The target electronic resource value is determined based on the feedback data. For example, if the object to be identified is a merchant and the target business is a rebate business, the target electronic resource value can be determined based on the rebate amount involved in the feedback data. For example, if the rebate amount involved in the feedback data is 1,000 yuan, then the target electronic resource value can be a resource value of 1,000 yuan.

[0176] As one embodiment, the training process of the data classification model, target scorecard model, and state recognition model involved in this application embodiment can be deployed on an offline system. After training and evaluating the data classification model, target scorecard model, and state recognition model, they can be deployed to a near-real-time system or a real-time system to improve the efficiency of object recognition based on feedback data. Specifically, models with high computational resource requirements can be deployed to a near-real-time system, and the final result can be returned to the real-time system after the computation is completed. For example, the aforementioned data classification model can be deployed to a near-real-time system. In the real-time system, after determining the recognition results corresponding to each feedback data based on the data classification model, the determined recognition results are written back to the real-time system. Since the target scorecard model requires relatively few computing resources, it can be directly deployed to the real-time system. After obtaining the recognition results corresponding to each feedback data and the object attribute information of the object to be identified (such as real-time cash flow and information data of the merchant's transactions when the object is a merchant) from the near-real-time system, the target operation evaluation value of the object to be identified for the target business can be determined directly on the real-time system based on the target scorecard model.

[0177] The following content of this application embodiment provides a complete example of an object state recognition method. In this example, a merchant is taken as the object to be identified, a rebate business (hereinafter referred to as rebate business) is taken as the target business, an object with fraudulent operation in the rebate business is taken as an abnormal object with an abnormal state, the Word2vec model is taken as an example of the language learning sub-model, and a classifier is taken as an example of the data prediction model.

[0178] Please see Figure 8In this example, the historical rebate feedback data of the merchant to be identified in the rebate business (i.e., the historical feedback data mentioned above) is first obtained. The data type of each historical rebate feedback data is labeled. If a historical rebate data contains target information, the data type of the historical rebate data can be labeled as valid rebate feedback. If a historical rebate data does not contain target information, the data type of the historical rebate data can be labeled as invalid rebate feedback. The above data classification model (including the Word2vec model and the above classifier) ​​is then trained on an offline system to obtain the trained data classification model. The trained data classification model is then deployed to a near real-time system for use. At the same time, the target scorecard model is trained offline and then deployed to a real-time system.

[0179] In the process of identifying whether a merchant is engaging in fraudulent activities in its cashback business, the system first acquires various cashback feedback data from the merchant. Then, based on a data classification model deployed in a near-real-time system (also known as a real-time transaction system), the acquired cashback feedback data is identified, determining the identification result (valid or invalid) for each data point. The identification results for each cashback feedback data are then written back to the real-time system (also known as a real-time transaction decision system). Further, based on a target scorecard model deployed in the real-time system and the identification results for each cashback feedback data, valid feedback information is determined. This valid feedback information and the object attribute information of the merchant are processed to obtain the target operation evaluation value for the cashback business. If the obtained target operation evaluation value is lower than a first evaluation value threshold, the merchant is automatically penalized (see above for details). If the obtained target operation evaluation value is not lower than the first evaluation value threshold, the merchant is allowed to proceed.

[0180] As an example, the training process of the data classification model in this example can be found in [link to example]. Figure 9In the training process of the data classification model, the Word2vec model is first pre-trained, and then a classifier is built on the pre-trained Word2vec model to form an initial data classification model. A certain amount of historical feedback data is then obtained, labeled, and used as training samples in the training sample set. This yields the text features to be processed for each historical feedback data point. The classifier is then trained based on these text features and labels. After training, the classifier is further trained based on these text features and labels in the test sample set. Specifically, the text features to be processed for a historical feedback data point can be obtained as follows: the Word2vec model is used to segment the historical feedback data point, and the word vectors (i.e., context information) of each word within the historical feedback data point are obtained. These word vectors are then summed (or concatenated) to obtain the text features to be processed for the historical feedback data point.

[0181] As one embodiment, this application also provides an overall system flow diagram of this example, which can be found in the following example. Figure 10 It mainly includes the data layer, model layer, and application layer; among which:

[0182] The data in the data layer may include cashback feedback data of the merchant to be identified in the cashback business, the merchant type of the merchant to be identified, and the object attribute information of the merchant to be identified (such as, but not limited to, the merchant classification, merchant cash flow, and merchant information flow shown in the figure).

