Object recognition method, device, electronic equipment and computer readable storage medium

Abstract

Embodiments of the present application provide an object identification method and device, electronic equipment and computer readable storage medium, and relate to the fields of big data, artificial intelligence and cloud technology. The method comprises: obtaining a training set and a target classification model trained based on the training set; the training set comprising a plurality of first samples of sample objects of a target type; obtaining related data of a to-be-processed object in a target application, the related data comprising object information and business transaction data; if the correlation degree of the to-be-processed object and the objects in the first samples is greater than a set value, determining the to-be-processed object as a candidate target object; determining a classification result based on the related data of the candidate target object through the target classification model; and determining the candidate target object with the classification result being the target type as a final target object. The embodiments of the present application simultaneously consider the object information and the business transaction data, adopt a double screening mechanism of correlation degree comparison and model screening, and can accurately determine the target object in the target application.

Description

Technical Field

[0001] This application relates to the fields of big data, artificial intelligence and cloud technology. Specifically, this application relates to an object recognition method, apparatus, electronic device, computer-readable storage medium and computer program product. Background Technology

[0002] With the development of big data and network technology, the reliance on big data and network technology to assist production and research in various industries is becoming increasingly mature.

[0003] In related technologies, when it is necessary to identify merchants within a preset industry, the main methods are either to check if a merchant's name contains keywords related to the preset industry, or to collect and analyze evaluation information related to the merchant to determine whether the merchant belongs to that industry. However, the former relies on the registration information filled in by the merchant themselves, which is greatly influenced by the merchant's subjective factors, while the latter relies on the subjective factors of the evaluators. Therefore, neither method is objective enough, has low reliability, and cannot accurately identify merchants within the preset industry, thus failing to meet business requirements. Summary of the Invention

[0004] This application provides a method, apparatus, electronic device, computer-readable storage medium, and computer program product for object identification, used to more accurately determine target objects in a target application.

[0005] According to one aspect of the embodiments of this application, an object recognition method is provided, the method comprising:

[0006] Obtain the training set and the target classification model trained based on the training set;

[0007] The training set includes multiple training samples. Each training sample includes relevant data for a sample object. The relevant data includes object information and business transaction data. The sample object is an object in the target application. The business transaction data includes business transaction data corresponding to the target application. The multiple training samples include multiple first samples. The sample object corresponding to each first sample is a first sample object of the target type.

[0008] Obtain relevant data for at least one object to be processed in the target application;

[0009] For each object to be processed, if the correlation between the object to be processed and any first sample object is greater than a set value, the object to be processed is determined as a candidate target object. The correlation between any two objects is determined based on the relevant data of these two objects.

[0010] For each candidate target object, based on the relevant data of that candidate target object, the classification result of the candidate target object is obtained through the target classification model;

[0011] The object that is classified as the target type among the candidate target objects is determined as the final target object.

[0012] According to another aspect of the embodiments of this application, an object recognition device is provided, the device comprising:

[0013] The first acquisition module is used to acquire the training set and the target classification model trained based on the training set.

[0014] The training set includes multiple training samples. Each training sample includes relevant data for a sample object. The relevant data includes object information and business transaction data. The sample object is an object in the target application. The business transaction data includes business transaction data corresponding to the target application. The multiple training samples include multiple first samples. The sample object corresponding to each first sample is a first sample object of the target type.

[0015] The second acquisition module is used to acquire relevant data of at least one object to be processed in the target application;

[0016] The candidate target object determination module is used to determine the object to be processed as a candidate target object if the correlation between the object to be processed and any first sample object is greater than a set value. The correlation between any two objects is determined based on the relevant data of any two objects.

[0017] The classification module is used to obtain the classification result of each candidate target object based on the relevant data of the candidate target object through the target classification model;

[0018] The target object determination module is used to identify the objects that are classified as target types from among the candidate target objects as the final target objects.

[0019] According to another aspect of the present application, an electronic device is provided, the electronic device including a memory, a processor and a computer program stored in the memory, the processor executing the computer program to implement the steps of the above method.

[0020] According to another aspect of the embodiments of this application, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the steps of the above-described method.

[0021] According to another aspect of the embodiments of this application, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps of the above-described method.

[0022] The beneficial effects of the technical solutions provided in this application are:

[0023] Based on the object recognition method provided in this application, a target classification model is trained by training on relevant data including object information of sample objects and business transaction data. This method can simultaneously consider object information and business transaction data, so that the obtained target classification model can fully combine the characteristics of multiple types of data (i.e., multi-source data), thereby achieving a more accurate determination of the classification result of an object based on the relevant data of a certain object.

[0024] When identifying the target object in at least one object to be processed in a target application, candidate target objects can be initially screened by the degree of association between the object to be processed and the first sample object of the target type. Then, based on the relevant data of the candidate target objects, the candidate target objects are classified and identified by a target classification model trained on multi-source data of the sample objects, thereby accurately identifying the final target object among the candidate target objects. That is, by adopting a dual screening mechanism of association degree comparison and model screening, the screening efficiency and accuracy are improved, thus more accurately determining the target object in the target application. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments of this application will be briefly introduced below.

[0026] Figure 1 A schematic diagram of an object recognition system provided in an embodiment of this application is shown;

[0027] Figure 2 Another schematic diagram of the object recognition system provided in this application embodiment is shown;

[0028] Figure 3 The diagram illustrates the signaling interaction of an object recognition method in an application scenario applicable to the embodiments of this application.

[0029] Figure 4 This diagram illustrates the structure of the neural network model provided in an embodiment of this application.

[0030] Figure 5 A flowchart illustrating the object recognition method provided in an embodiment of this application is shown;

[0031] Figure 6 A schematic diagram of an object network provided in an embodiment of this application is shown;

[0032] Figure 7 Another flowchart of an object recognition method applicable to an embodiment of this application is shown;

[0033] Figure 8 A schematic diagram of an object recognition device applicable to an embodiment of this application is shown;

[0034] Figure 9 A schematic diagram of the structure of an electronic device to which this application embodiment applies is shown. Detailed Implementation

[0035] The embodiments of this application are described below with reference to the accompanying drawings. It should be understood that the embodiments described below with reference to the accompanying drawings are exemplary descriptions for explaining the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions of the embodiments of this application.

[0036] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the terms “comprising” and “including” as used in embodiments of this application mean that the corresponding feature can be implemented as the presented feature, information, data, step, operation, element, and / or component, but do not exclude implementation as other features, information, data, step, operation, element, component, and / or combinations thereof supported by the art. It should be understood that when we say that an element is “connected” or “coupled” to another element, the one element can be directly connected or coupled to the other element, or it can mean that the one element and the other element establish a connection relationship through an intermediate element. Furthermore, “connected” or “coupled” as used herein can include wireless connection or wireless coupling. The term “and / or” as used herein indicates at least one of the items defined by the term; for example, “A and / or B” can be implemented as “A,” or as “B,” or as “A and B.”

[0037] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0038] The object recognition method provided in this application embodiment can be applied to any application scenario that requires target object recognition, including but not limited to the scenario of determining merchants in a preset industry as mentioned above. In this scenario, the target type object can be a merchant in a specified / specific industry (e.g., a specific high-risk industry, such as the risk control industry). This application embodiment does not limit the method of dividing the target type object into non-target type objects, and it can be configured according to actual needs.

[0039] Referring to the aforementioned related technologies, the methods for identifying merchants in preset industries need to be improved in terms of accuracy. In particular, for some specific sensitive industries, in order to meet the audit requirements, merchants often hide their actual business in their names, which increases the difficulty of identifying merchants in that industry and makes it impossible to accurately identify merchants in specific industries.

[0040] To address at least one of the aforementioned technical problems or areas requiring improvement in related technologies, embodiments of this application provide an object identification method, apparatus, electronic device, computer-readable storage medium, and computer program product. This method combines object information and business transaction data from multiple objects to train a target classification model. By determining the degree of association between the object to be processed and any pre-determined object belonging to a specific type, and based on the object information and business transaction data of the object to be processed that meet the association degree criteria, the target classification model determines the type of the object to be processed. This employs a dual screening mechanism of association degree comparison and model-based filtering, improving screening efficiency and accuracy, thereby more accurately identifying target objects belonging to the target type in the target application.

[0041] The technical solutions of this application and their effects are described below through several exemplary embodiments. It should be noted that the following embodiments can be referenced, borrowed from, or combined with each other. Identical terms, similar features, and similar implementation steps in different embodiments will not be repeated.

[0042] Figure 1 A schematic diagram of an object recognition system provided in an embodiment of this application is shown, such as... Figure 1 As shown, the system 10 includes a user terminal 101, an application server 102, and a model training server 103. The user terminal 101 and the application server 102, as well as the application server 102 and the model training server 103, can be directly or indirectly connected via wired or wireless communication. This application embodiment does not impose any restrictions on this.

