Customer label approval method, device, equipment, medium and program product

Abstract

The application provides a customer label approval method and device, equipment, storage medium and program product, which can be applied to the field of big data, and specifically relates to a big model in a customer label approval scene. The method comprises the following steps: extracting a label name corresponding to a customer label and a business definition text from a customer label approval request sent by a business end; calculating a first probability that a customer type indicated by the customer label is a target customer type based on the label name and the business definition text; performing prompt word reasoning on key information extracted from the business definition text, and calculating a completeness score of the customer label according to a reasoning result, wherein the completeness score indicates the completeness of the key information and the logical consistency of the business content, and the logical consistency indicates the matching degree of the target customer type and the business content; weighting the first probability and the completeness score to obtain a comprehensive score; and regarding a customer label with a comprehensive score greater than a preset threshold as a customer label approved.

Description

Technical Field

[0001] This application relates to the field of big data, specifically to the application of large models in customer tag approval scenarios, and more specifically to a customer tag approval method, device, equipment, medium, and program product. Background Technology

[0002] Customer tags are crucial in scenarios such as transaction risk control and financial product promotion. Whether customer tags match relevant business rules and customer types directly impacts their usability in financial settings. Therefore, approving customer tags to obtain more accurate ones is a way to improve their accuracy in financial applications. However, typical tag approval methods only verify the tag type or name and rely primarily on string matching to identify duplicate tags in the tag library. This results in low accuracy and an inability to obtain precise customer tags. Summary of the Invention

[0003] In view of the above problems, embodiments of this application provide a customer label approval method, apparatus, device, medium, and program product.

[0004] According to a first aspect of this application, a customer tag approval method is provided, comprising: extracting the tag name and business definition text corresponding to the customer tag from a customer tag approval request sent from the business end, wherein the business definition text indicates the business content corresponding to the customer tag; calculating a first probability that the customer type indicated by the customer tag is the target customer type based on the tag name and the business definition text; performing cue word reasoning on key information extracted from the business definition text, calculating a completeness score of the customer tag based on the reasoning result, wherein the completeness score indicates the completeness of the key information and the logical consistency of the business content, wherein the logical consistency indicates the degree of matching between the target customer type and the business content, and the key information indicates the contextual information of the business content; weighting the first probability and the completeness score to obtain a comprehensive score; and approving customer tags with a comprehensive score greater than a preset threshold as approved customer tags.

[0005] According to an embodiment of this application, the method further includes: obtaining a first semantic vector of the business definition text and inputting it into a tag type classifier to obtain a second probability that the tag type of the customer tag is the target tag type; and weighting the first probability, the second probability, and the integrity score to obtain a comprehensive score.

[0006] According to an embodiment of this application, before weighting the first probability, the second probability, and the completeness score, the method includes: matching the target label type with the weight adjustment rules to adjust the weights corresponding to the first probability, the second probability, and the completeness score respectively; wherein, the weight adjustment rules indicate the mapping relationship between the importance of the target customer type in the target business definition text and the weight of the first probability, the mapping relationship between the importance of the target label type in the target business definition text and the weight of the second probability, and the mapping relationship between the importance of the customer label's completeness in the target business definition text and the weight of the completeness score, the target business definition text representing the business definition text corresponding to the target label type, and the target customer type and the target label type changing with the change of the customer label respectively.

[0007] According to an embodiment of this application, matching target label types with weight adjustment rules to adjust the weights corresponding to the first probability, the second probability, and the integrity score includes: if there are multiple target label types, matching each target label type with the weight adjustment rules to obtain a weight adjustment strategy corresponding to each target label type; determining the initial weights of the first probability, the second probability, and the integrity score under the corresponding weight adjustment strategy based on each weight adjustment strategy; and calculating the mean of the initial weights of the first probability, the second probability, and the integrity score under each weight adjustment strategy to obtain the weights corresponding to the first probability, the second probability, and the integrity score.

[0008] According to an embodiment of this application, based on the tag name and business definition text, calculating the first probability that the customer type indicated by the customer tag is the target customer type includes: obtaining the second semantic vector of the tag name; performing a weighted fusion of the first semantic vector and the second semantic vector to obtain a fusion vector; inputting the fusion vector into a fully connected layer and performing a nonlinear transformation to obtain the first probability of each customer type corresponding to the customer tag; sorting the first probabilities of each customer type in descending order, and taking the customer type ranked first as the target customer type, wherein the first probability of the customer type ranked first is the first probability of the target customer type.

[0009] According to an embodiment of this application, the method further includes: obtaining a tag template corresponding to a customer tag based on the target tag type, and calculating a first matching degree between the customer tag and the tag template; if the first matching degree is greater than a first preset matching degree, comparing the words in the customer tag with a second matching degree in a preset thesaurus, wherein the words in the preset thesaurus are prohibited words; if the second matching degree is greater than a second preset matching degree, returning the customer tag to the business end.

[0010] According to an embodiment of this application, the method further includes: extracting the customer type of the customer tag from the customer tag approval request, calculating the similarity between the target customer type and the extracted customer type; calculating the confidence level of the target customer type based on the similarity; and if the confidence level is less than a first preset confidence level, sending the customer tag to the approval platform for review.

[0011] According to a second aspect of this application, a customer tag approval device is provided, comprising: an information extraction module, used to extract the tag name and business definition text corresponding to the customer tag from a customer tag approval request sent from a business terminal, wherein the business definition text indicates the business content corresponding to the customer tag; a first probability acquisition module, used to calculate a first probability that the customer type indicated by the customer tag is the target customer type based on the tag name and the business definition text; a completeness score acquisition module, used to perform prompt word reasoning on key information extracted from the business definition text, and calculate a completeness score of the customer tag based on the reasoning result, wherein the completeness score indicates the completeness of the key information and the logical consistency of the business content, the logical consistency indicates the degree of matching between the target customer type and the business content, and the key information indicates the contextual information of the business content; a comprehensive score acquisition module, used to weight the first probability and the completeness score to obtain a comprehensive score; and a tag approval module, used to approve customer tags whose comprehensive score is greater than a preset threshold as approved customer tags.

[0012] According to a third aspect of this application, an electronic device is provided, comprising: one or more processors; and a memory for storing one or more computer programs, wherein the one or more processors execute the one or more computer programs to implement the steps of the method described above.

[0013] According to a fourth aspect of this application, a computer-readable storage medium is also provided, on which a computer program or instructions are stored, wherein the computer program or instructions, when executed by a processor, implement the steps of the above-described method.

[0014] According to a fifth aspect of this application, a computer program product is also provided, including a computer program or instructions that, when executed by a processor, implement the steps of the above-described method. Attached Figure Description

[0015] The above-mentioned contents, other objects, features and advantages of this application will become clearer from the following description of embodiments with reference to the accompanying drawings, in which:

[0016] Figure 1 The illustrations depict application scenarios of customer tag approval methods, apparatus, devices, media, and program products according to embodiments of this application.