[0183] The models involved in the model layer may include, but are not limited to, data classification models and target scorecard models. Among them, data classification models may include, but are not limited to, Jieba segmentation, Word2vec models, and LR models.

[0184] The application layer may include, but is not limited to, operations related to mobile payments based on target applications (such as WeChat Business Payments, but not limited to), rebate fraud, rebate feedback (also known as business protection), and automatic penalties.

[0185] The method provided in this application embodiment achieves an accuracy of up to 90% in identifying abnormal objects in an abnormal state within the target business. For example, taking objects exhibiting fraudulent operations as abnormal objects, the method provided in this application embodiment achieves an accuracy of up to 90% in identifying abnormal objects exhibiting fraudulent operations within the target business. When using the method provided in this application embodiment to identify whether merchants are engaging in fraudulent operations in rebate programs, it can identify 90% of abnormal merchants exhibiting fraudulent operations in rebate programs. Furthermore, after obtaining rebate feedback data from the merchants to be identified, abnormal merchants can be identified within seconds, and automatic penalties can be imposed on the identified abnormal merchants, such as freezing their funds or closing their payment permissions, achieving timely loss mitigation. The accuracy rate of automatically penalizing abnormal merchants is as high as 99%. Compared to previous processing methods, this significantly improves the efficiency and timeliness of identifying abnormal objects exhibiting fraudulent operations in the target business, while reducing labor costs.

[0186] It should be noted that the feedback information, object attribute information and other user-related data involved in the embodiments of this application require user permission or consent when this application is applied to specific products or technologies using the above embodiments, and the collection, use and processing of related data must comply with the relevant laws and standards of the relevant countries and regions.

[0187] Please refer to Figure 11 Based on the same inventive concept, embodiments of this application provide an object state recognition device 1100, comprising:

[0188] The data acquisition unit 1101 is used to acquire various feedback data of the object to be identified in the target business; wherein, the feedback data is triggered by the business operation performed by the account using the target business for the object to be identified in the target business;

[0189] The first identification unit 1102 is used to identify whether each of the above feedback data contains target information, and to determine the identification result corresponding to each of the above feedback data. The target information represents the operation agreement information associated with the target business that the above business operation does not meet the operation agreement information associated with the target business.

[0190] The second identification unit 1103 is used to determine whether the object status of the object to be identified in the target service is abnormal based on the identification results corresponding to each of the above feedback data.

[0191] As one embodiment, the first identification unit 1102 is specifically used to perform the following operations for each of the above-mentioned feedback data:

[0192] Based on the contextual information of each word contained in one of the feedback data, extract the text features to be processed corresponding to the above feedback data.

[0193] Based on the aforementioned text features, the system identifies whether the aforementioned feedback data contains the aforementioned target information, and determines the identification result corresponding to the aforementioned feedback data.

[0194] As one embodiment, the first identification unit 1102 is specifically used to: input the aforementioned feedback data into a trained data classification model; based on the language learning sub-model in the aforementioned data classification model, extract features from the context information of each word contained in the aforementioned feedback data to obtain the text features to be processed corresponding to the aforementioned feedback data; wherein, the aforementioned language learning sub-model is trained using historical feedback data in the aforementioned target business as training samples, based on the context information of each word contained in the aforementioned training samples;

[0195] The second identification unit 1103 is specifically used to: predict the second correlation between the text features to be processed and the first identification information based on the first correlation learned by the data prediction sub-model in the data classification model; determine the identification result corresponding to the feedback data based on the second correlation; the first identification information represents that the feedback data contains the target information; and the first correlation is determined based on the degree of correlation between the historical text features corresponding to the historical feedback data and the first identification information.

[0196] As one embodiment, the first identification unit 1102 is specifically used to: if the second correlation degree is greater than the first preset threshold, determine that the identification result corresponding to the feedback data is that the feedback data contains the target information; if the second correlation degree is not greater than the first preset threshold, determine that the identification result corresponding to the feedback data does not contain the target information.

[0197] As one embodiment, the second identification unit 1103 is specifically used for:

[0198] Based on the first identification information in the identification results corresponding to each of the above feedback data, the valid feedback information corresponding to the above object to be identified is determined. The first identification information is used to characterize that the feedback data contains the above target information.

[0199] Obtain the object attribute information associated with the above-mentioned object to be identified;

[0200] Based on the above-mentioned valid feedback information and the above-mentioned object attribute information, it is determined whether the object status of the object to be identified in the above-mentioned target business is abnormal.

