Enterprise business processing method and device based on virtual mapping and data desensitization, equipment and medium

Abstract

The application discloses an enterprise business processing method and device based on virtual mapping and data desensitization, equipment and medium, relates to the field of artificial intelligence technology, including: determining the target business scene corresponding to the business request uploaded by the user, and determining the target data query template corresponding to the target business scene; in the target ontology type layer of the customer relationship management system of the enterprise, the focus entity in the target data query template is taken as the starting point to obtain the target entity; based on the pre-configured virtual mapping, the target business domain service corresponding to the target entity is called to analyze the associated instance of the target entity, and the business data is determined according to the target entity and the associated instance; the preset sensitive field in the business data is desensitized, and the target large model is used to process the business request based on the desensitized business data to obtain the target processing result. The application can realize the standardized acquisition of business data in the customer relationship management system of the enterprise under the premise of guaranteeing the safety of sensitive data.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to enterprise business processing methods, apparatus, equipment and media based on virtual mapping and data desensitization. Background Technology

[0002] With the rapid development of cross-border business, CRM (Customer Relationship Management) systems have accumulated massive amounts of structured data across multiple business domains, including customers, users, leads, opportunities, contacts, follow-up activities, orders, and GMV (Gross Merchandise Volume) metrics. To improve sales and operational efficiency, large language modeling capabilities are now being introduced into the CRM system, providing intelligent functions such as lead scoring, customer health scores, opportunity win rate prediction, business insights, and natural language question answering.

[0003] Since large language models themselves do not have access to real-time business data of enterprises, a retrieval-enhanced generation method must be adopted. This involves first retrieving relevant data from the business system, injecting it as contextual clues, and then passing it on to the model for inference.

[0004] The current common practice for implementing intelligent functions in CRM systems is to write separate code for each AI (Artificial Intelligence) scenario to call domain services and assemble context, then feed the context along with the original fields into a large model to generate results, and finally perform post-processing entity ID existence checks on the returned results. This approach has the following problems: highly repetitive and fragmented data retrieval logic, with the same relationship repeatedly expressed in multiple scenarios, leading to a linear increase in maintenance costs with the number of scenarios; weak anti-illusion capabilities, failing to identify errors such as illegal entity types, invalid relationships, and cardinality out-of-bounds errors, resulting in low output reliability; easy leakage of sensitive data, with complete phone numbers, email addresses, and other fields directly fed into the model, posing compliance risks; and poor scalability, requiring the modification of all relevant scenarios to integrate new business domains, hindering smooth evolution.

[0005] In conclusion, how to achieve standardized acquisition of business data from an enterprise's customer relationship management system while ensuring the security of sensitive data, in order to process enterprise business, is a pressing technical problem that needs to be solved. Summary of the Invention

[0006] In view of this, the purpose of this invention is to provide a method, apparatus, device, and medium for enterprise business processing based on virtual mapping and data anonymization, which can achieve standardized acquisition of business data in an enterprise's customer relationship management system while ensuring the security of sensitive data, thereby processing enterprise business. The specific solution is as follows: Firstly, this application provides an enterprise business processing method based on virtual mapping and data anonymization, including: Obtain the business request uploaded by the user, and determine the target business scenario corresponding to the business request based on the preset scenario mapping rules, and determine the target data query template corresponding to the target business scenario; In the target ontology type layer corresponding to the enterprise's customer relationship management system, the corresponding target entity is obtained by traversing from the focus entity in the target data query template. The target ontology type layer includes the entity corresponding to the business concept in the customer relationship management system, the edge corresponding to the business relationship between each business concept, and the attributes of each business concept. Based on a pre-configured virtual mapping, the target business domain service corresponding to the target entity is invoked to parse the associated instance of the target entity, and business data is determined according to the target entity and the associated instance; the virtual mapping is used to maintain the mapping relationship between business concepts and the data acquisition methods corresponding to the business domain service; The preset sensitive fields in the business data are de-identified, and the target large model is used to process the business request based on the de-identified business data to obtain the corresponding target processing result.

[0007] Optionally, determining the target business scenario corresponding to the business request based on a preset scenario mapping rule includes: The business request is subjected to keyword extraction to obtain the corresponding target keywords, and the target keywords are matched with the scenario keywords in the preset scenario mapping rules; If the target keyword successfully matches any of the scenario keywords, then the business scenario corresponding to the scenario keyword is determined as the target business scenario; If the target keyword fails to match any of the scenario keywords in the preset scenario mapping rules, then a pre-selected scenario is obtained and the pre-selected scenario is determined as the target business scenario.

[0008] Optionally, in the target ontology type layer corresponding to the enterprise's customer relationship management system, the corresponding target entity is obtained by traversing from the focus entity in the target data query template, including: According to the business relationship path in the target data query template, a jump-by-jump traversal is performed in the target ontology type layer starting from the focus entity; Accordingly, the process of traversing the target ontology type layer hop by hop starting from the focus entity also includes: If the current traversal depth does not exceed the preset traversal depth threshold, then continue traversing the next-hop entity in the target ontology type layer according to the business relationship path; If the current traversal depth exceeds the preset traversal depth threshold, then the traversal stops, and each entity obtained through the traversal is determined as the target entity.