[0043] User terminal 101 can be a terminal for conducting business transactions. Specifically, this terminal can have any application (i.e., the target application) that can be used for conducting business transactions installed on it, or a plugin for conducting business transactions can be set in the target application. By opening the target application or triggering the aforementioned plugin for conducting business transactions, business transactions can be conducted, and the business transaction data can be sent to application server 102.

[0044] The terminal 101 (also referred to as a user terminal or user equipment (UE)) can be a smartphone, tablet, laptop, desktop computer, wearable electronic device (such as a smartwatch), in-vehicle terminal, smart home appliance (such as a smart TV), AR (Augmented Reality) / VR (Virtual Reality) device, etc., but is not limited to these.

[0045] The model training server 103 can be used to train an initial neural network model based on a training set to obtain a trained target classification model. The trained target classification model is then sent to the application server 102 so that the application server 102 can deploy the trained target classification model.

[0046] Application server 102 can be used to obtain object information and business transaction data corresponding to the objects to be processed from user terminal 101, and based on the obtained object information and business transaction data corresponding to the objects to be processed, determine the classification result of the business objects through the deployed trained target classification model, and determine the list of target objects corresponding to user terminal 101 based on the classification result of each object to be processed.

[0047] If it is necessary to obtain object information from external sources, such as through a server other than the application server, the object recognition system may also include another application server, for example, evaluation information of the object to be processed contained in that other application server. Figure 2 As shown, in Figure 2 The object recognition system 20 shown includes a user terminal 201, a first application server 202, a second application server 203, and a model training server 204 corresponding to the user terminal. The user terminal 201 and the first application server 202, the first application server 202 and the second application server 203, and the first application server 202 and the model training server 204 can be directly or indirectly connected via wired or wireless communication. This embodiment does not limit this connection. The first application server 202 can obtain object information and business transaction data of the object to be processed from the user terminal 201, and obtain other object information of the object to be processed from the second application server 203. The object information of the object to be processed obtained from the user terminal 201 and the other object information obtained from the second application server 203 are combined as the object information of the object to be processed. The specific functions of the model training server 204 are the same as those of the model training server 103 described above, and the method by which the first application server 202 determines the list of target objects is the same as the method by which the application server 102 determines the list of target objects described above.

[0048] The object recognition method implemented by the object recognition system provided in this application can be based on artificial intelligence (AI) technology. For example, an initial neural network model is constructed based on AI technology, and the initial neural network model is trained on a training set to obtain a target classification model. Then, based on the relevant data of candidate target objects, the classification result of the candidate target objects is determined through the target classification model. AI refers to the theories, methods, technologies, and application systems that utilize digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use knowledge to obtain optimal results.

[0049] The aforementioned target classification model can also be a neural network model based on computer vision (CV) technology. Computer vision is a science that studies how to enable machines to "see." More specifically, it refers to machine vision, which uses cameras and computers to replace human eyes to identify and measure targets, and then performs further image processing to make the computer-processed images more suitable for human observation or transmission to instruments for detection.

[0050] The object recognition method provided in this application embodiment can also be implemented based on cloud technology. For example, the data computation involved in training the target classification model can be carried out using cloud computing. Cloud technology refers to a hosting technology that unifies a series of resources such as hardware, software, and networks within a wide area network or local area network to realize data computation, storage, processing, and sharing.

[0051] The face recognition method provided in this application embodiment can also be implemented in the field of intelligent transportation and applied to autonomous driving or transportation applications. For example, the object recognition method can be applied to the Electronic Toll Collection (ETC) system to identify vehicles that have missed multiple payments.

[0052] The object recognition method provided in this application embodiment can also be implemented based on blockchain technology. Specifically, the data used in the object recognition process, such as training datasets and images to be processed, can be stored on the blockchain.

[0053] It is understood that, in the specific implementation of this application, when the above embodiments of this application are applied to specific products or technologies, the relevant data such as object information and business transaction data involved need to obtain user permission or consent, and the collection, use and processing of the relevant data need to comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0054] The following is combined Figure 1The object recognition system shown uses a shopping scenario as an example to illustrate the flow of the object recognition method provided in this application embodiment. In this shopping scenario, the target application is a shopping application. Multiple merchants (i.e., sellers) register in this shopping application to provide consumers (i.e., buyers) with various product purchase services. The object recognition method provided in this application embodiment determines a list of target merchants belonging to the investment and wealth management category and a list of non-target merchants not belonging to the investment and wealth management category among the multiple merchants in the target application. In this example, preset keywords related to investment and wealth management include "finance," "wealth," and "investment"; preset product types include "funds," "gold," "bonds," and "insurance," etc.

[0055] Figure 3 The diagram illustrates the signaling interaction of an object recognition method in an application scenario applicable to the embodiments of this application. For example... Figure 3 As shown, the method may include the following steps S31 to S37.

[0056] Step S31: The application server obtains the initial merchant's relevant data from the user terminal.

[0057] For each merchant, when the merchant registers through a shopping application, the application server obtains the merchant's full name, abbreviation, and the names of the products offered by the merchant from the merchant's registration information. It also obtains relevant evaluation information (including positive and negative reviews, i.e., complaints) from the application and other applications, as well as information related to the merchant from other social media platforms (such as WeChat official accounts, Weibo posts, etc.). This information related to the merchant is then defined as the merchant's object information.

[0058] The application server uses the current time as the endpoint and traces back 30 days to retrieve transaction records generated through the shopping application within those 30 days. Each transaction record represents a single transaction data point. The server identifies the object identifier within each transaction data point and, based on the initial merchant corresponding to each object identifier, determines the business transaction data for that initial merchant. For example, a single transaction data point for a merchant represents the data generated when a consumer purchases any item offered by that merchant through the shopping application. Each transaction data point can include the consumer's object identifier, the merchant's object identifier, and the item identifier.

[0059] The business transaction data and object information corresponding to each initial merchant are used as the relevant data for that initial merchant.

[0060] Step S32: The application server processes the relevant data of each initial merchant to obtain the first dataset with tags.

[0061] First screening: The application server identifies initial merchants whose merchant names and abbreviations contain any of the keywords "finance," "wealth," and "investment" as seed merchants.

[0062] The first screening results and the object information of each initial merchant are sent to the user terminal so that the user corresponding to the user terminal can add the label "this merchant is an investment and wealth management merchant" to the merchants identified as seed merchants in the first screening results.

[0063] Second screening: For each initial merchant not determined after the first screening, the user on the corresponding user terminal determines the type of goods offered by the initial merchant based on the merchant's product name, reviews, and information obtained through other social media platforms related to the merchant. If the type of goods offered by the initial merchant includes one of "funds," "gold," "bonds," and "insurance," the initial merchant is designated as a seed merchant; otherwise, it is designated as a non-seed merchant. Based on the determination results, merchants designated as seed merchants are labeled "This merchant is an investment and wealth management merchant," and merchants designated as non-seed merchants are labeled "This merchant is not an investment and wealth management merchant."

[0064] The application server obtains the first dataset with tags based on the relevant data of each initial merchant with tags.

[0065] Step S33: The application server builds the initial merchant network.

[0066] After obtaining the initial labeled dataset, the application server traces back to the starting point of acquiring each initial merchant's data, using the current time as the endpoint, to obtain the business transaction data of each merchant within that time period. Based on the business transaction data of each merchant within that time period, it determines the number of consumers (i.e., transaction partners) corresponding to each merchant, as well as the number of common consumers between each pair of merchants. Based on the number of consumers corresponding to each pair of merchants and the number of common consumers, it determines the correlation coefficient between the two merchants.

[0067] Let merchant i be a merchant and merchant j be a merchant. Let C be the number of consumers corresponding to merchant i. i The number of consumers corresponding to merchant j is C. j The number of consumers shared by merchant i and merchant j is C. ij For example, the correlation coefficient α between the two ij for:

[0068]

[0069] Where, α ij The range of α is 0 to 1.ij The larger the value, the higher the degree of business overlap between merchant i and merchant j.

[0070] Treat each merchant as a node. If the correlation coefficient α > 0.01 and C > 5 between any two merchants, connect the nodes corresponding to these two merchants. Repeat this process to obtain the initial merchant network corresponding to each of the currently acquired merchants.

[0071] Based on the above method, the merchant network is updated continuously according to the business transaction data of merchants whose corresponding nodes do not exist in the current merchant network.

[0072] Step S34: The application server expands the first dataset with labels to obtain the training set.

[0073] According to the method in step S32, the application server determines the labels corresponding to each associated merchant based on the object information of the nodes connected to the labeled merchants in the initial merchant network and each labeled merchant in the first dataset. It then merges the labeled first dataset with the relevant data corresponding to each labeled associated merchant to obtain a training set, and sends the training set to the model training server. In this training set, the relevant data of a labeled merchant constitutes a training sample, and the corresponding merchant is the sample merchant.

[0074] Among them, sample merchants in the training set labeled "This merchant is an investment and wealth management merchant" can be added to the target merchant list corresponding to the shopping application.