[0017] Figure 2A flowchart illustrating a customer label approval method according to an embodiment of this application is shown schematically;

[0018] Figure 3 The illustration shows a schematic diagram of a comprehensive score calculation method according to an embodiment of this application;

[0019] Figure 4 This illustration schematically shows a weight adjustment method according to an embodiment of the present application;

[0020] Figure 5 This illustration schematically shows a method for determining a target customer type according to an embodiment of this application;

[0021] Figure 6 This illustration schematically shows another label approval method according to an embodiment of this application;

[0022] Figure 7 This schematic diagram illustrates a structural block diagram of a customer label approval device according to an embodiment of this application;

[0023] Figure 8 A block diagram schematically illustrates an electronic device suitable for implementing a customer tag approval method according to an embodiment of this application. Detailed Implementation

[0024] The embodiments of this application will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of this application. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of this application for ease of explanation. However, it will be apparent that one or more embodiments may be implemented without these specific details. Furthermore, descriptions of well-known structures and technologies are omitted in the following description to avoid unnecessarily obscuring the concepts of this application.

[0025] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0026] All terms used herein (including technical and scientific terms) have the meanings commonly understood by those skilled in the art, unless otherwise defined. It should be noted that the terms used herein are to be interpreted in a manner consistent with the context of this specification, and not in an idealized or overly rigid way.

[0027] When using expressions such as "at least one of A, B and C", they should generally be interpreted in accordance with the meaning that is commonly understood by those skilled in the art (e.g., "a system having at least one of A, B and C" should include, but is not limited to, a system having A alone, a system having B alone, a system having C alone, a system having A and B, a system having A and C, a system having B and C, and / or a system having A, B and C, etc.).

[0028] In the technical solution of this application, the user information (including but not limited to user personal information, user image information, user device information, such as location information) and data (including but not limited to data used for analysis, stored data, and displayed data) involved are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, storage, use, processing, transmission, provision, disclosure, and application of related data all comply with relevant laws, regulations, and standards, take necessary confidentiality measures, do not violate public order and good morals, and provide corresponding operation entry points for users to choose to authorize or refuse.

[0029] In scenarios where personal information is used for automated decision-making, the methods, devices, and systems provided in this application all provide users with corresponding operation entry points for users to choose to agree to or reject the automated decision results; if the user chooses to reject, the process enters the expert decision-making process.

[0030] The term "large model" as used in this application refers to a deep learning model with a large number of model parameters. Large models typically contain hundreds of millions, tens of billions, hundreds of billions, trillions, or even more than ten trillion model parameters. Large models can include large-scale language models (LLMs), large visual models, multimodal large models, and so on. The large models involved in the embodiments of this disclosure can be general-purpose large models, or they can be expert large models obtained through fine-tuning based on requirements. The embodiments of this disclosure do not limit this to either.

[0031] Figure 1 The illustration depicts an application scenario of the customer tag approval method according to an embodiment of this application. For example... Figure 1As shown, application scenario 100 according to an embodiment of this application may include a first terminal device 101, a second terminal device 102, a third terminal device 103, a network 104, and a server 105. The network 104 serves as a medium for providing a communication link between the first terminal device 101, the second terminal device 102, the third terminal device 103, and the server 105. The network 104 may include various connection types, such as wired or wireless communication links or fiber optic cables. For example, a user can use the first terminal device 101, the second terminal device 102, and the third terminal device 103 to interact with the server 105 through the network 104 to receive or send information, etc.

[0032] The first terminal device 101, the second terminal device 102, and the third terminal device 103 can be electronic devices such as smartphones, wearable devices, personal computers, intelligent voice interaction devices, smart home appliances, intelligent vehicles, in-vehicle terminals, aircraft, unmanned vending terminals, and extended reality devices. Extended reality devices can include virtual reality devices, augmented reality devices, and mixed reality devices. A client application for the target application can be installed and run on the terminal devices. This target application can include, but is not limited to, financial transaction applications, payment applications, shopping applications, web browser applications, search applications, instant messaging tools, email clients, and social media platform software (these are just examples). Furthermore, this application embodiment does not limit the form of the target application, and it can include, but is not limited to, applications, mini-programs, etc., installed on the terminal devices, and can also be in the form of web pages.

[0033] Server 105 can be a server providing various services, such as a backend management server supporting websites browsed by users using the first terminal device 101, the second terminal device 102, and the third terminal device 103 (this is just an example). The backend management server can analyze and process received user requests and other data, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal devices. 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 such as cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks, and basic cloud computing services such as big data. The server can be the backend server of the aforementioned target application, used to provide backend services to the clients of the target application.

[0034] It should be noted that the customer tag approval method provided in this application embodiment can generally be executed by server 105 and / or terminal devices 101-103. Accordingly, the customer tag approval device provided in this application embodiment can generally be set in server 105 and / or terminal devices 101-103.

[0035] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0036] Figure 2 A flowchart illustrating a customer label approval method according to an embodiment of this application is shown schematically. Figure 2 As shown, the customer tag approval method 200 according to an embodiment of this application may include steps S210 to S250.

[0037] In step S210, the tag name and business definition text corresponding to the customer tag are extracted from the customer tag approval request sent by the business side. The business definition text indicates the business content corresponding to the customer tag.

[0038] In step S220, based on the label name and business definition text, the first probability that the customer type indicated by the customer label is the target customer type is calculated.

[0039] In step S230, key information extracted from the business definition text is subjected to prompt word reasoning. Based on the reasoning results, the completeness score of the customer tag is calculated. The completeness score indicates the completeness of the key information and the logical consistency of the business content. Logical consistency indicates the degree of matching between the target customer type and the business content. Key information indicates the contextual information of the business content.

[0040] In step S240, the first probability and integrity scores are weighted to obtain a comprehensive score.

[0041] In step S250, customer tags with a comprehensive score greater than a preset threshold are approved as customer tags.

[0042] In some embodiments, in step S210, customer (or user) information can be sent from the business side to the server via a page form or application programming interface. In other words, the page form contains customer information and corresponding business content. The business side generates a customer tag approval request based on the information on the page form. This request contains information such as customer tags, tag names, and business definition text. Therefore, user consent or authorization is required before executing step S210. For example, a request to obtain user information can be sent to the user before step S210. If the user consents or authorizes the acquisition of user information, step S210 is executed.

[0043] In some embodiments, before step S210, a corresponding operation entry point can be provided to the user, allowing the user to choose to agree to or reject the automated decision result. That is, before collecting and processing / making decisions on user information, the user can provide their consent or rejection instruction through the corresponding operation entry point. If the user agrees to the processing / decision, the user information is collected and processed / decision is performed to facilitate the subsequent execution of step S210. If the user rejects the processing / decision, the expert decision-making process is initiated.

[0044] In some embodiments, in step S210, the customer tag can be "high-net-worth customer," "customer with large transfers in the past 7 days," or "total credit card spending of an individual customer in the past month," etc. The customer tag includes at least one of the following: tag name, business definition text, tag type, and customer type. Further, the target customer type refers to the customer type most likely to correspond to the customer tag inferred from the tag name and business definition text.