[0009] Optionally, the step of invoking the target business domain service corresponding to the target entity based on the pre-configured virtual mapping to parse the associated instance of the target entity, and determining business data based on the target entity and the associated instance, includes: The virtual mapping is queried according to the business concept corresponding to the target entity to determine the target business domain service corresponding to the target entity, and the target data acquisition method corresponding to the target business domain service is determined. The associated instance of the target entity is obtained based on the target data acquisition method. Using the target entity as a node, establish connections between different target entities with business relationships, label the business concepts corresponding to the nodes and the business relationships corresponding to the connections, and associate the target entity with the associated instance corresponding to the target entity to construct an entity neighborhood subgraph; The entity neighborhood subgraph is identified as the business data.

[0010] Optionally, the process of desensitizing preset sensitive fields in the business data includes: Identify preset sensitive fields in the business data; the preset sensitive fields include mobile phone number, email address, and ID number. The preset sensitive fields in the business data are de-identified to obtain the de-identified business data, and the preset sensitive fields are detected to be present in the de-identified business data. If the preset sensitive fields still exist in the de-identified business data, the call to the target large model will be blocked, and a preset error message will be returned. If the preset sensitive field is not present in the anonymized business data, the target large model is used to process the business request based on the anonymized business data.

[0011] Optionally, the step of using the target large model to process the business request based on the de-identified business data to obtain the corresponding target processing result includes: The anonymized business data and the business request are combined to form a target prompt word, and the target prompt word is sent to the target large model; The initial processing result of the incremental output of the target large model is obtained by using a server-side event stream method, and the initial processing result is verified to determine the target processing result based on the verification result.

[0012] Optionally, the step of verifying the initial processing result to determine the target processing result based on the verification result includes: Verify whether the entity referenced in the initial processing result exists in the target ontology type layer, whether the entity relationship in the initial processing result is declared in the target ontology type layer, whether the relationship cardinality corresponding to the entity relationship matches the relationship cardinality defined in the target ontology type layer, and verify whether the entity number corresponding to the entity referenced in the initial processing result exists through the target business domain service; If all entities and entity relationships referenced in the initial processing result pass the verification, then the initial processing result is directly determined as the target processing result; If there are entities and / or entity relationships that fail the verification in the initial processing result, then the entities and / or entity relationships that fail the verification in the initial processing result are removed, and the target processing result is determined based on the processing result after removal. If the entities and entity relationships referenced in the initial processing result fail the verification, then the entities and entity relationships referenced in the initial processing result are removed to obtain plain text data, and the plain text data is determined as the target processing result.

[0013] Secondly, this application provides an enterprise business processing apparatus based on virtual mapping and data anonymization, comprising: The data query template determination module is used to obtain the business request uploaded by the user, determine the target business scenario corresponding to the business request based on the preset scenario mapping rules, and determine the target data query template corresponding to the target business scenario. The target entity acquisition module is used to traverse the target ontology type layer corresponding to the enterprise's customer relationship management system, starting from the focus entity in the target data query template, to obtain the corresponding target entity; the target ontology type layer includes entities corresponding to business concepts in the customer relationship management system, edges corresponding to business relationships between the business concepts, and attributes of the business concepts; The business data determination module is used to call the target business domain service corresponding to the target entity based on a pre-configured virtual mapping to parse the associated instance of the target entity, and determine the business data according to the target entity and the associated instance; the virtual mapping is used to maintain the mapping relationship between business concepts and the data acquisition methods corresponding to the business domain service; The business processing module is used to de-identify preset sensitive fields in the business data, and use the target large model to process the business request based on the de-identified business data to obtain the corresponding target processing result.

[0014] Thirdly, this application provides an electronic device, comprising: Memory, used to store computer programs; A processor is used to execute the computer program to implement the aforementioned enterprise business processing method based on virtual mapping and data anonymization.

[0015] Fourthly, this application provides a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the aforementioned enterprise business processing method based on virtual mapping and data anonymization.

[0016] In this application, the user-uploaded business request is first obtained, and the target business scenario corresponding to the business request and the target data query template corresponding to the target business scenario are determined based on a preset scenario mapping rule. Then, in the target ontology type layer corresponding to the enterprise's customer relationship management system, the focus entity in the target data query template is traversed to obtain the corresponding target entity. The target ontology type layer includes entities corresponding to business concepts in the customer relationship management system, edges corresponding to business relationships between the business concepts, and attributes of each business concept. Next, based on a pre-configured virtual mapping, the target business domain service corresponding to the target entity is called to parse the associated instance of the target entity, and business data is determined based on the target entity and the associated instance. The virtual mapping is used to maintain the mapping relationship between business concepts and data acquisition methods corresponding to business domain services. Finally, preset sensitive fields in the business data are desensitized, and the target big model is used to process the business request based on the desensitized business data to obtain the corresponding target processing result. As can be seen from the above, this application unifies the data retrieval logic through the target ontology type layer. New scenarios only require declaring query templates without rewriting the data retrieval code, fundamentally eliminating the repetitive work and logical fragmentation issues caused by writing independent data retrieval logic and assembling code for each scenario, significantly reducing maintenance costs. Through virtual mapping, zero-materialization real-time data retrieval is achieved, allowing direct reuse of existing business domain services without introducing a graph database, greatly reducing implementation costs and data synchronization complexity. The anonymization and secondary detection mechanisms ensure that complete sensitive information is not sent to the large model, effectively guaranteeing data compliance. In this way, this application can achieve standardized acquisition of business data from an enterprise's customer relationship management system while ensuring the security of sensitive data, thereby accurately processing enterprise business. Attached Figure Description