[0075] Step S35: The model training server trains the initial neural network model based on the received training set to obtain the target classification model, and then sends the target classification model to the application server.

[0076] After receiving the training set, the model training server performs statistical analysis on the business transaction data corresponding to each training sample. This determines the indicator values ​​for each transaction evaluation metric (e.g., the unit price of each product offered by the sample merchant, repurchase rate, and the percentage of transaction amount per hour within 24 hours) for each sample merchant in the training sample. The indicator values ​​for each sample merchant are then concatenated in a predetermined order to form the business transaction feature vector for that sample merchant. This predetermined order can be, in sequence, the unit price of each product offered by the sample merchant, the repurchase rate, and the percentage of transaction amount per hour within 24 hours. Furthermore, the server converts the object information of different categories of the sample merchant (i.e., categorized according to the merchant's name (including full and abbreviated name), the product names of each product offered by the merchant, the merchant's review information, and public account information) into corresponding tokens (characters).

[0077] Figure 4 This diagram illustrates the structure of the neural network model provided in an embodiment of this application. Figure 4 As shown, the initial neural network model includes a BERT (bidirectional encoder representation from transformers) layer, a feature extraction layer, a Neural Network (NN) layer, a normalization layer, and a loss function layer. The feature extraction layer can be implemented using average pooling, and the normalization layer can be implemented using sigmoid (an activation function).

[0078] For each training sample, the BERT layer can be a set of BERT models corresponding to different types of object information. Different BERT models within the BERT layer can convert the corresponding tokens of the object information into corresponding word vectors. The CLS (Common Language Specification) flag of each word vector can be used to distinguish different types of object information; these word vectors will also be referred to as CLS vectors below. In this example, the word vectors corresponding to different types of object information are 768-dimensional CLS vectors. If there are N types of object information for each sample object, then N 768-dimensional CLS vectors can be obtained through the BERT layer. Average pooling can pool these 768-dimensional CLS vectors to obtain a 768-dimensional object feature vector. Since object information is usually in the form of text, it will be referred to as text information below, and correspondingly, the object feature vector can be referred to as the text feature vector. The neural network layer can concatenate the business transaction feature vector and the text feature vector of the sample merchant to obtain the target feature vector of the sample merchant. In this application, the business transaction feature vector can be concatenated before or after the text feature vector; the concatenation order is not restricted. The normalization layer determines the prediction result for each sample merchant based on its target feature vector. The loss layer calculates the training loss of the model based on the prediction results and corresponding labels of each training sample in the training set, using the cross-entropy loss function.

[0079] Training ends when the training loss is less than a set value, and the model with the loss less than the set value is used as the target classification model. If the training loss is greater than or equal to the set value, the model parameters of each layer in the neural network model are adjusted, and the adjusted neural network model is trained again based on the training set.

[0080] Step S36: The application server obtains relevant data of any merchant from the user terminal and determines the classification result of the merchant through the target classification model.

[0081] The target classification model can output the probability that a merchant is an investment and wealth management merchant. If the probability is greater than or equal to 0.75, the merchant is determined to be an investment and wealth management merchant; if the probability is less than 0.15, the merchant is determined not to be an investment and wealth management merchant. If the probability is equal to 0.15, or greater than 0.15 but less than 0.75, it is determined that the merchant cannot be classified according to the target classification model, and a merchant labeling method is used to determine whether the merchant is an investment and wealth management merchant based on relevant data.

[0082] The application server retrieves transaction data from any merchant within a specific time period from the user terminal. Based on this data, it performs statistical analysis to obtain the merchant's performance index value for each transaction evaluation metric. These index values ​​are then concatenated in a predetermined order to form the merchant's transaction feature vector. Finally, the server converts the merchant's different types of object information into corresponding tokens.

[0083] Input the merchant's business transaction feature vector and the tokens corresponding to the merchant's different types of object information into the target classification model to obtain the merchant's classification result, and add the merchants whose classification result is investment and financial management to the target merchant list corresponding to the shopping application.

[0084] Step S37: The application server obtains the merchant network corresponding to the current moment, identifies the unclassified merchants corresponding to each node in the merchant network that has an edge connected to the node corresponding to each merchant with a classification result of "investment and wealth management" (i.e., the first-order neighbor nodes of the nodes corresponding to each merchant with a classification result of "investment and wealth management"). Based on the relevant data of the identified merchants, the server determines the classification result of the identified merchants through the target classification model, and adds the merchants with a classification result of "investment and wealth management" to the target merchant list corresponding to the shopping application. This process continues until no new merchants appear, at which point the target merchant list for the shopping application is obtained. The determined target merchant list is then fed back to the user terminal.

[0085] Figure 5This diagram illustrates a flowchart of an object recognition method provided in an embodiment of this application. The object recognition method provided in this application can be executed by an object recognition device. This object recognition device can be any computer device, including but not limited to a terminal or server. The server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. The object recognition method can also be implemented by a processor calling computer-readable instructions stored in memory.

[0086] like Figure 5 As shown, the method includes:

[0087] Step S510: Obtain the training set and the target classification model trained based on the training set;

[0088] The training set includes multiple training samples. Each training sample includes relevant data for a sample object. The relevant data includes object information and business transaction data. The sample object is an object in the target application. The business transaction data includes business transaction data corresponding to the target application. The multiple training samples include multiple first samples. The sample object corresponding to each first sample is a first sample object of the target type.

[0089] The object information of a sample object includes basic text information related to that sample object, including but not limited to the basic text information of the sample object in the target application and the basic text information of the sample object in other applications (i.e., external search results). The basic text information includes, but is not limited to, the sample object's own identification information, the sample object's self-description information, and the evaluation information of the sample object. The sample object's own identification information can be composed of at least one of the following: text, letters, symbols, etc. The object information can be classified according to its function; that is, the object information including the sample object's own identification information, the sample object's self-description information, and the evaluation information of the sample object can be divided into three types: identification information, description information, and evaluation information, respectively.

[0090] Taking the merchant as an example, the object information of the sample object includes the merchant's main name, the merchant's abbreviation, the name of each product provided by the merchant, the merchant's self-introduction (i.e., description information), the evaluation information of the merchant, and information related to the merchant obtained through web pages, public accounts, etc.

[0091] A sample object's business transaction data includes at least one transaction record associated with that sample object. Each transaction record includes the identifier of the transaction object (including the identifiers of each of the two parties involved in the transaction), the object being traded, and the content incurred in generating the transaction. Taking an item as an example, the transaction data may include the identifiers of each of the two parties involved in the transaction, the identifier of the item, and the corresponding resource quantity of the item.

[0092] In this example, the multiple training samples may also include multiple second samples, each corresponding to a sample object of the target type. It should be noted that here, "first" and "second" are merely to distinguish different target types and do not have a specific order. This application embodiment does not limit the specific number of first samples and second samples among the multiple training samples; these can be determined according to actual circumstances. For example, the ratio of the number of first samples to the number of second samples can be set to 1:1.

[0093] In one possible implementation, obtaining the training set described above may include:

[0094] Obtain a first initial dataset and a keyword library corresponding to objects of the target type. The first initial dataset includes relevant data for multiple first initial objects.

[0095] Based on the object information of each first initial object, objects whose object information contains at least one keyword from the keyword library are used as samples of the selected target types.

[0096] The relevant data of each first initial object in the first initial dataset, excluding the selected target type sample objects, are provided to the annotators to obtain the labels corresponding to each first initial object, excluding the selected target type sample objects. The labels indicate whether the corresponding first initial object is a target type sample object or a non-target type sample object.

[0097] A training set is constructed based on the relevant data and corresponding labels of each initial object in the first initial dataset.

[0098] The target classification model trained on the training set mentioned above can include:

[0099] Multiple training samples from the training set are input into the initial neural network model, and the prediction results corresponding to the sample objects in the training samples are obtained through the initial classification model.

[0100] Based on the prediction results and labels corresponding to each sample object, the training loss of the neural network model is determined.

[0101] If the training loss meets the preset training termination condition, the neural network model that meets the training termination condition will be used as the target classification model.

[0102] If the training loss does not meet the training termination condition, the model parameters of the neural network model are adjusted based on the training loss.

[0103] In this implementation, the first initial dataset can be a collection of relevant data for each historical object in the target application. By processing the collection of relevant data for each historical object in the target application, a training set is obtained. Based on the target classification model trained on the training set, the classification results of candidate target objects are determined, which can improve the accuracy of the identification results of candidate target objects.

[0104] The keyword library can be determined based on the target type of the target object to be identified; that is, each target type corresponds to one keyword library. This can be determined according to the actual situation, and this application embodiment does not impose any restrictions on this. Furthermore, this application embodiment does not impose any restrictions on the number of keywords in each keyword library. The corresponding keyword library can be determined according to the identification requirements.

[0105] For example, if the target audience is in the live streaming industry (i.e., the target type is within the live streaming industry), the corresponding keywords in the keyword library could be "live streaming," "channel," "TV (television)," "Karaoke," and "show." Similarly, if the target audience is in the investment and wealth management industry (i.e., the target type is within the investment and wealth management industry), the corresponding keywords in the keyword library could be "finance," "wealth," and "investment."