[0045] The business side can perform structuring and data preprocessing operations on the page form information to obtain information such as the tag name and business definition text of customer tags, and generate customer tag approval requests. Data preprocessing operations can include: checking whether there are any missing fields in the structured data and whether the field format meets the requirements. For data with missing fields or incorrect field formats, an error code is generated and returned to the business side, ending the tag approval process. Data preprocessing operations can also include: cleaning the structured data and removing special characters. Furthermore, the structured data can include: tag names, business definition text, customer type, and tag type generated based on the page form information. For example, the tag name generated based on the page form information could be "Number of Credit Card Transactions in the Last Month," the business definition text could be "Total Number of Credit Card Transactions for Individual Customers in the Last Month," the customer type could be "Individual," and the tag type could be "Transaction Behavior."

[0046] In some embodiments, in step S230, a large model (e.g., a large-scale language model) can be used to perform cue word inference on key information in the business definition text in order to calculate a completeness score based on the inference results. For example, key information is first extracted from the business definition text. Key information includes time information, customer information, transaction behavior information, and constraint information. Time information indicates information related to transaction time in the business definition text, such as the last 7 days; customer information can indicate customer type, such as individual customer or corporate customer; transaction behavior information can indicate customer transaction behavior, such as large-amount transfer; constraint information can indicate the constraints corresponding to customer transaction behavior, such as a single transfer amount exceeding a certain amount is considered a large-amount transfer; key information can also include the source of transaction behavior information, i.e., data source information, such as the credit card system, which indicates that the customer's transaction behavior information comes from the credit card system. The data source information can also be a payment system, a transfer system, etc.

[0047] The extracted key information is input into the large model, which is then guided by prompts to perform contextual reasoning on the key information. These prompts can be derived from prompt templates, such as "Please check if the key information includes all of the following: time information, customer information, transaction behavior information, and constraint information, to verify whether the elements of the key information in the business definition text are complete." Alternatively, the prompt template could be "Please check if the time information matches the transaction behavior information, if the transaction behavior information matches the constraint information, and if the transaction behavior information matches the data source information, etc." Furthermore, for example, if the time information is the historical cumulative transaction amount, but the transaction behavior information reveals a single transaction amount, then the time information and transaction behavior... If the transaction behavior information indicates a high-frequency transaction with a first transaction frequency, and the constraint information indicates a high-frequency transaction with a higher transaction frequency than the second transaction frequency, then the transaction behavior information and constraint information do not match. If the transaction behavior information represents a transfer transaction, but the data source information represents a credit card system, then the transaction behavior information and data source information do not match. The prompt template can also be "Please determine whether there is a logical contradiction or ambiguity in the business content represented by the key information." Furthermore, logical contradictions can be a mismatch between the target customer type and the business content, or a mismatch between the time information and the transaction behavior information, etc.

[0048] Based on the above prompt templates, prompts can be derived. For example, a prompt could be: "Please verify whether the key information includes all of the following: time information, customer information, transaction behavior information, and constraint information. Analyze the logical relationships between these information, determine if any information is missing, and calculate the completeness score of the customer tag based on the completeness and logical consistency of this information." Specifically, information completeness can be obtained by calculating the ratio of the actual amount of information contained in the key information to the total amount of information that the key information needs to include. The scores for information completeness and logical consistency are then weighted to obtain the completeness score of the customer tag.

[0049] Furthermore, the logical consistency score of the business content can be calculated as follows: The business content indicated in the business definition text is structured to obtain structured data. This structured data includes time information, customer information, transaction behavior information, and data source information. For example, the structured data might be: "Time information: last 30 days, Customer information: individual customers, Transaction behavior information: credit card spending, Data source information: credit card system." By verifying whether the credit card spending occurred within the last 30 days, the matching degree between the time information and transaction behavior information can be calculated. By verifying whether the credit card spending is from an individual customer, the matching degree between customer information and transaction behavior information can be calculated. By verifying whether the credit card spending data originates from the credit card system, the matching degree between transaction behavior information and data source information can be calculated. These matching degrees are then weighted to obtain the logical consistency score. If the logical consistency score is low, it indicates a problem with the business logic of the business definition text. In this case, the customer tags can be returned to the business side with corresponding tag modification suggestions.

[0050] In some embodiments, in step S250, if the overall score is less than or equal to a preset threshold, a correction suggestion is generated based on a preset correction strategy, and the correction suggestion and customer tag are sent to the business end. For customer tags with particularly low overall scores, the customer tags are sent to the approval platform for manual approval.

[0051] Furthermore, the preset correction strategies include label name correction strategies, customer type correction strategies, and business definition text correction strategies. A label name correction strategy could be: obtaining the label template of a historical customer label based on the historical customer label with the highest similarity to the customer label, and correcting the customer label based on this label template; a label name correction strategy could also be: generating correction suggestions based on the differences between the label template corresponding to the customer label and the customer label itself. For example, if the customer label is "large-amount transfer customer," and the label template corresponding to this customer label also includes time information, it is recommended to change the customer label to a label similar to "large-amount transfer customer in the last 7 days"; a label name correction strategy could also be: modifying prohibited words in the customer label to healthy words based on prohibited words in a preset thesaurus.

[0052] Customer type correction strategies can be as follows: First, correct the customer type based on entity keywords in the business definition text. For example, if entity keywords such as "enterprise," "company," and "legal person" frequently appear in the business definition text, and the target customer type is determined to be an enterprise through reasoning, while the customer type in the customer tag is an individual, then the customer type in the customer tag needs to be changed to an enterprise. Second, correct the customer type based on the approval results of some historical customer tags that are highly similar to the customer tag. For example, if the customer type in the approval results of these historical customer tags is all individuals, then it is recommended to change the customer type of that customer tag to an individual.

[0053] Tag type correction strategies can be as follows: Based on the business scenario indicated by the business definition text, correct the tag type in the customer tag. For example, if the business scenario indicates that the department performing the corresponding business is the risk control department, it is recommended to add a risk tag if there is no risk tag in the customer tag. Tag type correction strategies can also be as follows: Based on the tag type distribution of historical customer tags with high semantic similarity, correct the tag type of the customer tag. For example, if the tag type of most historical customer tags is transaction behavior, while the tag type of this customer tag is marketing, then the tag type of this customer tag can be changed to transaction behavior.

[0054] Business definition text correction strategies can include: providing corresponding modification suggestions based on completeness scores. For example, by analyzing the completeness of the business definition text and the logical consistency of the business content, missing elements in the business definition text can be identified, such as missing time information. In this case, the correction suggestion could be to add time information to the customer tag. Alternatively, the strategy could be: determining the tag template corresponding to the tag type of the customer tag, and using this template as a basis to add missing information to the customer tag or correct errors in the customer tag. Another strategy could be: obtaining the historical business definition text of historical customer tags that are similar in tag type to the customer tag, extracting high-frequency words from the historical business definition text, and replacing words in the customer tag with words that have similar meanings to these high-frequency words to correct the customer tag.