[0017] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0018] Figure 1 This application provides a flowchart of an enterprise business processing method based on virtual mapping and data anonymization; Figure 2 A specific ontology type layer diagram is provided for this application; Figure 3 This application provides a specific CRM system architecture diagram; Figure 4 A schematic diagram of an enterprise business processing device based on virtual mapping and data desensitization is provided for this application; Figure 5 This application provides a structural diagram of an electronic device. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] With the rapid development of cross-border business, the CRM system has accumulated massive amounts of structured data across multiple business domains, including customers, users, leads, opportunities, contacts, follow-up activities, orders, and GMV metrics. To improve sales and operational efficiency, the CRM system is now incorporating large language model capabilities to provide intelligent functions such as lead scoring, customer health scores, opportunity win rate prediction, business insights, and natural language question answering.

[0021] The current common practice for implementing intelligent functions in CRM systems is to write separate code for each AI scenario to call domain services and assemble context, then feed the context along with the original fields into a large model to generate results, and finally perform post-processing entity ID existence checks on the returned results. This approach has the following problems: highly repetitive and fragmented data retrieval logic, with the same relationship repeatedly expressed in multiple scenarios, leading to a linear increase in maintenance costs with the number of scenarios; weak anti-illusion capabilities, failing to identify errors such as illegal entity types, invalid relationships, and cardinality out-of-bounds errors, resulting in low output reliability; easy leakage of sensitive data, with complete mobile phone numbers, email addresses, and other fields directly fed into the model, posing compliance risks; and poor scalability, requiring the modification of all relevant scenarios to integrate new business domains, hindering smooth evolution. Therefore, this application provides an enterprise business processing solution based on virtual mapping and data anonymization, which can achieve standardized acquisition of business data from an enterprise's customer relationship management system while ensuring the security of sensitive data, thereby processing enterprise business.

[0022] See Figure 1As shown, this embodiment of the invention discloses an enterprise business processing method based on virtual mapping and data anonymization, which may include: Step S11: Obtain the business request uploaded by the user, and determine the target business scenario corresponding to the business request based on the preset scenario mapping rules, and determine the target data query template corresponding to the target business scenario.

[0023] In this embodiment, the user-uploaded business request is first obtained. Then, the scene routing component determines the target business scene corresponding to the business request based on a preset scene mapping rule. The specific process may include: extracting keywords from the business request to obtain corresponding target keywords, and matching the target keywords with each scene keyword in the preset scene mapping rule; if the target keyword matches any of the scene keywords, the business scene corresponding to the scene keyword is determined as the target business scene; if the target keyword fails to match any of the scene keywords in the preset scene mapping rule, a pre-selected scene is obtained, and the pre-selected scene is determined as the target business scene.

[0024] Specifically, technical personnel pre-declare each AI function scenario as a data query template, including the focus concept, focus entity location method, relationship path, attribute projection list, and traversal depth threshold; and configure scenario routing rules (keywords, page context, etc.) and pre-selected scenarios, i.e., fallback scenarios. In actual application, the scenario routing component routes business requests to the corresponding target business scenario. If no match can be found, it routes to the pre-selected scenario and determines the target data query template corresponding to the target business scenario.

[0025] Step S12: In the target ontology type layer corresponding to the enterprise's customer relationship management system, traverse from the focus entity in the target data query template to obtain the corresponding target entity; the target ontology type layer includes the entity corresponding to the business concept in the customer relationship management system, the edge corresponding to the business relationship between each business concept, and the attributes of each business concept.

[0026] In this embodiment, technicians pre-configure the target ontology type layer TBox in the customer relationship management system using declarative methods. This includes: business concepts of the CRM system (such as customers, users, leads, opportunities, contacts, follow-up activities, responsible persons, etc.), data attributes of business concepts, object attributes between business concepts (relationships, domains, and value domains, with cardinality annotations such as 1:N, N:1, N:M), and constraint axioms (such as follow-up activities cannot be directly attributed to customers, and customers and users have a one-to-many relationship). When the customer relationship management system starts, it loads the target ontology type layer TBox and provides read-only query interfaces for concept validity, relationship cardinality, constraint quantity, and domain / value domain. If a definition is missing, a parsing error occurs, or the quantity is inconsistent, the system aborts the startup to ensure the integrity of the semantic model and records a locatable error log to prevent the service from being provided with an empty ontology or a partial ontology.

[0027] See target ontology type layer Figure 2 As shown; where solid arrows represent bound data mapping relationships (real-time parsing of instances at runtime), dashed arrows represent declared but unbound concepts (parsing returns an empty set), N:1 / 1:N is the relation cardinality, i.e., domain → value domain, nodes are concepts in the customer relationship management system, and lines are object attributes between business concepts.

[0028] It should be noted that in this embodiment, in the target ontology type layer corresponding to the enterprise's customer relationship management system, the traversal starts from the focus entity in the target data query template to obtain the corresponding target entity. The specific process may include: traversing the target ontology type layer hop by hop according to the business relationship path in the target data query template, starting from the focus entity; correspondingly, the above-mentioned hop by hop traversal in the target ontology type layer starting from the focus entity may also include: if the current traversal depth does not exceed a preset traversal depth threshold, then continue traversing the next hop entity in the target ontology type layer according to the business relationship path; if the current traversal depth exceeds the preset traversal depth threshold, then stop traversing, and determine each entity obtained by traversal as the target entity.