[0106] After obtaining the keyword library and the object information of each initial object, if the filtering result of the initial object is a seed sample object, the label of the initial object can be set to "object belonging to the target type", that is, a positive sample.

[0107] Considering that the basic object information of some objects cannot fully reflect the feature information of the object, and that the method of filtering by keywords is relatively simple, if the object information of the first initial object does not include any keyword in the keyword library, the first initial object is directly identified as a non-seed sample object and the label of the initial object is set to "object that does not belong to the target type". That is, the negative sample method is not accurate enough. In this embodiment of the application, the relevant data of each first initial object in the first initial dataset other than the selected target type sample objects can be provided to the annotators so that the labels corresponding to each first initial object other than the selected target type sample objects can be specifically determined by the annotators through manual annotation.

[0108] Specifically, when the annotators determine the labels for each initial object in the first initial dataset, excluding the selected target type sample objects, they can also treat each initial object as a candidate for filtering. A deep analysis of the object information for each candidate object is then performed to determine the corresponding real information. Based on this real information and the business transaction data of each candidate object, seed sample objects and non-seed sample objects are identified. Finally, based on the filtering results obtained from this second filtering, the labels for each object in the first dataset are determined.

[0109] For example, when the target to be screened is a merchant, since the product information provided by the merchant may not match the product information actually used in business transactions, a deep analysis of the merchant's information can determine the true product type of each product offered by the merchant. Then, based on the true product type, a further screening is performed. If the true product type of each product offered by the merchant includes the target product type, and the business transaction volume of the merchant's products corresponding to the target product type is determined to be greater than a preset value, such as 5, based on the merchant's business transaction data, then the merchant is identified as a seed merchant. Otherwise, the merchant is identified as a non-seed merchant.

[0110] By using objects that are not seed samples in the keyword-based filtering results as the objects to be filtered, and then conducting in-depth analysis of the object information of each object to be filtered through manual filtering, seed samples and non-seed samples are identified from each object to be filtered based on the results of the in-depth analysis. Based on the filtering results obtained from the second filtering, the labels of each object in the first dataset are determined. This can reduce the number of annotations required by the annotators while more accurately determining the labels of each training sample in the training set, thus reducing the reliance on annotators.

[0111] In one possible implementation, the above method may further include:

[0112] Obtain the second initial dataset, which includes relevant data for multiple second initial objects;

[0113] For each second initial object, if the correlation between the second initial object and any target type sample object in the first initial dataset is greater than a set value, the second initial object is determined as a candidate sample object of the target type.

[0114] The relevant data of each candidate sample object is provided to the annotators so that they can verify the type of each candidate sample object and obtain the verified label of each candidate sample object.

[0115] The training set constructed above, based on the relevant data and corresponding labels of each initial object in the initial dataset, may include:

[0116] A training set is constructed based on the relevant data and corresponding labels of each initial object in the first initial dataset, and the relevant data and verified labels of each candidate sample object in the second initial dataset.

[0117] In this embodiment, the expanded sample objects can be determined using the above method. However, this embodiment does not restrict the conditions for stopping the operation of determining expanded sample objects, and can be determined according to the actual situation. For example, the conditions for stopping the operation of determining expanded sample objects can be set to stop the operation when the number of times expanded sample objects are determined reaches a preset number, or it can be set to stop the operation when the number of determined expanded sample objects reaches a preset number, or it can be set to stop the operation when no new expanded sample objects appear.

[0118] In this implementation, the degree of association between each second initial object in the second initial dataset and any target type sample object in the first initial dataset is determined. If the degree of association is greater than a set value, the second initial object is identified as a candidate sample object of the target type. Furthermore, the type of each candidate sample object is verified by annotators to obtain the verified true label for each candidate sample object. Based on the relevant data of each labeled candidate sample object and the relevant data of each labeled first initial object, a training set is constructed. This effectively expands the training set, allowing the neural network model to be trained using the expanded training set, resulting in a more accurate target classification model.

[0119] In this implementation, before inputting multiple training samples into the initial neural network model, the object information of different types of sample objects in each training sample can be converted into corresponding tokens. Statistical analysis is then performed on the business transaction data of the sample objects in the training samples to obtain the indicator values ​​for each transaction evaluation metric corresponding to the sample objects. The tokens containing the different types of object information for each sample object, along with the indicator values ​​for each transaction evaluation metric for each sample object, are then input into the neural network model. Figure 4 In the neural network model shown, the object feature vector and business transaction feature vector corresponding to the sample object in each training sample are obtained. Then, the object feature vector and business transaction feature vector corresponding to each sample object are concatenated to obtain the target feature vector of each sample object. Then, prediction is made based on the target feature vector to obtain the prediction result of each sample object.

[0120] In this implementation, the prediction result for each sample object can include the confidence level (i.e., probability value) that the sample object is an object of the target type. If the obtained confidence level is greater than a preset value, the prediction result for the sample object is determined to be that the sample object is an object of the target type; if the obtained confidence level is less than the preset value, the prediction result for the sample object is determined to be that the sample object is not an object of the target type.

[0121] In this implementation, the cross-entropy loss function can be used to determine the training loss of the neural network model until a target classification model that meets the training termination condition is obtained.

[0122] The preset training termination conditions can be configured according to needs, and may include, but are not limited to, loss function convergence, training loss being less than a set value, or the number of training iterations reaching a set number. The smaller this set value, the higher the accuracy of the obtained target classification model.

[0123] In this implementation, the training set can also be divided into a training dataset and an evaluation dataset according to a preset ratio. The preset ratio can be configured according to actual needs (e.g., it can be an empirical value or an experimental value), and this application does not impose any restrictions on it. For example, the preset ratio can be set to 5:1, that is, after the split, the ratio of the number of training samples in the training dataset to the number of training samples in the evaluation dataset is 5:1.

[0124] After obtaining the training and evaluation datasets, the initial neural network model can be trained using the training dataset according to the training method described above to obtain a first classification model. This first classification model is then evaluated using the evaluation dataset. If the evaluation result meets the preset evaluation criteria, the first classification model is determined as the target classification model. If the evaluation result does not meet the preset evaluation criteria, the first classification model is trained again using the training dataset until a target classification model that meets the preset evaluation criteria is obtained. This application does not impose restrictions on the evaluation metrics; they can be determined according to the actual situation.

[0125] Taking accuracy as the evaluation metric, the preset evaluation condition can be that the accuracy of the first classification model is greater than or equal to the accuracy threshold. The accuracy threshold ACC can be any value between 0.9 and 1; this embodiment does not impose any limitation on this. The higher the accuracy threshold, the higher the accuracy of the final target classification model.

[0126] Of course, it's also possible to determine the specific classification result value based on the probability output by the model, using the evaluation dataset as a specific basis. Specifically, assuming the set of training samples in the evaluation dataset is X, the probability value of the predicted result of a training sample in the evaluation dataset obtained through the first classification model is p, and the label of that training sample is y, assuming that based on the training samples in the evaluation dataset, the classification result is determined... and ACC1 can be the same as or different from ACC2, depending on the specific circumstances. Then, the condition that satisfies... and t0 is determined as the probability threshold that the sample object does not belong to the target type, and t1 is determined as the probability threshold that the sample object belongs to the target industry. In specific implementation, if the probability p output by the model is less than t0, the current object is determined to not belong to the target type; if the probability p output by the model is greater than t1, the current object is determined to belong to the target type.

[0127] By using the methods described above, the target classification model can achieve better accuracy, stronger generalization ability, and higher coverage and precision.

[0128] Step S520: Obtain relevant data for at least one object to be processed in the target application.

[0129] Step S530: For each object to be processed, if the correlation between the object to be processed and any first sample object is greater than a set value, the object to be processed is determined as a candidate target object. The correlation between any two objects is determined based on the relevant data of the above two objects.

[0130] In this implementation, the degree of association between any two objects can be an indicator value that represents their association, determined based on the relevant data of the two objects. This application embodiment does not limit the specific method for determining the degree of association. For example, the degree of association between the two objects can be characterized by the number of common trading objects, or by a correlation coefficient determined by the number of trading objects for each of the two objects and the number of common trading objects, or by the number of shared items in the trading goods provided by each of the two objects.

[0131] Accordingly, the corresponding setting value can be determined according to the method of determining the degree of association, and this application embodiment does not limit this.

[0132] In this implementation, if the correlation between the object to be processed and the first sample object of any target type is less than a set value, the relevant data of the object to be processed may not be processed.

[0133] In one possible implementation, for any object, the business transaction data includes relevant information about the corresponding transaction object, including the object identifier; the degree of association between the aforementioned object to be processed and any first sample object is determined in the following way:

[0134] Based on the object identifiers of the transaction objects corresponding to the object to be processed and any first sample object, determine the first number of common transaction objects corresponding to the object to be processed and the first sample object;

[0135] Based on the first quantity, determine the degree of association between the object to be processed and the first sample object.