[0055] According to embodiments of this application, by reasoning about key information in the business definition text and calculating the completeness score of customer tags, the degree of matching between the target customer type and the business definition text can be quantified. This allows for the determination of whether the business definition text and customer tags are compatible, reducing the problem of customer tags not accurately representing the business definition text and thus improving the accuracy of tag approval. Based on the tag name and the business definition text, a first probability is calculated that the customer type represented by the customer tag is the target customer type. Combining the first probability, it is determined whether the customer tag matches its corresponding business content. Compared to the method of string matching customer tags with historical customer tags in the tag library, this method can reduce misjudgments caused by string matching. This application combines the business definition text to obtain the target customer type that is closest to the customer tag and its corresponding probability, which can more accurately quantify the customer type of the customer tag and thus ensure that the approved customer tag can accurately reveal the customer type indicated by the tag.

[0056] Figure 3 The illustration shows a schematic diagram of a comprehensive score calculation method according to an embodiment of this application.

[0057] like Figure 3 As shown, the first semantic vector 301 of the business definition text is obtained and input into the tag type classifier 302 to obtain the second probability 303 that the customer tag's tag type is the target tag type; the first probability 304, the second probability 303, and the completeness score 305 are weighted to obtain the comprehensive score 306. Further, the target tag type refers to the tag type most likely to correspond to the customer tag inferred from the business definition text.

[0058] Specifically, a large language model is used to perform semantic parsing on the business definition text, such as word segmentation, part-of-speech tagging, and contextual semantic parsing, to obtain the first semantic vector of the business definition text. The first semantic vector contains key information and some verb phrases. The time window information and the format of the numerical information involved in the first semantic vector are uniform. The verb phrases in the first semantic vector can represent the business processing method. For example, the verb phrase "monitoring" indicates that the main processing method of the business may be monitoring certain indicators, and the verb phrase "filtering" indicates that the business mainly filters certain data. The first semantic vector is input into a label type classifier to obtain the probability of the first semantic vector corresponding to each preset label type. The preset label types can include risk, transaction behavior, statistics, compliance monitoring, and marketing. The preset label types with the highest or relatively high probabilities are taken as the label types corresponding to the first semantic vector. In other words, by performing semantic parsing on the business definition text, the target label type corresponding to the customer label can be determined. The target label type refers to the preset label types with the highest or relatively high probabilities. For example, if a customer is labeled as a "high-risk customer", the probability that this customer label belongs to both the risk category and the compliance monitoring category is greater than the first preset probability. Therefore, the target label type corresponding to "high-risk customer" is the risk category and the compliance monitoring category. The target label type of a customer label can also be only one type. For example, if "the number of transfers by individual customers in the past 7 days" has the highest probability of belonging to the transaction behavior category, then the target label type of this customer label is the transaction behavior category.

[0059] Furthermore, by parsing the first semantic vector, business scenario information can be obtained. This information indicates the department that performs the corresponding business. For example, if the department performing the business indicated by the first business definition text is the risk control department, then a relatively high weight can be assigned to the probability corresponding to risk-related tags, and a relatively low weight can be assigned to the probability corresponding to other types of tags. If the department performing the business indicated by the second business definition text is the marketing department, then in the process of obtaining the target tag type, a relatively high weight can be assigned to the probability corresponding to the marketing tag to improve the accuracy of the target tag type.

[0060] The customer tag approval request can also retrieve the tag type based on the page form information. The similarity between this tag type and the target tag type is calculated. If the similarity is low, it indicates that the customer tag classification is inaccurate, and the customer tag can be sent to the approval platform for manual review. If the probability that the customer tag belongs to the target tag type is less than the second preset probability, it indicates that the customer tag classification is inaccurate, or that the customer tag does not accurately reflect its tag type. In this case, the customer tag can be sent to the approval platform for manual review, where the first preset probability is less than the second preset probability.

[0061] According to the embodiments of this application, the target tag type that is closest to the customer tag type is determined based on the first semantic vector, which can reduce the bias of human subjective judgment, quickly screen out tags to be approved with large semantic deviations, and improve approval efficiency and accuracy.

[0062] In some embodiments, an indexing method is used to retrieve historical tag names in the tag library that are similar to the customer tag name, and the semantic similarity between the historical tag name and the current tag name is calculated. If the semantic similarity is greater than a first preset semantic similarity, it indicates that the tag name is highly similar to the historical tag name, and the tag name is duplicated. In other words, the customer tag has a duplicate part with the historical customer tag in the tag library, and the customer tag can be returned to the business end for modification. If the semantic similarity is less than a second preset semantic similarity, it indicates that the customer tag does not duplicate the historical customer tag, and the customer tag can be saved to the tag library. If the semantic similarity is greater than the second preset semantic similarity and less than the first preset semantic similarity, the following suggestion can be generated: merge the historical customer tag corresponding to the semantic similarity with the current customer tag to generate a new customer tag, and return the suggestion and the customer tag to the business end. The semantic similarity, first probability, second probability, and completeness score are weighted to obtain a comprehensive score.

[0063] In some embodiments, before weighting the first probability, the second probability, and the completeness score, the method includes: matching the target label type with a weight adjustment rule to adjust the weights corresponding to the first probability, the second probability, and the completeness score; wherein the weight adjustment rule indicates the mapping relationship between the importance of the target customer type in the target business definition text and the weight of the first probability, the mapping relationship between the importance of the target label type in the target business definition text and the weight of the second probability, and the mapping relationship between the importance of the completeness of the customer label in the target business definition text and the weight of the completeness score, the target business definition text representing the business definition text corresponding to the target label type, and the target customer type and the target label type changing with the change of the customer label.

[0064] In some embodiments, different customer tags may correspond to different target customer types and different target tag types. Therefore, neither the target customer type nor the target tag type is constant. Based on the weight adjustment rules, for different customer tags, the weight can be assigned to the first probability corresponding to the target customer type by judging the importance of the corresponding target customer type in the target business definition text corresponding to the customer tag. Alternatively, the weight can be assigned to the second probability corresponding to the target tag type by judging the importance of the corresponding target tag type in the target business definition text. Furthermore, the weight can be assigned to the integrity score by judging the importance of the integrity of the customer tag in the target business definition text.

[0065] The first probability, second probability, and completeness scores are weighted using the above methods to obtain a comprehensive score. The comprehensive score indicates the degree of matching between the customer tag and multiple factors, such as the target customer type, target tag type, completeness of key information, and logical consistency of business content. A low comprehensive score indicates that the customer type in the customer tag may not be consistent with the target customer type, and / or the tag type in the customer tag may not be consistent with the target tag type, and / or there is a large gap between the business definition text of the customer tag and the target business definition text. All of these indicate that the customer tag cannot accurately reflect its actual business content and cannot pass the approval.

[0066] According to the embodiments of this application, based on the target tag type, corresponding weights are assigned to the first probability, the second probability, and the integrity score. Different types of customer tags have different focuses in their corresponding business definition texts, that is, factors such as customer type, tag type, and the integrity of customer tags have different importance in different business definition texts. Therefore, for different types of customer tags, the weights corresponding to the first probability, the second probability, and the integrity score are also different. This method of dynamically adjusting weights can make the approval results closer to actual business needs, reduce the false interception rate of high-risk tags, and improve the accuracy of approval.