[0029] Specifically, the ontology query service starts with the focus entity corresponding to the focus concept and traverses the business relationship path in the target data query template hop by hop. During the traversal, the current traversal depth is compared with a preset traversal depth threshold. If it does not exceed the preset traversal depth threshold, the traversal continues; if it exceeds the preset traversal depth threshold, the traversal stops, the target entity is generated, and out-of-bounds data is truncated at the nearest boundary. The preset traversal depth threshold is limited to a preset range, such as 1 to 5.

[0030] In addition, the ontology definition and scene query template in this embodiment are both distributed through the configuration center and support hot updates.

[0031] Step S13: Based on the pre-configured virtual mapping, call the target business domain service corresponding to the target entity to parse the associated instance of the target entity, and determine the business data according to the target entity and the associated instance; the virtual mapping is used to maintain the mapping relationship between business concepts and the data acquisition methods corresponding to the business domain service.

[0032] In this embodiment, technicians configure data retrieval mapping for each implemented business concept and business relationship: business concept / relationship of the ontology → data retrieval method of existing business domain service, that is, specifying the data retrieval method of the existing business domain service called corresponding to the business concept and the input / output parameter conversion rules. For concepts that have not yet been integrated (such as contracts and merchants), they are only declared in the target ontology type layer TBox, without configuring mapping.

[0033] It should be noted that ontology instances are not materialized for storage and are not imported into graph databases; during queries, existing domain services are dynamically invoked to resolve instances based on mappings. Mappings support declaration before binding; for mappings that have been declared but not yet bound, the mapping resolver returns an empty set instead of throwing an exception.

[0034] In this embodiment, based on a pre-configured virtual mapping, the target business domain service corresponding to the target entity is invoked to parse the associated instance of the target entity, and business data is determined according to the target entity and the associated instance. The specific process may include: firstly, querying the virtual mapping according to the business concept corresponding to the target entity to determine the target business domain service corresponding to the target entity, and determining the target data acquisition method corresponding to the target business domain service; obtaining the associated instance of the target entity based on the target data acquisition method; using the target entity as a node, establishing connections between different target entities with business relationships, labeling the business concepts corresponding to the nodes and the business relationships corresponding to the connections, and associating the target entity with the associated instance corresponding to the target entity to construct an entity neighborhood subgraph; and determining the entity neighborhood subgraph as the business data.

[0035] Specifically, during the hop-by-hop traversal of the target ontology type layer, the target business domain service is invoked via virtual mapping to resolve the associated instances of the target entity. Using the target entity as a node, connections are established between different target entities with business relationships, constructing an entity neighborhood subgraph where "nodes carry concept types and edges carry relationship types." An upper limit is set on the number of instances that a single node can expand along the same relationship (e.g., no more than 200) to prevent the subgraph from exploding in size.

[0036] Step S14: Desensitize the preset sensitive fields in the business data, and use the target large model to process the business request based on the desensitized business data to obtain the corresponding target processing result.

[0037] In this embodiment, the process of de-identifying preset sensitive fields in business data may include: first, identifying preset sensitive fields in the business data; the preset sensitive fields include mobile phone numbers, email addresses, and ID numbers; then, de-identifying the preset sensitive fields in the business data to obtain de-identified business data, and detecting whether the preset sensitive fields exist in the de-identified business data; if the preset sensitive fields still exist in the de-identified business data, blocking the call to the target large model and returning a preset error message; if the preset sensitive fields do not exist in the de-identified business data, using the target large model to process the business request based on the de-identified business data.

[0038] Specifically, sensitive fields such as mobile phone numbers, email addresses, and ID numbers in the entity neighborhood subgraph are anonymized. A second check is performed before injecting the anonymized business data into the target large model. If the complete sensitive information can still be matched after anonymization, the anonymization is deemed to have failed, the large model call is blocked, and no data is sent. Subsequent large model calls are only allowed after the anonymization is successful.

[0039] Next, the target big model is used to process business requests based on the de-identified business data to obtain the corresponding target processing result. The specific process may include: first, assembling the de-identified business data and the business request into a target prompt word, and sending the target prompt word to the target big model; then, using a server-side event stream method to obtain the initial processing result of the incremental output of the target big model, and verifying the initial processing result to determine the target processing result based on the verification result.

[0040] Specifically, after data anonymization, the anonymized business data and business requests are assembled into prompts and subgraph contexts, which are then invoked on the target large model / Agent platform. In dialogue scenarios, server-side event streams (SSE) incrementally return processing results, and a structured initial processing result is output at the end.

[0041] It should be noted that when calling the large model times out or the returned content cannot be parsed, the large model can be called again, at most once. If it still fails, the result will be returned as plain text without a score and no exception will be thrown to the caller.

[0042] Furthermore, the initial processing result is validated to determine the target processing result based on the validation result. The specific process may include: validating whether the entities referenced in the initial processing result exist in the target ontology type layer; whether the entity relationships in the initial processing result are declared in the target ontology type layer; whether the relationship cardinality corresponding to the entity relationship matches the relationship cardinality defined in the target ontology type layer; and validating whether the entity number corresponding to the entity referenced in the initial processing result exists through the target business domain service. If both the entities and entity relationships referenced in the initial processing result pass the validation, the initial processing result is directly determined as the target processing result. If there are entities and / or entity relationships in the initial processing result that fail the validation, the entities and / or entity relationships that fail the validation are removed from the initial processing result, and the target processing result is determined based on the processed result after removal. If both the entities and entity relationships referenced in the initial processing result fail the validation, the entities and entity relationships referenced in the initial processing result are removed to obtain plain text data, and the plain text data is determined as the target processing result.