[0136] For any two objects, the object identifier of each transaction object can be determined based on the business transaction data corresponding to each of the two objects. Then, based on the object identifier of each transaction object in the two objects, the object identifier of the common transaction object can be determined, and the number of object identifiers of the common transaction object can be determined as the first quantity.

[0137] As described above, the business transaction data of an object includes multiple transaction data entries for that object. When determining the object identifier of a transaction object corresponding to an object based on its business transaction data, if at least two of the multiple transaction data entries for that object contain the same object identifier, the duplicate object identifiers can be directly used to repeatedly determine the object identifiers of the same transaction objects, resulting in multiple identical object identifiers. Alternatively, the duplicate object identifiers can be directly used as the object identifier of a single transaction object. This application does not limit this approach, but the former method is more accurate.

[0138] In this implementation, the larger the number of common transaction objects corresponding to any two objects, the higher the degree of association between the two objects. Specifically, a first threshold can be set. When the first number exceeds this first threshold, it indicates that the degree of association between the two objects is greater than a set value, and the association between them is considered valid. Specifically, when the first number of the object to be processed and any first sample object exceeds the first threshold, the object to be processed can be identified as a candidate target object. The first threshold can be determined based on the length of the time period corresponding to the acquisition of business transaction data and the target type. For example, if the target type is investment and wealth management and the length of the time period for acquiring business transaction data is 30 days, the first threshold can be set to 5. That is, when the first number corresponding to any two objects is greater than 5, it can be determined that the degree of association between the two objects is greater than a set value, and the association between them is considered valid.

[0139] By using the above method, the correlation between the object to be processed and the transaction object corresponding to any first sample object is accurately determined, thereby more accurately identifying candidate target objects from the objects to be processed.

[0140] After determining the initial number of common trading partners for any two objects, a correlation coefficient can be further determined based on these initial numbers. Thus, the initial numbers and correlation coefficient are used to define the degree of correlation between the two objects. The specific determination method is as follows:

[0141] In one possible implementation, the determination of the correlation between the object to be processed and the first sample object based on the first quantity includes:

[0142] For each object in the object to be processed and the first sample object, determine the second number of transaction objects corresponding to that object based on the object identifier of the transaction object corresponding to that object;

[0143] The correlation coefficient between the object to be processed and the first sample object is determined based on the first quantity and the second quantity corresponding to the object to be processed and the first sample object, respectively.

[0144] Based on the first quantity and the correlation coefficient, the degree of correlation between the object to be processed and the first sample object is determined.

[0145] In this implementation, the number of object identifiers of the transaction objects of the object can be determined as the second number of transaction objects corresponding to the object. Correspondingly, if at least two of the multiple transaction data corresponding to the object have the same object identifier, the object identifiers of the duplicate transaction objects can be directly determined as duplicates, that is, the number of the same object identifiers of the transaction objects is the number of repetitions. Of course, the object identifiers of the duplicate transaction objects can also be directly determined as the object identifiers of a single transaction object, that is, the number of the same object identifiers of the transaction objects is 1. This application embodiment does not limit this, but the former determination method is more accurate.

[0146] Given objects a and b, the second quantity corresponding to object a is C. a The second quantity corresponding to object b is C. b The first quantity corresponding to object a and object b is C. ab For example, the correlation coefficient α between the two ab for:

[0147]

[0148] Where, α ab The range of α is 0 to 1. abThe larger the value, the higher the correlation coefficient between object a and object b, the higher the degree of business overlap between them, and the higher the degree of correlation between them.

[0149] In one possible implementation, the correlation between the object to be processed and the first sample object includes a first quantity and a correlation coefficient. The statement that if the correlation between the object to be processed and any of the first sample objects is greater than a set value, the object to be processed is determined as a candidate target object includes:

[0150] If the first quantity corresponding to the object to be processed and the first sample object is greater than the first threshold, and the correlation coefficient between the object to be processed and the first sample object is greater than the second threshold, then the object to be processed is determined as a candidate target object. The set values ​​include the first threshold and the second threshold.

[0151] In this implementation, the specific values ​​of the first threshold and the second threshold can be determined according to the actual situation. For example, the first threshold can be set to 5 and the second threshold to 0.01.

[0152] By combining the first number of common transaction objects corresponding to any two objects and the second number of transaction objects corresponding to each object to determine the correlation coefficient between them, and by combining the first number and the correlation coefficient between them to determine the degree of correlation between them, the candidate target objects can be more accurately identified from the objects to be processed.

[0153] Step S540: For each candidate target object, based on the relevant data of the candidate target object, obtain the classification result of the candidate target object through the target classification model.

[0154] In one possible implementation, based on relevant data of the candidate target objects, the classification result of the candidate target objects is obtained through a target classification model, which may include:

[0155] Statistical analysis is performed on the business transaction data of the candidate target objects to obtain the indicator value of at least one transaction evaluation target corresponding to the candidate target objects;

[0156] Input the various indicator values ​​and object information corresponding to the candidate target objects into the target classification model, and then perform the following operations through the target classification model to obtain the classification results of the candidate target objects:

[0157] Based on the object information of the candidate target objects, determine the object feature vector of the candidate target objects;

[0158] Based on the various indicator values ​​corresponding to the candidate target objects, determine the business transaction feature vector of the candidate target objects;

[0159] The object feature vector and the business transaction feature vector are concatenated to obtain the target feature vector of the candidate target object;

[0160] Based on the target feature vector, the classification result of the candidate target object is obtained.

[0161] In this implementation, the transaction evaluation metrics for an object may include, but are not limited to, the unit price of each item offered by the object, the repurchase rate of transactions with the object (which can be determined by the ratio of the number of object identifiers of the same transaction object to the number of transactions), and the percentage of transaction amount per hour within 24 hours. Of course, the transaction evaluation metrics can also be specifically determined according to the target type, and this application embodiment does not impose any restrictions on this.

[0162] In this implementation, the object information of the candidate target object and the values ​​of each indicator can be input to... Figure 4 In the neural network model shown, the object feature vector and business transaction feature vector corresponding to the candidate sample object are obtained. The object feature vector and business transaction feature vector of the candidate sample object are then concatenated to obtain the target feature vector of the candidate sample object. Based on the target feature vector, a prediction is made to obtain the classification result of the candidate sample object.

[0163] When determining the classification result of each candidate sample object through the target classification model, for one candidate sample object, if the model output result is less than t0, the candidate sample object is determined to be a non-target type object; if the model output result is greater than t1, the candidate sample object is determined to be a target type object; if the model output result is greater than t0 and less than t1, or the model output result is equal to t0, or the model output result is less than t1, the candidate sample object is determined to be an object that the target classification model cannot determine. Then, the classification result of the candidate sample object can be determined according to the above method of determining the labels of each initial object.

[0164] By considering the correlation between the object to be processed and the first sample object of any target type, which is greater than a set value, the object to be processed is then identified as a candidate target object. The relevant data of the candidate target object is then input into the target classification model to obtain the classification result of the candidate target object. That is, only the correlation between the object to be processed and the first sample object of any target type is considered. If the business overlap between the object to be processed and the first sample object of the target type is greater than a set value, the corresponding classification result is determined by the target classification model. This can reduce the amount of data processing and determine the classification result of the object to be processed more quickly.

[0165] Step S550: Among the candidate target objects, the objects whose classification result is the target type are determined as the final target objects.

[0166] In this implementation, after determining the final target objects, a target object list can be formed based on these targets—that is, a list of all target objects existing in the target application. This target object list is then fed back to the terminal corresponding to the target application, enabling the terminal to perform relevant processing on the objects in the target object list. The terminal corresponding to the target application can be a management client that manages the objects within the target application.

[0167] As an example, when the target application is a shopping application and the target object is a merchant with a risk of fraud, the relevant actions taken by the management client on the objects in the target object list may include, but are not limited to, sending reminders to the target object in the shopping application, deleting content related to the target object, or closing the account of the target object.

[0168] In this implementation, a preset time period (e.g., 30 days) can also be set, and the determined target object list is updated every preset time period based on the object identification method provided in the embodiments of this application, so as to ensure the timeliness of the obtained target object list.

[0169] The object identification method provided in this application can simultaneously consider object information and business transaction data, enabling the obtained target classification model to fully integrate the characteristics of multiple data sources. This allows for a more accurate determination of the object's classification result based on its relevant data. Furthermore, by employing a dual screening mechanism of correlation degree comparison and model-based filtering, the screening efficiency and accuracy are improved, thereby more accurately identifying the target object in the target application.

[0170] In one possible implementation, the method may further include:

[0171] Based on the relevant data of each reference object, the degree of correlation between each reference object is determined. Each reference object includes each sample object corresponding to multiple training samples and at least one object to be processed mentioned above.

[0172] Each reference object is treated as a node, and edges are established between the nodes corresponding to two reference objects whose correlation degree is greater than a set value to obtain the object network.