[0067] Figure 4 The illustration shows a schematic diagram of a weight adjustment method according to an embodiment of this application.

[0068] In some embodiments, matching target label types with weight adjustment rules to adjust the weights corresponding to the first probability, the second probability, and the integrity score includes: if there are multiple target label types, matching each target label type with the weight adjustment rules to obtain a weight adjustment strategy corresponding to each target label type; determining the initial weights of the first probability, the second probability, and the integrity score under the corresponding weight adjustment strategy based on each weight adjustment strategy; and calculating the mean of the initial weights of the first probability, the second probability, and the integrity score under each weight adjustment strategy to obtain the weights corresponding to the first probability, the second probability, and the integrity score.

[0069] like Figure 4 As shown, assuming there are N target label types, where N is an integer greater than 1, i.e., the target label types include the first target label type 401, the second target label type 402, ..., and the Nth target label type 403, these N target label types are matched with weight adjustment rule 404 respectively to obtain the first weight adjustment strategy 405, the second weight adjustment strategy 406, ..., and the Nth weight adjustment strategy 407. The first weight adjustment strategy is the weight adjustment strategy corresponding to the first target label type, the second weight adjustment strategy is the weight adjustment strategy corresponding to the second target label type, ..., and the Nth weight adjustment strategy is the weight adjustment strategy corresponding to the Nth target label type. The first weight adjustment strategy indicates one way to allocate the initial weights corresponding to the first probability, the second probability, and the integrity score, respectively, while the second weight adjustment strategy indicates another way to allocate the initial weights corresponding to the first probability, the second probability, and the integrity score, respectively. The allocation method, specifically the Nth weight adjustment strategy, indicates another way to allocate the initial weights corresponding to the first probability, the second probability, and the completeness score. For example, if the first target label type is a risk-related label, the matched first weight adjustment strategy is to appropriately increase the initial weights corresponding to the first probability and the completeness score, and decrease the initial weight of semantic similarity. If the second target label type is a marketing-related label, the second weight adjustment strategy could be to increase the initial weights corresponding to the completeness score and semantic similarity, and decrease the initial weight of the first probability. If the third target label type is a compliance monitoring label, the third weight adjustment strategy could be to increase the initial weights corresponding to the first probability and the completeness score, and decrease the initial weight of semantic similarity. If the Nth target label type is a transaction behavior label, the Nth weight adjustment strategy could be to keep the initial weights of the first probability, the second probability, semantic similarity, and the completeness score unchanged.

[0070] In other words, based on the first weight adjustment strategy, the first initial weights 408 corresponding to the first probability, the second probability, and the integrity score can be obtained. Based on the Nth weight adjustment strategy, the Nth initial weights 409 corresponding to the first probability, the second probability, and the integrity score can be obtained. After obtaining the initial weights corresponding to the first probability, the second probability, and the integrity score under each weight adjustment strategy, operation S410 is executed, that is, the mean of each initial weight corresponding to the first probability, the mean of each initial weight corresponding to the second probability, and the mean of each initial weight corresponding to the integrity score are calculated respectively. In other words, based on the first initial weight to the Nth initial weight of the first probability, the mean of the initial weights of the first probability is calculated, and this mean is used as the final weight of the first probability; based on the first initial weight to the Nth initial weight of the second probability, the mean of the initial weights of the second probability is calculated, and this mean is used as the final weight of the second probability; based on the first initial weight to the Nth initial weight of the integrity score, the mean of the initial weights of the integrity score is calculated, and this mean is used as the final weight of the integrity score; it is also possible to obtain the first initial weight to the Nth initial weight of semantic similarity, and then calculate the mean of the initial weights of semantic similarity, and use this mean as the final weight of semantic similarity.

[0071] According to the embodiments of this application, each target label type is matched with a weight adjustment rule to generate its own independent weight adjustment strategy. For any indicator (e.g., the first probability, the second probability, or the completeness score), the average weight under each weight adjustment strategy is taken as the final weight of the indicator. This can avoid the weight imbalance problem caused by the weight adjustment strategy of a single target label type, improve the balance of the adjusted weights, and ensure the accuracy and reliability of the weighted comprehensive score.

[0072] Figure 5 The illustration shows a schematic diagram of a method for determining a target customer type according to an embodiment of this application.

[0073] In some embodiments, based on the label name and business definition text, calculating the first probability that the customer type indicated by the customer label is the target customer type includes: obtaining the second semantic vector of the label name; performing a weighted fusion of the first semantic vector and the second semantic vector to obtain a fusion vector; inputting the fusion vector into a fully connected layer and performing a nonlinear transformation to obtain the first probability of each customer type corresponding to the customer label; sorting the first probabilities of each customer type in descending order, and taking the customer type ranked first as the target customer type, wherein the first probability of the customer type ranked first is the first probability of the target customer type.

[0074] like Figure 5As shown, a weighted fusion of the first semantic vector 501 and the second semantic vector 502 yields a fused vector 503. This fused vector is then input into a fully connected layer 504 to perform operation S510, i.e., a nonlinear transformation, to obtain the first probability corresponding to each customer type. Then, operation S520 is performed to sort these first probabilities, obtaining the first probability corresponding to the target customer type. Specifically, the tag name is input into the large language model, and contextual information from the tag name is captured through a multi-layer encoder to obtain a second semantic vector. The first semantic vector and the second semantic vector are concatenated to obtain a concatenated vector. The first semantic vector is used as the query vector, and the second semantic vector as the key vector. An attention score is calculated, and this attention score is used as the weight coefficient of the first semantic vector. Corresponding weight coefficients are assigned to the second semantic vector. Based on the respective weight coefficients of the first and second semantic vectors, a weighted fusion of the first and second semantic vectors is performed to obtain the fused vector. The weight coefficients corresponding to the second semantic vector and the attention score are... The sum is 1. The fusion vector is input into the fully connected layer, and through activation functions and nonlinear transformations, it is mapped to the classification space. Finally, a normalization function is used to calculate the probability of each customer type (individual, enterprise, legal person, etc.) corresponding to the fusion vector. Furthermore, semantic parsing can be performed on the first semantic vector to identify whether keywords such as "enterprise," "individual," "legal person," and "customer" exist in the business definition text. If a keyword exists, the first probability corresponding to the customer type indicated by that keyword is biased and adjusted. For example, if the keyword "individual" exists, the first probability of the customer type "individual" is increased by a preset probability value to increase the probability of that customer type. The customer type with the higher or highest first probability is selected as the target customer type.

[0075] According to the embodiments of this application, by fusing the first semantic vector and the second semantic vector, deep correlation features between the two vectors can be mined. Based on these deep correlation features, the target customer type is finally determined. Based on the first probability corresponding to the target customer type, it is determined whether the customer tag can accurately reveal the target customer type, thereby effectively reducing the problem of mismatch between tags and customer types and improving the accuracy of customer tag classification and approval efficiency.

