Cross-model customer portrait collaborative updating intelligent customer service method and system

Abstract

The invention relates to the technical field of intelligent customer service and customer portrait data management, and discloses a cross-model customer portrait collaborative updating intelligent customer service method and system. The method is used for solving the problem of leakage risk caused by non-uniform shared calibers or insufficient information caused by excessive desensitization in a traditional method. The method comprises the following steps: firstly, receiving a multi-channel dialogue request by a unified entrance, completing identity authentication, generating a dialogue number and a link identifier, and constructing a retrieval key capable of being reused across the entrances; and aggregating the multi-information-source portrait elements based on the retrieval key and labeling sensitive levels to form an entry list. And matching a sharing rule according to the scene and the target model, determining a sharing range and a desensitization aperture, and generating a version identifier. And according to the input constraint, formatting and cutting the load, splicing the historical message injection model, updating the portrait version when the session ends or the condition is triggered, and writing the portrait version into the auditing index.

Description

Technical Field

[0001] This invention relates to the field of intelligent customer service and customer profile data governance technology, specifically to an intelligent customer service method and system for cross-model customer profile collaborative updating. Background Technology

[0002] With the development of large language models and their engineering deployment capabilities, enterprises often introduce multiple models or model versions simultaneously in customer service and operation scenarios to adapt to the cost, latency, inference capabilities, and compliance requirements of different business processes. In this type of multi-model parallel application, enterprises usually need to enable the models to understand customers' historical interactions, service requests, preference characteristics, and key factual information related to the business system, thereby improving dialogue coherence and reducing the risk of model illusion. In existing technologies, customer profiles and session contexts are mostly derived from enterprise customer relationship management systems, online and offline service logs, and business process data. The aggregation, processing, storage, and retrieval of these data across different systems typically rely on interfaces and data processing links built separately by each business system, or on each model application independently extracting and maintaining relevant tag information at its own stage. When multiple models participate in the same customer's full-process service, the above approach can easily lead to customer tags and context information being scattered in multiple places, increasing the cost of unified security control and consistency maintenance. Furthermore, when business information changes over time, cross-model profile updates and data consistency assurance become more difficult. Furthermore, different models differ in context window size, input format, and processing capabilities. To ensure a consistent service experience across models, differentiated supply and adaptation of relevant information for the same customer are necessary. Otherwise, issues such as repeated communication after model switching, inconsistent background understanding, and disjointed response styles may arise. Meanwhile, the number of customer profile tags may be large. Directly injecting all tags would cause information overload and token waste, affecting model attention and response efficiency. To meet customer experience requirements, related processing needs to be kept within a low latency range. Traditional sequential assembly and temporary trimming methods cannot balance completeness, relevance, and real-time performance. In addition, customer profiles and conversation contexts typically contain sensitive data such as contact information, identity information, and order contracts. While existing technologies can use regular expression replacement and keyword matching for filtering or desensitization, when multiple models, scenarios, and deployment forms coexist, there is a lack of a controllable sharing mechanism that matches model capabilities, scenario requirements, and context budget. This can easily lead to leakage risks due to inconsistent sharing standards or insufficient information due to excessive desensitization, thereby affecting service continuity and consistency. Therefore, existing technologies still lack a technical solution that can achieve controllable sharing and rapid adaptation of customer profiles and session contexts between models under the condition of multi-model parallel service, so as to reduce the maintenance costs and security risks caused by information dispersion while meeting the constraints of different model windows and formats and latency requirements, and improve the continuity and consistency of cross-model services. Summary of the Invention

[0003] To address the shortcomings of existing technologies, this invention provides a cross-model customer profile collaborative update intelligent customer service method and system, which solves the problems of leakage risk caused by inconsistent sharing standards or insufficient information caused by excessive desensitization in traditional methods.

[0004] To achieve the goal of controllable sharing and rapid adaptation of customer profiles and session contexts between models, this invention provides the following technical solution: A cross-model customer profile collaborative update intelligent customer service method includes: S1: After receiving a dialogue request at the unified enterprise portal, perform identity authentication, obtain the customer's unique identifier, generate a session number, and establish a binding record between the customer's unique identifier and the session number; S2: Write the binding record into the session metadata, generate a search key that can be reused across entry points, and write the touchpoint identifier and business scenario identifier into the session metadata; S3: Based on the search key, read the profile elements in the customer information source, classify and label the profile elements according to type and sensitivity level, and generate a list of profile items to be shared; S4: Based on the matching and sharing rules of scene identifiers and model identifiers, determine the sharing scope and desensitization criteria of the item list, generate the shared payload, and write the shared identifier and version identifier; S5: Format and trim the shared payload according to the input constraints of the target model, inject the session context into the target model, and record the call link identifier and payload summary; S6: When the session ends or an update condition is triggered, summarize the interaction logs and profile change information, update the profile entry version, and archive the shared records and audit index.