[0173] For each object to be processed, the above-mentioned determination of the object to be processed as a candidate target object if the correlation between the object to be processed and any first sample object is greater than a set value may include:

[0174] Identify at least one node in the object network that has an edge connected to the node corresponding to the object to be processed;

[0175] If at least one node contains a first sample object among the reference objects corresponding to each node, the object to be processed is determined as a candidate target object.

[0176] For example, for a reference object, there may be multiple reference objects whose association with the reference object is greater than a set value; that is, in the object network, the node corresponding to the reference object will have multiple edges. Correspondingly, in the object network, each node that has an edge connected to a node corresponding to a sample object corresponds to multiple objects to be processed, and these multiple objects to be processed can all be identified as candidate target objects.

[0177] Taking a sample size of 9 objects and a quantity of 5 objects to be processed as an example, Figure 6 A schematic diagram of an object network provided in an embodiment of this application is shown. Figure 6 As shown, 1-9 represent the nodes corresponding to the sample objects, and a, b, c, d, and e represent the nodes corresponding to the objects to be processed. The node corresponding to sample object 1 is connected to the nodes corresponding to sample object 2, sample object 3, and the node corresponding to the object to be processed (b). The node corresponding to sample object 2 is also connected to the node corresponding to sample object 7. The node corresponding to sample object 7 is also connected to the node corresponding to the object to be processed (a). The node corresponding to sample object 3 is also connected to the nodes corresponding to the objects to be processed (c) and sample object 9. The node corresponding to sample object 6 is connected to the node corresponding to the object to be processed (d). The nodes corresponding to sample object 5, sample object 8, and the object to be processed (e) are all isolated nodes in the object network and are not connected to nodes corresponding to other reference objects. Figure 6 The object network shown can identify objects a, b, c, d, and e as candidate target objects.

[0178] In this implementation, each sample object and each reference object in each object to be processed is treated as a node. An edge is established between the corresponding nodes of two reference objects whose correlation degree is greater than a set value to obtain an object network. This visualizes the relationship between the reference objects. Thus, other reference objects with a business overlap degree higher than a preset value with each reference object can be identified more clearly and explicitly through the object network. In other words, the candidate target objects in each object to be processed can be identified more quickly through this object network.

[0179] Of course, after obtaining the first dataset, the corresponding object network can also be determined based on the degree of correlation between any two objects in the first dataset and each candidate sample object, and then the expanded sample objects in the candidate sample objects can be determined based on the object network.

[0180] In one possible implementation, the plurality of training samples may further include a plurality of second samples, each second sample corresponding to a sample object that is a non-target type second sample object. The above method may further include:

[0181] The relevant data of each final target object is used as the first new sample, and the relevant data of each candidate object other than the final target object is used as the second new sample to expand the training set.

[0182] The target classification model is updated and trained based on the expanded training set to obtain the updated target classification model.

[0183] In this implementation, the embodiments of this application do not limit the method of expanding the training set based on the newly added first sample and the newly added second sample. For example, the newly added first sample, the newly added second sample, and multiple training samples from the original training set can be directly mixed to obtain the expanded training set. The original training set can be expanded according to a certain ratio. For example, a 4:5 ratio can be set, selecting 4 / 5 of the samples from the newly added first sample, 4 / 5 of the samples from the newly added second sample, and 4 / 5 of the positive samples (first samples) and 4 / 5 of the negative samples (second samples) from the original training set, and then mixing the selected samples to obtain the expanded training set. Alternatively, the same number of first samples as the newly added first samples and the same number of second samples as the newly added second samples can be deleted from the original training set, and the training set after deleting some data can be mixed with the newly added first and second samples to obtain the expanded training set.

[0184] After identifying the final target object, the relevant data of the final target object is used as the first new sample, and the relevant data of all candidate objects other than the final target object are used as the second new sample. This continuous expansion of the training set, along with the updated training set, allows for continuous training and updating of the target classification model. This results in an updated target classification model that is more accurate, has wider coverage, broader applicability, and better timely, thus better meeting practical needs. Over time, through continuous updates and iterations of the target classification model, the training process can gradually become fully automated, eliminating the need for manual label determination and reducing reliance on human intervention.

[0185] To more clearly illustrate the object recognition method shown in the embodiments of this application, the following is combined with... Figure 7 The flowchart shown illustrates the object identification method.

[0186] Step 1: Using the current time point as the time endpoint, obtain the relevant data corresponding to multiple initial objects within a preset time period in the target application, and use the obtained relevant data corresponding to the multiple initial objects as the initial dataset.

[0187] Step 2: Based on the relevant data of each initial object in the initial dataset, determine the label of each initial object and add labels to each initial object in the initial dataset.

[0188] Step 3: Construct an object network based on the relevant data corresponding to each initial object in the initial dataset, and update the object network continuously based on the relevant data of objects appearing in the target application within a preset time period.

[0189] Step 4: If there is a node in the updated object network that is connected to the node corresponding to the object labeled "This object is an object of the target type" in each of the initial objects, determine that the object corresponding to that node is a new object. Continue in this manner until there are no nodes corresponding to new objects in the updated object network.

[0190] Step 5: Based on the relevant data of each new object, determine the label of each new object, and expand the initial dataset with the added labels to obtain the training set.

[0191] Step 6: Train the target classification model based on the training set.

[0192] Step 7: Obtain relevant data for the object to be processed.

[0193] Step 8: Obtain the current object network including the nodes corresponding to the object to be processed, and determine whether the object to be processed is a candidate target object.

[0194] Step 9: Determine the classification results of the candidate target objects using the target classification model. Expand the training set based on the classification results of each candidate target object and update the target classification model.

[0195] Step 10: Determine if there are any new objects to be processed in the current object network. If so, continue with Step 7 until there are no new objects to be processed.

[0196] Step 11: If there are no new objects to be processed, generate a list of target categories in the target application based on the identified target objects.

[0197] Based on the same principle as the object recognition method provided in the embodiments of this application, an apparatus is also provided in the embodiments of this application. Figure 8 A schematic diagram of an object recognition device applicable to an embodiment of this application is shown. For example... Figure 8 As shown, the device 800 includes:

[0198] The first acquisition module 801 is used to acquire the training set and the target classification model trained based on the training set;

[0199] The training set includes multiple training samples. Each training sample includes relevant data of a sample object. The relevant data includes object information and business transaction data. The sample object is an object in the target application. The business transaction data includes business transaction data corresponding to the target application. The multiple training samples include multiple first samples. The sample object corresponding to each first sample is a first sample object of the target type.

[0200] The second acquisition module 802 is used to acquire relevant data of at least one object to be processed in the target application;

[0201] The candidate target object determination module 803 is used to determine the object to be processed as a candidate target object if the correlation between the object to be processed and any first sample object is greater than a set value for each object to be processed. The correlation between any two objects is determined based on the relevant data of any two objects.

[0202] The classification module 804 is used to obtain the classification result of each candidate target object based on the relevant data of the candidate target object through the target classification model;

[0203] The target object determination module 805 is used to determine the objects that are classified as target types among the candidate target objects as the final target objects.

[0204] In one possible implementation, the device 800 may further include an object network construction module for:

[0205] Based on the relevant data of each reference object, the degree of correlation between each reference object is determined. Each reference object includes each sample object corresponding to multiple training samples and at least one object to be processed.

[0206] Each reference object is treated as a node, and edges are established between the nodes corresponding to two reference objects whose correlation degree is greater than a set value to obtain the object network.

[0207] The aforementioned candidate target object determination module 803, for each object to be processed, determines the object to be processed as a candidate target object if the correlation between the object to be processed and any first sample object is greater than a set value. Specifically, it can be used for:

[0208] Identify at least one node in the object network that has an edge connected to the node corresponding to the object to be processed;

[0209] If at least one node contains a first sample object among the reference objects corresponding to each node, the object to be processed is determined as a candidate object.

[0210] In one possible implementation, the plurality of training samples may further include a plurality of second samples, each second sample corresponding to a sample object that is a non-target type second sample object. The device 800 may further include a training set expansion module and a model update module.

[0211] The training set expansion module is used to expand the training set by taking the relevant data of each determined final target object as the first new sample and taking the relevant data of each candidate target object other than the final target object as the second new sample.

[0212] The model update module is used to update and train the target classification model based on the expanded training set to obtain the updated target classification model.

[0213] In one possible implementation, for any object, the business transaction data may include relevant information about the transaction object corresponding to that object, and the relevant information may include an object identifier; the degree of association between the above-mentioned object to be processed and any first sample object may be determined by the candidate target object determination module 803 in the following way:

[0214] Based on the object identifiers of the transaction objects corresponding to the object to be processed and any first sample object, determine the first number of common transaction objects corresponding to the object to be processed and the first sample object;

[0215] Based on the first quantity, determine the degree of association between the object to be processed and the first sample object.