[0076] Figure 6 The illustration shows a schematic diagram of another label approval method according to an embodiment of this application.

[0077] In some embodiments, method 200 further includes: obtaining a tag template 602 corresponding to a customer tag according to a target tag type 601, and calculating a first matching degree 604 between the customer tag 603 and the tag template; if the first matching degree is greater than a first preset matching degree, comparing the words in the customer tag with a second matching degree in a preset thesaurus, wherein the words in the preset thesaurus are prohibited words; if the second matching degree is greater than a second preset matching degree, returning the customer tag to the business end.

[0078] like Figure 6 As shown, after obtaining the first matching degree 604 based on the target tag type, operation S610 is executed to determine whether the first matching degree is greater than the first preset matching degree. If so, operation S620 is executed, that is, the customer tag is compared with the preset thesaurus to obtain the second matching degree. Then, operation S630 is executed to determine whether the second matching degree is greater than the second preset matching degree. If so, operation S630 is executed to return the customer tag to the business end. If the second matching degree is not greater than the second preset matching degree, operation S650 is executed, that is, the approval operation continues according to method 200. If the first matching degree is less than or equal to the first preset matching degree, operation S630 is executed. Further, the tag template includes parts representing customer information, parts representing transaction behavior information, and parts representing business content, etc.

[0079] According to the embodiments of this application, customer tags that match the corresponding tag template can be filtered out based on the first matching degree between the customer tag and the corresponding tag template, which is conducive to standardizing the form of customer tags. For customer tags with standardized forms, it is possible to detect whether they contain prohibited words, which can reduce the business statistics deviation and compliance risks caused by non-compliant tags. In this way, the cost of manual review can be greatly reduced and the review efficiency can be improved.

[0080] In some embodiments, method 200 further includes: extracting the customer type of the customer tag from the customer tag approval request, calculating the similarity between the target customer type and the extracted customer type; calculating the confidence level of the target customer type based on the similarity; and if the confidence level is less than a first preset confidence level, sending the customer tag to the approval platform for review.

[0081] Furthermore, if the confidence level of the target customer type is greater than the second preset confidence level, the customer tag is approved and saved to the tag library. If the confidence level is greater than the first preset confidence level but less than the second preset confidence level, the customer tag is sent to the approval platform for review.

[0082] In some embodiments, for new business definition text, i.e., business content appearing for the first time, the customer tags in the customer tag approval request are directly sent to the approval platform for manual review to ensure approval accuracy. If the tag name contains preset words such as "test" or "temporary," the customer tag is directly sent to the approval platform.

[0083] In some embodiments, the approval platform can display information such as customer tags, tag names, and business definition text contained in the customer tag approval request. It can also display the target customer type and its first probability, the target tag type and its second probability, semantic similarity and its corresponding historical customer tags, completeness score, confidence levels of the first and second probabilities, and approval results for similar customer tags, as well as correction suggestions returned to the business side. The platform also includes a manual approval window to re-approve customer tags with low confidence or vague descriptions. After manual approval, approved customer tags can be directly saved to the tag library, while unapproved tags are returned to the business side with reasons for return, such as suggested modifications. Furthermore, the operation data during the manual approval process can be saved in a log for later verification.

[0084] In some embodiments, information displayed on the approval platform is used as training samples to train the large model. For example, approval information of customer tags with low confidence but approved manually can be used as the first training sample to fine-tune the large model; approval information from the previous 24 hours can also be used as the second training sample for incremental training. The approval information includes the final approval conclusion, comprehensive score, target customer type and its first probability, target tag type and its second probability, semantic similarity and corresponding historical customer tags, completeness score, and correction suggestions. Furthermore, based on the information displayed on the approval platform, newly appearing customer types can be added to the tag library, and frequently corrected content can be added to the corresponding tag templates. Further, the frequency of manual approval and the frequency of tag return to the business end can be calculated periodically, and a large model optimization scheme can be generated for subsequent large model optimization.

[0085] According to the embodiments of this application, sending customer tags with low confidence levels of target customer types to the approval platform for review can not only significantly reduce the workload of manual approval in high-confidence scenarios, but also effectively reduce business mismatch problems caused by customer type judgment bias. Therefore, this method can improve approval accuracy while ensuring approval efficiency.

[0086] In some embodiments, the specific approval process for customer tags is as follows: A customer tag approval request is received from the business terminal; the customer tag, its tag name, business definition text, customer type, and tag type are extracted from the request; the semantic similarity between the historical tag name and the current tag name is calculated to determine if the current customer tag is duplicated; a first matching degree between the customer tag and the corresponding tag template, and a second matching degree between the customer tag and a preset thesaurus are calculated. The first matching degree indicates the degree of matching between the customer tag and the corresponding tag template, and the second matching degree indicates the degree to which the customer tag contains prohibited words. The reasonableness of the tag name is evaluated by combining the semantic similarity, the first matching degree, and the second matching degree; a reasonableness score and its confidence level are calculated, and problems with the tag name are recorded. The tag name and business definition text are input into a large model to predict the most likely customer type represented by the customer tag (target customer type); a first probability and its confidence level of the target customer type are calculated, and problems with the customer type are recorded. The first semantic vector is input into a tag type classifier to obtain the target tag type and its corresponding second probability and confidence level, and problems with the tag type are recorded. By reasoning through the business definition text, a completeness score for the customer tag is obtained, and problems with the completeness and consistency of the business logic of the customer tag are recorded.

[0087] A comprehensive score is obtained by combining the reasonableness score, the first probability, the second probability, and the completeness score. Customer tags with a comprehensive score greater than a preset threshold are approved. For customer tags that fail to pass approval, if the comprehensive score is within a certain range, a correction suggestion is generated and returned to the business end along with the correction suggestion. If the business end accepts the correction suggestion and makes corrections, it can receive the corrected customer tag approval request from the business end again and re-perform the approval operation. If the business end does not make corrections, the correction suggestion and customer tag are sent to the approval platform for manual review. The large model is optimized and the tag library is updated based on the approval information from the manual review, and the approved customer tags are stored in the tag library. The total confidence score is obtained based on the confidence scores corresponding to the reasonableness score, the first probability, the second probability, and the completeness score. If the total confidence score is low, the customer tag and the recorded issues are sent to the approval platform for manual review. Furthermore, the approval platform can display the approval process and results at each stage, and also allow manual reviewers to directly store customer tags that they believe can be approved into the tag library through the corresponding interface, or return the customer tags to the business end with correction suggestions.

[0088] Based on the above-described customer label approval method, embodiments of this application also provide a customer label approval device. The following will be combined with... Figure 7The device is described in detail.

[0089] Figure 7 A schematic block diagram of a customer label approval device according to an embodiment of this application is shown.

[0090] like Figure 7 As shown, the customer tag approval device 700 of this embodiment includes an information extraction module 710, a first probability acquisition module 720, an integrity score acquisition module 730, a comprehensive score acquisition module 740, and a tag approval module 750.