[0043] Specifically, after obtaining the initial processing result returned by the target large model, the initial processing result is subjected to a four-level progressive anti-illusion verification using the target ontology type layer TBox: type level + relation level + relation cardinality level + entity existence level. That is, it verifies whether each entity type referenced in the initial processing result is a legal concept, whether the entity number actually exists after reverse lookup through the domain service, and whether the entity relationship is within the ontology and its cardinality is compliant. If all entities and entity relationships referenced in the initial processing result pass the verification, the initial processing result is directly determined as the target processing result; if there are entities and / or entity relationships that fail the verification, the parts that fail the verification are removed and the legal parts are retained, and the target processing result is determined based on the processing result after removal; if all are illegal, it is downgraded to a plain text result without entity relationships. The verification process does not throw exceptions to ensure the stability of the service.

[0044] Understandably, for pre-calculated results of scoring categories, the entity identifier and feature version number are used as cache keys. The results are cached with a maximum lifespan. When the feature version number changes, the old cache becomes invalid, and recalculation occurs on the next read, ensuring the result is consistent with the current feature version number. Asynchronous pre-calculation can also be combined with distributed scheduling to reduce real-time latency. Furthermore, passing the same tracing identifier between the interface layer, orchestration service, and model calls facilitates problem localization and performance analysis.

[0045] It should be noted that the entire business processing flow of the CRM system is implemented by overlaying the existing interface layer and service layer. No new independent gateway is built, and no graph database is introduced. The instance is always a real-time projection of the existing business database, which minimizes the implementation cost.

[0046] As can be seen from the above, this embodiment first obtains the user-uploaded business request, and determines the target business scenario corresponding to the business request and the target data query template corresponding to the target business scenario based on the preset scenario mapping rules. Then, in the target ontology type layer corresponding to the enterprise's customer relationship management system, the focus entity in the target data query template is traversed as the starting point to obtain the corresponding target entity. The target ontology type layer includes the entity corresponding to the business concept in the customer relationship management system, the edge corresponding to the business relationship between each business concept, and the attributes of each business concept. Next, based on the pre-configured virtual mapping, the target business domain service corresponding to the target entity is called to parse the associated instance of the target entity, and the business data is determined according to the target entity and the associated instance. The virtual mapping is used to maintain the mapping relationship between the business concept and the data acquisition method corresponding to the business domain service. Finally, the preset sensitive fields in the business data are desensitized, and the target big model is used to process the business request based on the desensitized business data to obtain the corresponding target processing result. As can be seen from the above, this embodiment unifies the data retrieval logic through the target ontology type layer. New scenarios only require declaring query templates without rewriting the data retrieval code, fundamentally eliminating the repetitive work and logical fragmentation issues caused by writing independent data retrieval logic and assembling code for each scenario, significantly reducing maintenance costs. Virtual mapping enables zero-materialization real-time data retrieval, allowing direct reuse of existing business domain services without introducing a graph database, greatly reducing implementation costs and data synchronization complexity. De-identification and secondary detection mechanisms ensure that complete sensitive information is not sent to the large model, effectively guaranteeing data compliance. In this way, this embodiment can achieve standardized acquisition of business data from an enterprise's customer relationship management system while ensuring the security of sensitive data, thereby accurately processing enterprise business.

[0047] In one specific implementation, the architecture diagram of the AI ​​Native CRM (AI-native customer relationship management) system can be found here. Figure 3 As shown. Specifically, it can include: ① Presentation layer: The front end of the AI ​​Native CRM system, which includes user interaction entry points such as natural language question answering and AI field rendering.

[0048] ② Interface layer: Provides AI endpoints and page aggregation endpoints to receive front-end requests.

[0049] ③ AI Orchestration and Ontology Semantic Layer (Core Layer): This includes scene routing (routing requests to the corresponding business scenario), ontology query service (assembling entity neighborhood subgraphs according to ontology paths), desensitization and blocking (desensitizing sensitive fields and performing secondary checks), ontology validator (performing anti-illusion checks on model output based on TBox), and feature versioning caching (caching scoring results using "entity + feature version number"). The AI ​​Orchestration and Ontology Semantic Layer relies on declarative virtual ontology—the ontology type layer TBox defines business concepts, data attributes, object attributes (including relations and cardinality), domain / range, and constraint axioms; the virtual mapping layer maintains the mapping relationship between concepts / relationships and the data acquisition methods of existing domain services, supporting declaration before binding, without introducing a graph database. After desensitization, the subgraph context is streamed into the large language model / Agent platform via SSE for inference. The inference result is returned to the ontology validator for verification, and then sent back to the front-end presentation layer via the interface layer.

[0050] ④ Existing business domain service layer: includes existing domain services such as customers, users, leads, business opportunities, events, GMV, etc., and the data comes from relational databases, search engines and caches.