[0201] As one embodiment, the above-mentioned valid feedback information includes one or any combination of the following: the number of first identification information in the identification results corresponding to each of the above-mentioned feedback data; the ratio of the number of the first identification information to the total number of the above-mentioned feedback data in the identification results corresponding to each of the above-mentioned feedback data.

[0202] As one embodiment, the second identification unit 1103 is specifically used to: process the above-mentioned feedback valid information and the above-mentioned object attribute information based on the above-mentioned target scoring card model to obtain the target operation evaluation value of the object to be identified for the above-mentioned target business; and determine whether the object status of the object to be identified in the above-mentioned target business is abnormal based on the above-mentioned target operation evaluation value.

[0203] As one embodiment, the second identification unit 1103 is further configured to: determine the object type of the object to be identified before processing the above-mentioned feedback valid information and the above-mentioned object attribute information based on the above-mentioned target scorecard model to obtain the target operation evaluation value of the object to be identified for the above-mentioned target business; and obtain the target scorecard model corresponding to the above-mentioned object type from the target scorecard model set associated with the above-mentioned target business.

[0204] As one embodiment, the second identification unit 1103 is specifically used for:

[0205] If the target operation evaluation value is determined to be lower than the evaluation value threshold, then the object to be identified is determined to be an abnormal object; wherein, the abnormal object includes objects whose object state is abnormal in the target business, and the operation evaluation value is negatively correlated with the degree of suspicion that the object's object state is abnormal in the target business; or

[0206] Based on the operation evaluation value range mapped to at least one object anomaly level, the operation evaluation value range to which the target operation evaluation value belongs is determined, and the object anomaly level mapped to the determined operation evaluation value range is determined as the object anomaly level corresponding to the object to be identified; wherein, the object anomaly level is determined based on the degree of suspicion that the object's state in the target business is an anomaly object.

[0207] As one embodiment, the second identification unit 1103 is specifically used to: input the above-mentioned feedback valid information and the above-mentioned object attribute information into the trained state recognition model, and perform the following processing through the above-mentioned state recognition model: determine the above-mentioned feedback valid information and the above-mentioned object attribute information as the features of the object to be processed corresponding to the object to be identified; predict the fourth correlation degree between the features of the object to be processed and the target object identification result based on the third correlation degree learned by the above-mentioned state recognition model; if the fourth correlation degree is greater than the second set threshold, determine that the object to be identified is an abnormal object, wherein the third correlation degree is determined based on the correlation degree between the historical object features corresponding to the historical object and the target identification result, the abnormal object includes objects whose object state is abnormal in the target business, and the target object identification result is used to characterize that the object's object state is abnormal in the target business.

[0208] As one embodiment, the second identification unit 1103 is further configured to: after determining that the object to be identified is an abnormal object, cancel at least one of the first permission and the second permission of the object to be identified; the first permission includes the permission to perform electronic resource transfer operations, and the second permission includes electronic resource transfer operations for a target electronic resource value, the target electronic resource value being determined based on the feedback data.

[0209] As one embodiment, the above-mentioned object attribute information includes at least one of the following:

[0210] Transfer information of electronic resources of the object to be identified; fraud anomaly information corresponding to the object to be identified, wherein the fraud anomaly information indicates the degree of suspicion that the object to be identified is involved in network fraud; operation information associated with abnormal business operations, wherein the abnormal business operations include business operations associated with feedback data containing the above target information.

[0211] As one embodiment, the object attribute information mentioned above includes operation information associated with the abnormal business operation. The operation information includes at least one of the following: the electronic resource value associated with the abnormal business operation; the account associated with the abnormal business operation; the business object associated with the abnormal business operation; and the trigger time information of the abnormal business operation.

[0212] As one example, Figure 11 The device described above can be used to implement any of the object state recognition methods discussed earlier.

[0213] Based on the same inventive concept as the above-described method embodiments, this application also provides a computer device. This computer device can be used for data processing based on pushed content. In one embodiment, the computer device can be a server, such as... Figure 1The server 120 is shown. In this embodiment, the structure of the computer device can be as follows: Figure 12 As shown, it includes a memory 1201, a communication module 1203, and one or more processors 1202.

[0214] The memory 1201 is used to store computer programs executed by the processor 1202. The memory 1201 may mainly include a program storage area and a data storage area. The program storage area may store the operating system and programs required to run instant messaging functions, etc.; the data storage area may store various instant messaging information and operation instruction sets, etc.