[0091] The information extraction module 710 is used to extract the tag name and business definition text corresponding to the customer tag from the customer tag approval request sent by the business terminal. The business definition text indicates the business content corresponding to the customer tag. In one embodiment, the information extraction module 710 can be used to perform step S210 described above, which will not be repeated here.

[0092] The first probability acquisition module 720 is used to calculate the first probability that the customer type indicated by the customer tag is the target customer type based on the tag name and business definition text. In one embodiment, the first probability acquisition module 720 can be used to perform step S220 described above, which will not be repeated here.

[0093] The integrity score acquisition module 730 is used to perform cue word inference on key information extracted from the business definition text, and calculate the integrity score of the customer tag based on the inference result. The integrity score indicates the integrity of the key information and the logical consistency of the business content. Logical consistency indicates the degree of matching between the target customer type and the business content, and key information indicates the contextual information of the business content. In one embodiment, the integrity score acquisition module 730 can be used to execute step S230 described above, which will not be repeated here.

[0094] The comprehensive score acquisition module 740 is used to weight the first probability and integrity scores to obtain a comprehensive score. In one embodiment, the comprehensive score acquisition module 740 can be used to perform step S240 described above, which will not be repeated here.

[0095] The tag approval module 750 is used to approve customer tags whose comprehensive score is greater than a preset threshold. In one embodiment, the tag approval module 750 can be used to perform step S250 described above, which will not be repeated here.

[0096] According to an embodiment of this application, label approval can be performed using the customer label approval device 700, which can improve the accuracy of label approval.

[0097] In some embodiments, the customer tag approval device 700 is further configured to: obtain a first semantic vector of the business definition text and input it into a tag type classifier to obtain a second probability that the tag type of the customer tag is the target tag type; and weight the first probability, the second probability and the integrity score to obtain a comprehensive score.

[0098] In some embodiments, the customer tag approval device 700 is further configured to: match the target tag type with the weight adjustment rules before weighting the first probability, the second probability, and the completeness score, so as to adjust the weights corresponding to the first probability, the second probability, and the completeness score respectively; wherein, the weight adjustment rules indicate the mapping relationship between the importance of the target customer type in the target business definition text and the weight of the first probability, the mapping relationship between the importance of the target tag type in the target business definition text and the weight of the second probability, and the mapping relationship between the importance of the customer tag's completeness in the target business definition text and the weight of the completeness score, the target business definition text representing the business definition text corresponding to the target tag type, and the target customer type and the target tag type changing with the change of the customer tag respectively.

[0099] In some embodiments, the customer label approval device 700 is further configured to: if there are multiple target label types, match each target label type with a weight adjustment rule to obtain a weight adjustment strategy corresponding to each target label type; based on each weight adjustment strategy, determine the initial weights of the first probability, the second probability, and the integrity score under the corresponding weight adjustment strategy; calculate the mean of the initial weights of the first probability, the second probability, and the integrity score under each weight adjustment strategy to obtain the weights corresponding to the first probability, the second probability, and the integrity score.

[0100] In some embodiments, the first probability acquisition module 720 is further configured to: acquire a second semantic vector of the tag name; perform weighted fusion of the first semantic vector and the second semantic vector to obtain a fusion vector; input the fusion vector into a fully connected layer and perform a nonlinear transformation to obtain a first probability of each customer type corresponding to the customer tag; sort the first probabilities of each customer type in descending order, and take the customer type ranked first as the target customer type, wherein the first probability of the customer type ranked first is the first probability of the target customer type.

[0101] In some embodiments, the customer tag approval device 700 is further configured to: obtain a tag template corresponding to the customer tag according to the target tag type, and calculate a first matching degree between the customer tag and the tag template; if the first matching degree is greater than a first preset matching degree, compare the words in the customer tag with a second matching degree of a preset thesaurus, wherein the words in the preset thesaurus are prohibited words; if the second matching degree is greater than a second preset matching degree, return the customer tag to the business end.

[0102] In some embodiments, the customer tag approval device 700 is further configured to: extract the customer type of the customer tag from the customer tag approval request, calculate the similarity between the target customer type and the extracted customer type; calculate the confidence level of the target customer type based on the similarity; and if the confidence level is less than a first preset confidence level, send the customer tag to the approval platform for review.

[0103] According to embodiments of this application, any multiple modules among the information extraction module 710, the first probability acquisition module 720, the integrity score acquisition module 730, the comprehensive score acquisition module 740, and the tag approval module 750 can be merged into one module, or any one of these modules can be split into multiple modules. Alternatively, at least some of the functions of one or more of these modules can be combined with at least some of the functions of other modules and implemented in one module. According to embodiments of this application, at least one of the information extraction module 710, the first probability acquisition module 720, the integrity score acquisition module 730, the comprehensive score acquisition module 740, and the tag approval module 750 can be at least partially implemented as hardware circuits, such as field-programmable gate arrays, programmable logic arrays, systems-on-a-chip, systems-on-a-substrate, systems-on-package, application-specific integrated circuits, or other reasonable means of integrating or packaging circuits, or implemented in hardware or firmware, or in any one of software, hardware, and firmware implementations, or in a suitable combination of any of these. Alternatively, at least one of the information extraction module 710, the first probability acquisition module 720, the integrity score acquisition module 730, the comprehensive score acquisition module 740, and the label approval module 750 can be at least partially implemented as a computer program module, which can perform corresponding functions when the computer program module is run.

[0104] Figure 8 A block diagram schematically illustrates an electronic device suitable for implementing a customer tag approval method according to an embodiment of this application.

[0105] like Figure 8As shown, an electronic device 800 according to an embodiment of this application includes a processor 801, which can perform various appropriate actions and processes according to a program stored in a read-only memory 802 or a program loaded from a storage portion 808 into a random access memory 803. The processor 801 may include, for example, a general-purpose microprocessor, an instruction set processor and / or an associated chipset and / or a dedicated microprocessor. The processor 801 may also include onboard memory for caching purposes. The processor 801 may include a single processing unit or multiple processing units for executing different steps of the method flow according to an embodiment of this application.

[0106] Random access memory 803 stores various programs and data required for the operation of electronic device 800. Processor 801, read-only memory 802, and random access memory 803 are interconnected via bus 804. Processor 801 executes various steps of the method flow according to embodiments of this application by executing programs stored in read-only memory 802 and / or random access memory 803. It should be noted that the programs may also be stored in one or more memories other than read-only memory 802 and random access memory 803. Processor 801 may also execute various steps of the method flow according to embodiments of this application by executing programs stored in said one or more memories.

[0107] According to embodiments of this application, the electronic device 800 may further include an input / output interface 805, which is also connected to a bus 804. The electronic device 800 may also include one or more of the following components connected to the input / output interface 805: an input section 806 including a keyboard, mouse, etc.; an output section 807 including a cathode ray tube, liquid crystal display, etc., and a speaker, etc.; a storage section 808 including a hard disk, etc.; and a communication section 809 including a network interface card, such as a local area network card, modem, etc. The communication section 809 performs communication processing via a network such as the Internet. A drive 810 is also connected to the input / output interface 805 as needed. A removable medium 811, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 810 as needed so that computer programs read from it can be installed into the storage section 808 as needed.