[0216] In one possible implementation, when determining the correlation between the object to be processed and the first sample object based on the first quantity, the candidate target object determination module 803 can specifically be used for:

[0217] For each object in the object to be processed and the first sample object, determine the second number of transaction objects corresponding to that object based on the object identifier of the transaction object corresponding to that object;

[0218] The correlation coefficient between the object to be processed and the first sample object is determined based on the first quantity and the second quantity corresponding to the object to be processed and the first sample object, respectively.

[0219] Based on the first quantity and the correlation coefficient, the degree of correlation between the object to be processed and the first sample object is determined.

[0220] In one possible implementation, the correlation between the object to be processed and the first sample object includes a first quantity and a correlation coefficient. Specifically, when the candidate target object determination module 803 determines the object to be processed as a candidate target object if the correlation between the object to be processed and any first sample object is greater than a set value, it can be used to:

[0221] If the first quantity corresponding to the object to be processed and the first sample object is greater than the first threshold, and the correlation coefficient between the object to be processed and the first sample object is greater than the second threshold, then the object to be processed is determined as a candidate target object. The set values ​​include the first threshold and the second threshold.

[0222] In one possible implementation, the first acquisition module 801, when acquiring the training set, can specifically be used for:

[0223] Obtain a first initial dataset and a keyword library corresponding to objects of the target type. The first initial dataset includes relevant data for multiple first initial objects.

[0224] Based on the object information of each first initial object, objects whose object information contains at least one keyword from the keyword library are used as samples of the selected target types.

[0225] The relevant data of each first initial object in the first initial dataset, excluding the selected target type sample objects, are provided to the annotators to obtain the labels corresponding to each first initial object, excluding the selected target type sample objects. The labels indicate whether the corresponding first initial object is a target type sample object or a non-target type sample object.

[0226] A training set is constructed based on the relevant data and corresponding labels of each initial object in the first initial dataset.

[0227] The first acquisition module 801 mentioned above can also be specifically used for:

[0228] Obtain the second initial dataset, which includes relevant data for multiple second initial objects;

[0229] For each second initial object, if the correlation between the second initial object and any target type sample object in the first initial dataset is greater than a set value, the second initial object is determined as a candidate sample object of the target type.

[0230] The relevant data of each candidate sample object is provided to the annotators so that they can verify the type of each candidate sample object and obtain the verified label of each candidate sample object.

[0231] When the first acquisition module 801 constructs a training set based on the relevant data and corresponding labels of each first initial object in the first initial dataset, it can be specifically used for:

[0232] A training set is constructed based on the relevant data and corresponding labels of each initial object in the first initial dataset, and the relevant data and verified labels of each candidate sample object in the second initial dataset.

[0233] In one possible implementation, when the classification module 804 obtains the classification result of the candidate target object based on the relevant data of the candidate target object through the target classification model, it can be specifically used for:

[0234] Statistical analysis is performed on the business transaction data of the candidate target objects to obtain the indicator values ​​of at least one transaction evaluation indicator corresponding to the candidate target objects;

[0235] Input the various indicator values ​​and object information corresponding to the candidate target objects into the target classification model, and then perform the following operations through the target classification model to obtain the classification results of the candidate target objects:

[0236] Based on the object information of the candidate target objects, determine the object feature vector of the candidate target objects;

[0237] Based on the various indicator values ​​corresponding to the candidate target objects, determine the business transaction feature vector of the candidate target objects;

[0238] The object feature vector is concatenated with the business transaction feature vector to obtain the target feature vector of the candidate target object;

[0239] Based on the target feature vector, the classification results of the candidate target feature vectors are obtained.

[0240] The apparatus in this application embodiment can execute the method provided in this application embodiment, and the implementation principle is similar. The actions performed by each module in the apparatus of each embodiment of this application correspond to the steps in the method of each embodiment of this application. For detailed functional descriptions of each module of the apparatus, please refer to the descriptions in the corresponding methods shown above, which will not be repeated here.

[0241] Based on the same principle as the object recognition method and apparatus provided in the embodiments of this application, the embodiments of this application also provide an electronic device (such as a server), which may include a memory, a processor and a computer program stored in the memory, the processor executing the computer program to implement the steps of the method provided in any optional embodiment of this application.

[0242] Optionally, Figure 9 A schematic diagram of the structure of an electronic device to which this application embodiment applies is shown, such as... Figure 9 As shown, Figure 9 The illustrated electronic device 4000 includes a processor 4001 and a memory 4003. The processor 4001 and the memory 4003 are connected, for example, via a bus 4002. Optionally, the electronic device 4000 may further include a transceiver 4004, which can be used for data interaction between the electronic device and other electronic devices, such as sending and / or receiving data. It should be noted that in practical applications, the transceiver 4004 is not limited to one type, and the structure of the electronic device 4000 does not constitute a limitation on the embodiments of this application.

[0243] Processor 4001 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 4001 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0244] Bus 4002 may include a pathway for transmitting information between the aforementioned components. Bus 4002 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Bus 4002 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 9 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.

[0245] The memory 4003 may be ROM (Read Only Memory) or other types of static storage devices capable of storing static information and instructions, RAM (Random Access Memory) or other types of dynamic storage devices capable of storing information and instructions, or EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media, other magnetic storage devices, or any other medium capable of carrying or storing computer programs and capable of being read by a computer, without limitation herein.

[0246] The memory 4003 stores computer programs that execute embodiments of this application, and its execution is controlled by the processor 4001. The processor 4001 executes the computer programs stored in the memory 4003 to implement the steps shown in the foregoing method embodiments.

[0247] This application provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, it can implement the steps and corresponding content of the aforementioned method embodiments.

[0248] This application also provides a computer program product, including a computer program that, when executed by a processor, can implement the steps and corresponding content of the aforementioned method embodiments.

[0249] The terms "first," "second," "third," "fourth," "1," "2," etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in a sequence other than that shown in the figures or text.

[0250] It should be understood that although arrows indicate various operation steps in the flowcharts of this application's embodiments, the order in which these steps are implemented is not limited to the order indicated by the arrows. Unless explicitly stated herein, in some implementation scenarios of this application's embodiments, the implementation steps in each flowchart can be executed in other orders as required. Furthermore, some or all steps in each flowchart, based on the actual implementation scenario, may include multiple sub-steps or multiple stages. Some or all of these sub-steps or stages can be executed at the same time, and each sub-step or stage can also be executed at different times. In scenarios where execution times differ, the execution order of these sub-steps or stages can be flexibly configured according to requirements, and this application's embodiments do not limit this.

[0251] The above description is only an optional implementation method for some implementation scenarios of this application. It should be noted that for those skilled in the art, other similar implementation methods based on the technical concept of this application without departing from the technical concept of this application also fall within the protection scope of the embodiments of this application.

Claims

1. An object recognition method, characterized in that, include: Obtain a training set and a target classification model trained based on the training set; The training set includes multiple training samples, each training sample includes relevant data of a sample object, the relevant data includes object information and business transaction data, the sample object is an object in the target application, the business transaction data includes business transaction data corresponding to the target application, the multiple training samples include multiple first samples, and the sample object corresponding to each first sample is a first sample object of the target type; Obtain relevant data of at least one object to be processed in the target application; For each object to be processed, a first number of common transaction objects corresponding to the object to be processed and the first sample object are determined based on the object identifier of the transaction object corresponding to the object to be processed and any of the first sample objects respectively; based on the first number, the degree of association between the object to be processed and the first sample object is determined; if the degree of association between the object to be processed and any of the first sample objects is greater than a set value, the object to be processed is determined as a candidate target object; For each candidate target object, based on the relevant data of the candidate target object, the classification result of the candidate target object is obtained through the target classification model; The object that is classified as the target type among the candidate target objects is determined as the final target object.

2. The method according to claim 1, characterized in that, The method further includes: Based on the relevant data of each reference object, the degree of correlation between each reference object is determined. Each reference object includes each sample object corresponding to the multiple training samples and the at least one object to be processed. Each reference object is treated as a node, and an edge is established between the nodes corresponding to two reference objects whose correlation degree is greater than the set value to obtain an object network. For each of the objects to be processed, the step of determining the object to be processed as a candidate target object if the correlation between the object to be processed and any of the first sample objects is greater than a set value includes: Identify at least one node in the object network that has an edge connected to the node corresponding to the object to be processed; If a first sample object exists among the reference objects corresponding to each of the at least one node, the object to be processed is determined to be a candidate target object.

3. The method according to claim 1 or 2, characterized in that, The plurality of training samples further includes a plurality of second samples, each second sample corresponding to a sample object that is a non-target type second sample object; the method further includes: The relevant data of each final target object is determined as the newly added first sample, and the relevant data of each object other than the final target object in each candidate target object is added as the newly added second sample, thereby expanding the training set; The target classification model is updated and trained based on the expanded training set to obtain the updated target classification model.