[0051] It should be noted that, on the one hand, the business relationships in this embodiment are declared once in the ontology and reused in all scenarios. Adding a new AI scenario only requires declaring the query template without rewriting the data retrieval code, fundamentally eliminating the duplication and fragmentation caused by writing independent data retrieval logic and assembly code for each scenario. Maintenance costs no longer accumulate linearly with the number of scenarios. On the other hand, this embodiment upgrades from "checking entity ID existence one by one after the fact" to "type-level + relationship-level + cardinality-level verification based on the ontology type layer," which can identify and automatically eliminate illegal entity types fabricated in the large model, non-existent relationships, and out-of-bounds relationships that violate business rules, significantly improving the credibility and business compliance of AI output. Furthermore, this embodiment uses context anonymization and a pre-injection secondary detection blocking mechanism to prevent complete sensitive information from entering the large model, reducing data leakage and compliance risks. Secondly, this embodiment supports "declaration before binding" of business concepts. When accessing a new business domain, only virtual mapping needs to be added; the semantic model and query templates for each scenario remain unchanged, resulting in low evolution costs. In addition, the ontology in this embodiment adopts virtual mapping, without introducing a graph database or adding a new independent storage unit. The instance data is taken from existing business domain services in real time, and can be directly superimposed on the existing system architecture.

[0052] Next, a detailed explanation of the business processing procedures, such as high-intent lead scoring and customer management insight Q&A, will be provided in the cross-border payment CRM system.

[0053] 1. Operating environment and component versions: Operating environment: Java 1.8, Spring Boot 1.5.x.

[0054] The configuration center uses Apollo to store the large model platform address, key, and timeout parameters (default timeout is 15 seconds), and hardcoding is prohibited.

[0055] Persistence is handled by MySQL 5.7 (compatible with 8.0), retrieval / behavioral signals are handled by ElasticSearch, and results are cached using Redis (the score cache lifespan does not exceed 30 minutes).

[0056] Serialization uses Fastjson (autoType disabled), tracing uses SkyWalking, and log masking uses a unified masking component.

[0057] Large models / agents use self-built agent platforms (such as 72B level models), and dialogue scenarios are incrementally returned via server event streams (SSE).

[0058] 2. Ontology Declaration (excerpt): Concepts include customer, user, lead, opportunity, contact person, follow-up activity, and person in charge; relationships include "user belongs to customer (N:1)", "user owns opportunity (1:N)", "activity belongs to one of lead / user / opportunity", etc., and the cardinality is marked; constraint axioms include "activity cannot be directly attributed to customer"; at the same time, concepts such as contract and merchant are declared but not bound to mapping.

[0059] 3. Clue scoring scenario: The focus concept in the query template is "clue", the relationship path is "clue → behavioral signal + clue → person in charge", and the maximum traversal depth is 2; at runtime, the clue neighborhood subgraph is assembled, and after desensitization, the model is called to generate a score; the score value is constrained to a closed interval of 0 to 100, and the level is determined by a unique mapping (not less than 80 is high intention HOT, 60 to 80 is medium intention WARM, and less than 60 is low intention COLD); the results are cached as "clue + feature version".

[0060] 4. Business Insight Q&A Scenario: Users ask questions in natural language on the front end. The scenario routing matches the "Business Insight" scenario. The ontology query service assembles a multi-concept aggregation subgraph of customers / opportunities / activities / GMV. After desensitization, the model is called in a streaming manner. The returned structured results are rendered into text, indicators and suggested actions after the ontology validator removes illegal entities and relationships fabricated by the model. Among them, suggested actions involving funds and outbound calls are forcibly marked as "requires secondary confirmation".

[0061] 5. When the issue involves concepts such as "contract" and "merchant" that are not yet integrated, the relevant relationship path resolution returns an empty set, and the front end displays a unified placeholder without reporting an error; when the corresponding business domain is integrated in the future, only the virtual mapping needs to be added for it to take effect, and the ontology and scene template do not need to be modified.

[0062] Accordingly, see Figure 4 As shown in the embodiments of this application, an enterprise business processing device based on virtual mapping and data anonymization is also provided, which may include: The data query template determination module 11 is used to obtain the business request uploaded by the user, and determine the target business scenario corresponding to the business request based on the preset scenario mapping rules, and determine the target data query template corresponding to the target business scenario; The target entity acquisition module 12 is used to traverse the target ontology type layer corresponding to the enterprise's customer relationship management system, starting from the focus entity in the target data query template, to obtain the corresponding target entity; the target ontology type layer includes entities corresponding to business concepts in the customer relationship management system, edges corresponding to business relationships between the business concepts, and attributes of the business concepts. The business data determination module 13 is used to call the target business domain service corresponding to the target entity based on a pre-configured virtual mapping to parse the associated instance of the target entity, and determine the business data according to the target entity and the associated instance; the virtual mapping is used to maintain the mapping relationship between business concepts and the data acquisition methods corresponding to the business domain services; The business processing module 14 is used to desensitize the preset sensitive fields in the business data, and use the target big model to process the business request based on the desensitized business data to obtain the corresponding target processing result.

[0063] In some specific embodiments, the data query template determination module 11 may include: The scenario keyword matching unit is used to extract keywords from the business request to obtain corresponding target keywords, and match the target keywords with the scenario keywords in the preset scenario mapping rules; The target business scenario determination unit is used to determine the business scenario corresponding to the scenario keyword as the target business scenario if the target keyword successfully matches any of the scenario keywords. The alternative scenario determination unit is used to obtain a pre-selected scenario and determine the pre-selected scenario as the target business scenario if the target keyword fails to match the scenario keywords in the preset scenario mapping rules.