[0215] Memory 1201 may be volatile memory, such as random-access memory (RAM); memory 1201 may also be non-volatile memory, such as read-only memory, flash memory, hard disk drive (HDD), or solid-state drive (SSD); or memory 1201 may be any other medium capable of carrying or storing desired program code having the form of instructions or data structures and accessible by a computer, but is not limited thereto. Memory 1201 may be a combination of the above-described memories.

[0216] Processor 1202 may include one or more central processing units (CPUs) or digital processing units, etc. Processor 1202 is used to implement the above-mentioned method for extracting account features when calling a computer program stored in memory 1201.

[0217] The communication module 1203 is used to communicate with terminal devices and other servers.

[0218] This application embodiment does not limit the specific connection medium between the memory 1201, communication module 1203, and processor 1202. This application embodiment... Figure 12 The memory 1201 and the processor 1202 are connected via a bus 1204, and the bus 1204 is in Figure 12 The connections between other components are shown in thick lines only and are not intended to be limiting. The 1204 bus can be divided into address bus, data bus, control bus, etc. For ease of illustration, Figure 12 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0219] The memory 1201 stores a computer storage medium, which stores computer-executable instructions for implementing the account feature extraction method of this application embodiment. The processor 1202 is used to execute the above-described account feature extraction method, such as... Figure 3 As shown.

[0220] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0221] Alternatively, if the integrated unit described above is implemented as a software functional module and sold or used as an independent product, it can also be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the embodiments of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as mobile storage devices, ROM, RAM, magnetic disks, or optical disks.

[0222] Based on the same technical concept, embodiments of this application also provide a computer-readable storage medium storing computer instructions that, when executed on a computer, cause the computer to perform the startup method for an instant messaging application as described above.

[0223] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0224] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A method for object state recognition, characterized in that, include: Obtain various feedback data of the object to be identified in the target service; wherein, the feedback data is triggered by the business operations performed by the object to be identified in the target service using the account of the target service; Each feedback data is identified as containing target information, and the identification result corresponding to each feedback data is determined. The target information represents the operation that does not meet the operation convention information associated with the target business. Based on the first identification information in the identification results corresponding to each of the feedback data, the valid feedback information corresponding to the object to be identified is determined, and the first identification information is used to characterize that the feedback data contains the target information. Based on the valid feedback information and the object attribute information associated with the object to be identified, it is determined whether the object status of the object to be identified in the target service is abnormal. The step of determining whether the object status of the object to be identified in the target service is abnormal based on the valid feedback information and the object attribute information associated with the object to be identified includes: The valid feedback information and the object attribute information are input into the target scoring card model to obtain the target operation evaluation value of the object to be identified for the target service, and based on the target operation evaluation value, it is determined whether the object status of the object to be identified in the target service is abnormal; or The valid feedback information and the object attribute information are input into the trained state recognition model, and the following steps are performed: the valid feedback information and the object attribute information are determined as the features of the object to be identified, the fourth correlation degree between the features of the object to be identified and the target object recognition result is predicted, and based on the fourth correlation degree, it is determined whether the object state of the object to be identified in the target business is abnormal.

2. The method as described in claim 1, characterized in that, The step of identifying whether each piece of feedback data contains target information and determining the identification result corresponding to each piece of feedback data includes: For each of the feedback data, perform the following operations respectively: Based on the context information of each word contained in one of the feedback data, extract the text features to be processed corresponding to the feedback data; Based on the features of the text to be processed, it is determined whether the feedback data contains the target information, and the recognition result corresponding to the feedback data is determined.

3. The method as described in claim 2, characterized in that, The step of extracting the text features corresponding to a given feedback data point based on the context information of each word contained in that feedback data point includes: Input the feedback data into the trained data classification model; Based on the language learning sub-model in the data classification model, feature extraction is performed on the contextual information of each word contained in the feedback data to obtain the text features to be processed corresponding to the feedback data; wherein, the language learning sub-model is trained by using historical feedback data in the target business as training samples and based on the contextual information of each word contained in the training samples. The step of identifying whether a feedback data contains the target information based on the features of the text to be processed, and determining the identification result corresponding to the feedback data, includes: Based on the first correlation degree learned by the data prediction sub-model in the data classification model, a second correlation degree is predicted between the text features to be processed and the first identification information. Based on the second correlation degree, the identification result corresponding to the feedback data is determined. The first identification information represents that the feedback data contains the target information. The first correlation degree is determined based on the degree of correlation between the historical text features corresponding to the historical feedback data and the first identification information.