4. The method according to claim 1 or 2, characterized in that, The step of determining the correlation between the object to be processed and the first sample object based on the first quantity includes: For each object in the object to be processed and the first sample object, a second number of transaction objects corresponding to that object is determined based on the object identifier of the transaction object corresponding to that object; Based on the first quantity and the second quantities corresponding to the object to be processed and the first sample object, the correlation coefficient between the object to be processed and the first sample object is determined; Based on the first quantity and the correlation coefficient, the degree of correlation between the object to be processed and the first sample object is determined.

5. The method according to claim 4, characterized in that, The degree of correlation between the object to be processed and the first sample object includes the first quantity and the correlation coefficient. The step of determining the object to be processed as a candidate target object if the degree of correlation between the object to be processed and any of the first sample objects is greater than a set value includes: If the first quantity corresponding to the object to be processed and the first sample object is greater than the first threshold, and the correlation coefficient between the object to be processed and the first sample object is greater than the second threshold, then the object to be processed is determined as a candidate target object, wherein the set value includes the first threshold and the second threshold.

6. The method according to claim 1, characterized in that, The acquisition of the training set includes: Obtain a first initial dataset and a keyword library corresponding to the target type object. The first initial dataset includes relevant data for multiple first initial objects. Based on the object information of each of the first initial objects, objects whose object information contains at least one keyword from the keyword library are selected as sample objects of the target type. The relevant data of each first initial object in the first initial dataset, excluding the sample objects of the selected target type, are provided to the annotator to obtain the label corresponding to each first initial object, excluding the sample objects of the selected target type. The label indicates whether the corresponding first initial object is a sample object of the target type or a sample object of a non-target type. The training set is constructed based on the relevant data and corresponding labels of each of the first initial objects in the first initial dataset.

7. The method according to claim 6, characterized in that, The method further includes: Obtain a second initial dataset, which includes relevant data for multiple second initial objects; For each second initial object, if the correlation between the second initial object and any target type sample object in the first initial dataset is greater than a set value, the second initial object is determined as a candidate sample object of the target type. The relevant data of each of the identified candidate sample objects are provided to the annotator so that the annotator can verify the type of each candidate sample object and obtain the verified label of each candidate sample object; The training set is constructed based on the relevant data and corresponding labels of each of the first initial objects in the first initial dataset, including: The training set is constructed based on the relevant data and corresponding labels of each of the first initial objects in the first initial dataset, and the relevant data and verified labels of each of the candidate sample objects in the second initial dataset.

8. The method according to claim 1, characterized in that, The process of obtaining the classification result of the candidate target object based on the relevant data of the candidate target object and through the target classification model includes: Statistical analysis is performed on the business transaction data of the candidate target objects to obtain the indicator values ​​of at least one transaction evaluation indicator corresponding to the candidate target objects; The various indicator values ​​and object information corresponding to the candidate target objects are input into the target classification model. The target classification model then performs the following operations to obtain the classification result of the candidate target objects: Based on the object information of the candidate target object, determine the object feature vector of the candidate target object; Based on the various indicator values ​​corresponding to the candidate target objects, the business transaction feature vector of the candidate target objects is determined; The object feature vector is concatenated with the business transaction feature vector to obtain the target feature vector of the candidate target object; Based on the target feature vector, the classification result of the candidate target object is obtained.

9. An object recognition device, characterized in that, include: The first acquisition module is used to acquire a training set and a target classification model trained based on the training set. The training set includes multiple training samples, each training sample includes relevant data of a sample object, the relevant data includes object information and business transaction data, the sample object is an object in the target application, the business transaction data includes business transaction data corresponding to the target application, the multiple training samples include multiple first samples, and the sample object corresponding to each first sample is a first sample object of the target type; The second acquisition module is used to acquire relevant data of at least one object to be processed in the target application; The candidate target object determination module is used to, for each object to be processed, determine a first number of common transaction objects corresponding to the object to be processed and the first sample object based on the object identifiers of the transaction objects corresponding to the object to be processed and any of the first sample objects respectively; determine the degree of association between the object to be processed and the first sample object based on the first number; if the degree of association between the object to be processed and any of the first sample objects is greater than a set value, determine the object to be processed as a candidate target object, wherein the degree of association between any two objects is determined based on the relevant data of the two objects; The classification module is used to obtain the classification result of each candidate target object based on the relevant data of the candidate target object and through the target classification model. The target object determination module is used to determine the objects that are classified as target types from the candidate target objects as the final target objects.

10. The apparatus according to claim 9, characterized in that, The device further includes an object network construction module for: Based on the relevant data of each reference object, the degree of correlation between each reference object is determined. Each reference object includes each sample object corresponding to the multiple training samples and the at least one object to be processed. Each reference object is treated as a node, and an edge is established between the nodes corresponding to two reference objects whose correlation degree is greater than the set value to obtain an object network. For each of the objects to be processed, if the correlation between the object to be processed and any of the first sample objects is greater than a set value, the candidate target object determination module, when determining the object to be processed as a candidate target object, is used to: Identify at least one node in the object network that has an edge connected to the node corresponding to the object to be processed; If a first sample object exists among the reference objects corresponding to each of the at least one node, the object to be processed is determined to be a candidate target object.

11. The apparatus according to claim 9 or 10, characterized in that, The plurality of training samples also includes a plurality of second samples, each second sample corresponding to a non-target type second sample object. The device further includes a training set expansion module and a model update module. The training set expansion module is used to expand the training set by taking the relevant data of each determined final target object as the newly added first sample and taking the relevant data of each object other than the final target object in each candidate target object as the newly added second sample. The model update module is used to update and train the target classification model based on the expanded training set to obtain the updated target classification model.

12. The apparatus according to claim 9 or 10, characterized in that, When determining the correlation between the candidate target object and the first sample object based on the first quantity, the candidate target object determination module is used to: For each object in the object to be processed and the first sample object, a second number of transaction objects corresponding to that object is determined based on the object identifier of the transaction object corresponding to that object; Based on the first quantity and the second quantities corresponding to the object to be processed and the first sample object, the correlation coefficient between the object to be processed and the first sample object is determined; Based on the first quantity and the correlation coefficient, the degree of correlation between the object to be processed and the first sample object is determined.

13. The apparatus according to claim 12, characterized in that, The correlation between the object to be processed and the first sample object includes the first quantity and the correlation coefficient. If the correlation between the object to be processed and any first sample object is greater than a set value, the candidate target object determination module, when determining the object to be processed as a candidate target object, is used to: If the first quantity corresponding to the object to be processed and the first sample object is greater than the first threshold, and the correlation coefficient between the object to be processed and the first sample object is greater than the second threshold, then the object to be processed is determined as a candidate target object, wherein the set value includes the first threshold and the second threshold.

14. The apparatus according to claim 9, characterized in that, When acquiring the training set, the first acquisition module is used to: Obtain a first initial dataset and a keyword library corresponding to the target type object. The first initial dataset includes relevant data for multiple first initial objects. Based on the object information of each of the first initial objects, objects whose object information contains at least one keyword from the keyword library are selected as sample objects of the target type. The relevant data of each first initial object in the first initial dataset, excluding the sample objects of the selected target type, are provided to the annotator to obtain the label corresponding to each first initial object, excluding the sample objects of the selected target type. The label indicates whether the corresponding first initial object is a sample object of the target type or a sample object of a non-target type. The training set is constructed based on the relevant data and corresponding labels of each of the first initial objects in the first initial dataset.

15. The apparatus according to claim 14, characterized in that, The first acquisition module is also used for: Obtain a second initial dataset, which includes relevant data for multiple second initial objects; For each second initial object, if the correlation between the second initial object and any target type sample object in the first initial dataset is greater than a set value, the second initial object is determined as a candidate sample object of the target type. The relevant data of each of the identified candidate sample objects are provided to the annotator so that the annotator can verify the type of each candidate sample object and obtain the verified label of each candidate sample object; When the first acquisition module constructs the training set based on the relevant data and corresponding labels of each of the first initial objects in the first initial dataset, it is used to: The training set is constructed based on the relevant data and corresponding labels of each of the first initial objects in the first initial dataset, and the relevant data and verified labels of each of the candidate sample objects in the second initial dataset.

16. The apparatus according to claim 9, characterized in that, When the classification module obtains the classification result of the candidate target object based on the relevant data of the candidate target object and through the target classification model, it is used to: Statistical analysis is performed on the business transaction data of the candidate target objects to obtain the indicator values ​​of at least one transaction evaluation indicator corresponding to the candidate target objects; The various indicator values ​​and object information corresponding to the candidate target objects are input into the target classification model. The target classification model then performs the following operations to obtain the classification result of the candidate target objects: Based on the object information of the candidate target object, determine the object feature vector of the candidate target object; Based on the various indicator values ​​corresponding to the candidate target objects, the business transaction feature vector of the candidate target objects is determined; The object feature vector is concatenated with the business transaction feature vector to obtain the target feature vector of the candidate target object; Based on the target feature vector, the classification result of the candidate target object is obtained.

17. An electronic device comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1-8.

18. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1-8.

19. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1-8.

Images