[0064] In some specific embodiments, the target entity acquisition module 12 may include: The jump-by-jump traversal submodule is used to perform jump-by-jump traversal in the target ontology type layer, starting from the focus entity, according to the business relationship path in the target data query template. Correspondingly, the jump-by-jump traversal submodule may further include: The depth determination unit is used to continue traversing the next-hop entity in the target ontology type layer according to the business relationship path if the current traversal depth does not exceed the preset traversal depth threshold. The traversal stop unit is used to stop traversing if the current traversal depth exceeds the preset traversal depth threshold, and to determine each entity obtained by traversal as the target entity.

[0065] In some specific embodiments, the business data determination module 13 may include: The mapping query unit is used to query the virtual mapping according to the business concept corresponding to the target entity, so as to determine the target business domain service corresponding to the target entity, and determine the target data acquisition method corresponding to the target business domain service, and obtain the associated instance of the target entity based on the target data acquisition method; The subgraph construction unit is used to establish connections between different target entities with business relationships, with the target entity as the node, label the business concept corresponding to the node and the business relationship corresponding to the connection, and associate the target entity with the associated instance corresponding to the target entity to construct an entity neighborhood subgraph. A business data determination unit is used to determine the entity neighborhood subgraph as the business data.

[0066] In some specific embodiments, the business processing module 14 may include: A sensitive field identification unit is used to identify preset sensitive fields in the business data; the preset sensitive fields include mobile phone number, email address, and ID number. The desensitization detection unit is used to desensitize the preset sensitive fields in the business data to obtain the desensitized business data, and to detect whether the preset sensitive fields exist in the desensitized business data; The large model call blocking unit is used to block the call to the target large model and return a preset error message if the preset sensitive field still exists in the de-identified business data. The business request processing unit is used to process the business request based on the desensitized business data using the target big model if the preset sensitive field does not exist in the desensitized business data.

[0067] In some specific embodiments, the business processing module 14 may include: The prompt word assembly submodule is used to assemble the de-identified business data and the business request into a target prompt word, and send the target prompt word to the target large model; The verification result determination submodule is used to obtain the initial processing result of the incremental output of the target large model using a server-side event stream method, and to verify the initial processing result in order to determine the target processing result based on the verification result.

[0068] In some specific implementations, the verification result determination submodule may include: The processing result verification unit is used to verify whether the entity referenced in the initial processing result exists in the target ontology type layer, whether the entity relationship in the initial processing result is declared in the target ontology type layer, whether the relationship cardinality corresponding to the entity relationship matches the relationship cardinality defined in the target ontology type layer, and to verify whether the entity number corresponding to the entity referenced in the initial processing result exists through the target business domain service. The first processing result determination unit is used to directly determine the initial processing result as the target processing result if the entities and entity relationships referenced in the initial processing result are all verified. The second processing result determination unit is used to remove entities and / or entity relationships that fail verification in the initial processing result if such entities and / or entity relationships fail verification in the initial processing result, and to determine the target processing result based on the processing result after removal. The third processing result determination unit is used to remove all entities and entity relationships referenced in the initial processing result if the verification fails, thereby obtaining plain text data, and determining the plain text data as the target processing result.

[0069] Furthermore, embodiments of this application also disclose an electronic device, Figure 5 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content of the diagram should not be construed as limiting the scope of this application. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the enterprise business processing method based on virtual mapping and data anonymization disclosed in any of the foregoing embodiments. Furthermore, the electronic device 20 in this embodiment may specifically be an electronic computer.

[0070] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.

[0071] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or optical disk, etc. The resources stored thereon can include operating system 221, computer program 222, etc., and the storage method can be temporary storage or permanent storage.

[0072] The operating system 221 is used to manage and control the various hardware devices on the electronic device 20 and the computer program 222, which may be Windows Server, Netware, Unix, Linux, etc. In addition to including computer programs capable of performing the enterprise business processing methods based on virtual mapping and data anonymization disclosed in any of the foregoing embodiments, the computer program 222 may further include computer programs capable of performing other specific tasks.

[0073] Furthermore, this application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the aforementioned disclosed enterprise business processing method based on virtual mapping and data anonymization. Specific steps of this method can be found in the corresponding content disclosed in the foregoing embodiments, and will not be repeated here.

[0074] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.

[0075] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0076] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0077] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0078] The technical solutions provided in this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for enterprise business processing based on virtual mapping and data anonymization, characterized in that, include: Obtain the business request uploaded by the user, and determine the target business scenario corresponding to the business request based on the preset scenario mapping rules, and determine the target data query template corresponding to the target business scenario; In the target ontology type layer corresponding to the enterprise's customer relationship management system, the corresponding target entity is obtained by traversing from the focus entity in the target data query template. The target ontology type layer includes the entity corresponding to the business concept in the customer relationship management system, the edge corresponding to the business relationship between each business concept, and the attributes of each business concept. Based on a pre-configured virtual mapping, the target business domain service corresponding to the target entity is invoked to parse the associated instance of the target entity, and business data is determined according to the target entity and the associated instance; the virtual mapping is used to maintain the mapping relationship between business concepts and the data acquisition methods corresponding to the business domain service; The preset sensitive fields in the business data are de-identified, and the target large model is used to process the business request based on the de-identified business data to obtain the corresponding target processing result.