[0108] Embodiments of this application also provide a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or it may exist independently and not assembled into the device / apparatus / system. The computer-readable storage medium carries one or more programs, which, when executed, implement the method according to the embodiments of this application.

[0109] According to embodiments of this application, the computer-readable storage medium can be a non-volatile computer-readable storage medium, such as including but not limited to: portable computer disks, hard disks, random access memory, read-only memory, erasable programmable read-only memory, portable compact disk read-only memory, optical storage devices, magnetic storage devices, or any suitable combination thereof. In embodiments of this application, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. For example, according to embodiments of this application, the computer-readable storage medium may include the read-only memory 802 described above, and / or random access memory 803, and / or one or more memories other than read-only memory 802 and random access memory 803.

[0110] Embodiments of this application also include a computer program product comprising a computer program containing program code for performing the methods shown in the flowchart. When the computer program product is run on a computer system, the program code is used to cause the computer system to implement the methods provided in the embodiments of this application.

[0111] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and may be downloaded and installed via the communication section 809, and / or installed from a removable medium 811. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0112] In embodiments of this application, the computer program can be downloaded and installed from a network via communication section 809, and / or installed from removable medium 811. When the computer program is executed by processor 801, it performs the functions defined in the system of embodiments of this application. According to embodiments of this application, the systems, devices, apparatuses, modules, units, etc., described above can be implemented by computer program modules.

[0113] According to embodiments of this application, program code for executing the computer programs provided in the embodiments of this application can be written in any combination of one or more programming languages. Specifically, these computational programs can be implemented using high-level procedural and / or object-oriented programming languages, and / or assembly / machine languages. The program code can be executed entirely on the user's computing device, partially on the user's device, partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0114] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0115] Those skilled in the art will understand that the features described in the various embodiments of this application can be combined and / or combined in various ways, even if such combinations or combinations are not explicitly described in this application. In particular, the features described in the various embodiments of this application can be combined and / or combined in various ways without departing from the spirit and teachings of this application. All such combinations and / or combinations fall within the scope of this application.

Claims

1. A customer tag approval method, characterized in that, The method includes: Extract the tag name and business definition text corresponding to the customer tag from the customer tag approval request sent from the business side. The business definition text indicates the business content corresponding to the customer tag. Based on the tag name and the business definition text, calculate the first probability that the customer type indicated by the customer tag is the target customer type; The key information extracted from the business definition text is subjected to prompt word reasoning. Based on the reasoning results, the integrity score of the customer tag is calculated. The integrity score indicates the integrity of the key information and the logical consistency of the business content. The logical consistency indicates the degree of matching between the target customer type and the business content. The key information indicates the contextual information of the business content. The first probability and the integrity score are weighted to obtain a comprehensive score; Customer tags whose comprehensive score is greater than a preset threshold are used as approved customer tags.

2. The method according to claim 1, characterized in that, The method further includes: Obtain the first semantic vector of the business definition text and input it into the tag type classifier to obtain the second probability that the tag type of the customer tag is the target tag type; The first probability, the second probability, and the integrity score are weighted to obtain the comprehensive score.

3. The method according to claim 2, characterized in that, Before weighting the first probability, the second probability, and the integrity score, the following steps are included: The target label type is matched with the weight adjustment rules to adjust the weights corresponding to the first probability, the second probability, and the integrity score. The weight adjustment rules indicate the mapping relationship between the importance of the target customer type in the target business definition text and the weight of the first probability, the mapping relationship between the importance of the target tag type in the target business definition text and the weight of the second probability, and the mapping relationship between the importance of the completeness of the customer tag in the target business definition text and the weight of the completeness score. The target business definition text represents the business definition text corresponding to the target tag type, and the target customer type and the target tag type change with the change of the customer tag.

4. The method according to claim 3, characterized in that, The step of matching the target label type with the weight adjustment rules to adjust the weights corresponding to the first probability, the second probability, and the completeness score includes: If there are multiple target label types, then each target label type is matched with the weight adjustment rule to obtain the weight adjustment strategy corresponding to each target label type. Based on each weight adjustment strategy, determine the initial weights of the first probability, the second probability, and the integrity score under the corresponding weight adjustment strategy. Calculate the mean of the initial weights for the first probability, the mean of the initial weights for the second probability, and the mean of the initial weights for the integrity score under each weight adjustment strategy to obtain the weights corresponding to the first probability, the second probability, and the integrity score.

5. The method according to claim 2, characterized in that, The step of calculating the first probability that the customer type indicated by the customer tag is the target customer type based on the tag name and the business definition text includes: Obtain the second semantic vector of the tag name, and perform a weighted fusion of the first semantic vector and the second semantic vector to obtain a fusion vector; The fusion vector is input into a fully connected layer and subjected to a nonlinear transformation to obtain the first probability of each customer type corresponding to the customer tag. The first probabilities of each customer type are sorted in descending order, and the customer type with the highest probability is taken as the target customer type, wherein the first probability of the customer type with the highest probability is the first probability of the target customer type.

6. The method according to claim 2, characterized in that, The method further includes: Based on the target tag type, obtain the tag template corresponding to the customer tag, and calculate the first matching degree between the customer tag and the tag template; If the first matching degree is greater than the first preset matching degree, then compare the words in the customer tag with the second matching degree of the preset word library, wherein the words in the preset word library are prohibited words; If the second matching degree is greater than the second preset matching degree, the customer tag is returned to the business terminal.

7. The method according to claim 1, characterized in that, The method further includes: Extract the customer type of the customer tag from the customer tag approval request, and calculate the similarity between the target customer type and the extracted customer type; Based on the similarity, calculate the confidence level of the target customer type; If the confidence level is less than the first preset confidence level, the customer tag will be sent to the approval platform for review.

8. A customer label approval device, characterized in that, The device includes: The information extraction module is used to extract the tag name and business definition text corresponding to the customer tag from the customer tag approval request sent by the business side. The business definition text indicates the business content corresponding to the customer tag. The first probability acquisition module is used to calculate the first probability that the customer type indicated by the customer tag is the target customer type based on the tag name and the business definition text; The integrity score acquisition module is used to perform cue word reasoning on key information extracted from the business definition text, and calculate the integrity score of the customer tag based on the reasoning result. The integrity score indicates the integrity of the key information and the logical consistency of the business content. The logical consistency indicates the degree of matching between the target customer type and the business content. The key information indicates the contextual information of the business content. The comprehensive score acquisition module is used to weight the first probability and the integrity score to obtain a comprehensive score; The tag approval module is used to approve customer tags whose comprehensive score is greater than a preset threshold.

9. An electronic device, comprising: One or more processors; Memory, used to store one or more computer programs. The characteristic feature is that the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program or instructions stored thereon, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 7.

11. A computer program product, comprising a computer program or instructions, characterized in that, When the computer program or instructions are executed by a processor, they implement the steps of the method according to any one of claims 1 to 7.

Images