4. The method as described in claim 3, characterized in that, Determining the recognition result corresponding to the feedback data based on the second correlation includes: If the second correlation degree is greater than the first set threshold, then the identification result corresponding to the feedback data is determined to be that the feedback data contains the target information; If the second correlation degree is not greater than the first set threshold, then the identification result corresponding to the feedback data is determined to be that the feedback data does not contain the target information.

5. The method as described in claim 1, characterized in that, The valid feedback information includes one or any combination of the following: The number of first identification information in the identification results corresponding to each of the feedback data; The ratio of the number of the first identification information to the total number of all feedback data in the identification results corresponding to each of the feedback data.

6. The method as described in claim 1, characterized in that, Before inputting the valid feedback information and the object attribute information into the target scoring card model to obtain the target operation evaluation value of the object to be identified for the target business, the process further includes: Determine the object type of the object to be identified; Obtain the target scorecard model corresponding to the object type from the target scorecard model set associated with the target business.

7. The method as described in claim 1, characterized in that, The step of determining whether the object status of the object to be identified in the target service is abnormal based on the target operation evaluation value includes: If the target operation evaluation value is determined to be lower than the evaluation value threshold, then the object to be identified is determined to be an abnormal object; wherein, the abnormal object includes objects whose object state is abnormal in the target service, and the operation evaluation value is negatively correlated with the degree of suspicion that the object is in an abnormal state in the target service; or Based on the operation evaluation value range mapped to at least one object anomaly level, the operation evaluation value range to which the target operation evaluation value belongs is determined, and the object anomaly level mapped to the determined operation evaluation value range is determined as the object anomaly level corresponding to the object to be identified; wherein, the object anomaly level is determined based on the degree of suspicion that the object's object status in the target business is an abnormal state.

8. The method as described in claim 1, characterized in that, The fourth correlation degree between the predicted features of the object to be processed and the target object recognition result includes: Based on the third correlation degree already learned by the state recognition model, a fourth correlation degree is predicted between the features of the object to be processed and the target object recognition result, wherein the third correlation degree is determined based on the degree of correlation between the historical object features corresponding to the historical object and the target recognition result.

9. The method as described in claim 1, characterized in that, The step of determining whether the object status of the object to be identified in the target service is abnormal based on the fourth correlation degree includes: If the fourth correlation degree is greater than the second set threshold, then the object to be identified is determined to be an abnormal object. The abnormal object includes objects whose object state is abnormal in the target service. The target object identification result is used to characterize that the object's object state is abnormal in the target service.

10. The method as described in claim 7 or 9, characterized in that, After determining that the object to be identified is an abnormal object, the process further includes: Cancel at least one of the first and second permissions of the object to be identified; the first permission includes the permission to perform electronic resource transfer operations, and the second permission includes electronic resource transfer operations for a target electronic resource value, the target electronic resource value being determined based on the feedback data.

11. An object state recognition device, characterized in that, include: The data acquisition unit is used to acquire various feedback data of the object to be identified in the target business; wherein, the feedback data is triggered by the business operation performed by the object to be identified in the target business using the account of the target business; The first identification unit is used to identify whether each feedback data contains target information, and to determine the identification result corresponding to each feedback data. The target information represents that the business operation does not meet the operation agreement information associated with the target business. The second identification unit is used to determine the valid feedback information corresponding to the object to be identified based on the first identification information in the identification results corresponding to each of the feedback data, and to determine whether the object status of the object to be identified in the target service is abnormal based on the valid feedback information and the object attribute information associated with the object to be identified; the first identification information is used to characterize that the feedback data contains the target information; The second identification unit is specifically used for: The valid feedback information and the object attribute information are input into the target scoring card model to obtain the target operation evaluation value of the object to be identified for the target service, and based on the target operation evaluation value, it is determined whether the object status of the object to be identified in the target service is abnormal; or The valid feedback information and the object attribute information are input into the trained state recognition model, and the following steps are performed: the valid feedback information and the object attribute information are determined as the features of the object to be identified, the fourth correlation degree between the features of the object to be identified and the target object recognition result is predicted, and based on the fourth correlation degree, it is determined whether the object state of the object to be identified in the target business is abnormal.

12. A computer program product, characterized in that, The computer program product includes computer instructions stored in a computer-readable storage medium, a processor of a computer device reading the computer instructions from the computer-readable storage medium, and the processor executing the computer instructions to cause the computer device to perform the method of any one of claims 1-10.

13. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the method according to any one of claims 1-10.

14. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that, when executed on a computer, cause the computer to perform the method as described in any one of claims 1-10.

Images