2. The enterprise business processing method based on virtual mapping and data anonymization according to claim 1, characterized in that, The step of determining the target business scenario corresponding to the business request based on preset scenario mapping rules includes: The business request is subjected to keyword extraction to obtain the corresponding target keywords, and the target keywords are matched with the scenario keywords in the preset scenario mapping rules; If the target keyword successfully matches any of the scenario keywords, then the business scenario corresponding to the scenario keyword is determined as the target business scenario; If the target keyword fails to match any of the scenario keywords in the preset scenario mapping rules, then a pre-selected scenario is obtained and the pre-selected scenario is determined as the target business scenario.

3. The enterprise business processing method based on virtual mapping and data anonymization according to claim 1, characterized in that, In the target ontology type layer corresponding to the enterprise's customer relationship management system, the corresponding target entities are obtained by traversing from the focus entity in the target data query template, including: According to the business relationship path in the target data query template, a jump-by-jump traversal is performed in the target ontology type layer starting from the focus entity; Accordingly, the process of traversing the target ontology type layer hop by hop starting from the focus entity also includes: If the current traversal depth does not exceed the preset traversal depth threshold, then continue traversing the next-hop entity in the target ontology type layer according to the business relationship path; If the current traversal depth exceeds the preset traversal depth threshold, then the traversal stops, and each entity obtained through the traversal is determined as the target entity.

4. The enterprise business processing method based on virtual mapping and data anonymization according to claim 1, characterized in that, The step of invoking the target business domain service corresponding to the target entity based on the pre-configured virtual mapping to parse the associated instance of the target entity, and determining business data based on the target entity and the associated instance, includes: The virtual mapping is queried according to the business concept corresponding to the target entity to determine the target business domain service corresponding to the target entity, and the target data acquisition method corresponding to the target business domain service is determined. The associated instance of the target entity is obtained based on the target data acquisition method. Using the target entity as a node, establish connections between different target entities with business relationships, label the business concepts corresponding to the nodes and the business relationships corresponding to the connections, and associate the target entity with the associated instance corresponding to the target entity to construct an entity neighborhood subgraph; The entity neighborhood subgraph is identified as the business data.

5. The enterprise business processing method based on virtual mapping and data anonymization according to claim 1, characterized in that, The process of desensitizing preset sensitive fields in the business data includes: Identify preset sensitive fields in the business data; the preset sensitive fields include mobile phone number, email address, and ID number. The preset sensitive fields in the business data are de-identified to obtain the de-identified business data, and the preset sensitive fields are detected to be present in the de-identified business data. If the preset sensitive fields still exist in the de-identified business data, the call to the target large model will be blocked, and a preset error message will be returned. If the preset sensitive field is not present in the anonymized business data, the target large model is used to process the business request based on the anonymized business data.

6. The enterprise business processing method based on virtual mapping and data anonymization according to any one of claims 1 to 5, characterized in that, The process of using the target large model to process the business request based on the anonymized business data to obtain the corresponding target processing result includes: The anonymized business data and the business request are combined to form a target prompt word, and the target prompt word is sent to the target large model; The initial processing result of the incremental output of the target large model is obtained by using a server-side event stream method, and the initial processing result is verified to determine the target processing result based on the verification result.

7. The enterprise business processing method based on virtual mapping and data anonymization according to claim 6, characterized in that, The step of verifying the initial processing result to determine the target processing result based on the verification result includes: Verify whether the entity referenced in the initial processing result exists in the target ontology type layer, whether the entity relationship in the initial processing result is declared in the target ontology type layer, whether the relationship cardinality corresponding to the entity relationship matches the relationship cardinality defined in the target ontology type layer, and verify whether the entity number corresponding to the entity referenced in the initial processing result exists through the target business domain service; If all entities and entity relationships referenced in the initial processing result pass the verification, then the initial processing result is directly determined as the target processing result; If there are entities and / or entity relationships that fail the verification in the initial processing result, then the entities and / or entity relationships that fail the verification in the initial processing result are removed, and the target processing result is determined based on the processing result after removal. If the entities and entity relationships referenced in the initial processing result fail the verification, then the entities and entity relationships referenced in the initial processing result are removed to obtain plain text data, and the plain text data is determined as the target processing result.

8. An enterprise business processing device based on virtual mapping and data anonymization, characterized in that, include: The data query template determination module is used to obtain the business request uploaded by the user, determine the target business scenario corresponding to the business request based on the preset scenario mapping rules, and determine the target data query template corresponding to the target business scenario. The target entity acquisition module is used to traverse the target ontology type layer corresponding to the enterprise's customer relationship management system, starting from the focus entity in the target data query template, to obtain the corresponding target entity; the target ontology type layer includes entities corresponding to business concepts in the customer relationship management system, edges corresponding to business relationships between the business concepts, and attributes of the business concepts; The business data determination module is used to call the target business domain service corresponding to the target entity based on a pre-configured virtual mapping to parse the associated instance of the target entity, and determine the business data according to the target entity and the associated instance; the virtual mapping is used to maintain the mapping relationship between business concepts and the data acquisition methods corresponding to the business domain service; The business processing module is used to de-identify preset sensitive fields in the business data, and use the target large model to process the business request based on the de-identified business data to obtain the corresponding target processing result.

9. An electronic device, characterized in that, The electronic device includes a processor and a memory; wherein the memory is used to store a computer program, which is loaded and executed by the processor to implement the enterprise business processing method based on virtual mapping and data desensitization as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, Used to store computer programs, which, when executed by a processor, implement the enterprise business processing method based on virtual mapping and data desensitization as described in any one of claims 1 to 7.

Images