[0005] In a preferred embodiment, after receiving a dialogue request at the enterprise unified portal, identity authentication is performed, a unique customer identifier is obtained, a session number is generated, and a binding record between the unique customer identifier and the session number is established, including: The login token, single sign-on ticket, or short-term access credential are verified separately according to the entry channel, and the replay is determined based on the request random number and request timestamp. The determination parameters are given by the versioning configuration item. The customer identifier is mapped to a unified customer primary key, a session number and a link identifier are generated and deduplication is performed, the binding record and initial metadata are written by an idempotent key, abnormal replay terminates, and conflict callback is returned.

[0006] In a preferred embodiment, the binding record is written to the session metadata, a retrieval key for cross-entry reuse is generated, and the touchpoint identifier and business scenario identifier are written to the session metadata, including: The session management service reads the bound records by session number, generates session metadata entries by field model, and unifies the time field to the same time base. A search key is generated using the customer's unique identifier and the enterprise's dimension identifier, and a two-way index is established. Other identifiers are written into the alias extension field and do not participate in the search key generation. Touchpoint identifiers and business scenario identifiers are determined and recorded based on entry priority, mapping completion, or default completion, and source conflicts are recorded. Link mapping and version number are atomically written and registered according to versioning rules.

[0007] In a preferred embodiment, profile elements from customer information sources are retrieved based on search keys. These profile elements are then categorized and labeled according to type and sensitivity level to generate a list of profile items to be shared, including: The profile aggregation service reads the search key, enterprise dimension identifier, and entry channel identifier by session number, determines the access order of information sources, and uses the customer's primary key as the access criterion. When the primary key is missing in the search key, perform master data mapping and record the source time; Load the versioned directory and mapping configuration, and pull and merge fields according to the validity period. When a failure or missing item occurs, it is marked according to the status code mapping table and integrity rules, switched according to priority, and the collection log is recorded. Generate standard field keys and field value references in the order of conflict, mark the sensitivity level, and write them into the list summary, directory version number, and reference identifier.

[0008] In a preferred embodiment, based on the scene identifier and model identifier matching and sharing rules, the sharing scope and desensitization criteria of the item list are determined, a shared payload is generated, and a shared identifier and version identifier are written into it, including: The rule matching service reads the scene identifier, touchpoint identifier, entry channel identifier, target model identifier, and profile entry set by session number, and loads the versioned rule base and rule mapping table; Filter scenario rule entries according to the priority conditions corresponding to the touch point identifier and the entry channel identifier, perform compatibility verification with the model rule entries corresponding to the target model, generate a shared scope list and a desensitization caliber table, and record the source of conflict. Generate a shared identifier and combine the rule version with the directory version to form a version identifier. The directory version is taken from the session extension field and atomically written to the session storage.

[0009] In a preferred embodiment, the shared payload is formatted and trimmed according to the input constraints of the target model, a session context is injected into the target model, and the call chain identifier and payload summary are recorded, including: The payload generation service reads the shared identifier, version identifier, scene identifier, and model identifier by session number, and loads the versioned template, caliber priority table, and pruning rules. Entries are extracted based on the shared scope. If a missing entry is found, it is either placed in place or skipped according to the shared granularity and a missing tag is written. Field value references are processed according to the desensitization criteria. When multiple rules are matched, the final criteria are determined according to the criteria priority and the criteria entry identifier is recorded. First, exclude required fields and fields that cannot be clipped according to model constraints, then clip them in stages and write the reason codes. The fragment boundaries and paragraph numbers are set according to the scene template. The model calls the gateway to splice historical fragments and writes the link node tags and payload summary references.

[0010] In a preferred implementation, when the session ends or an update condition is triggered, the interaction logs and profile change information are aggregated, the profile entry version is updated, and shared records and audit indexes are archived, including: The session management service writes interaction logs by session number and maintains session status, writing payload summary references, shared identifier references, and version identifier references; When a session ends or triggers a key field change flag, tag flag, or cross-entry session count flag, an update task is generated according to the versioned time window and threshold, carrying the retrieval key, link identifier, shared scope list reference, and desensitization caliber table reference. The profile update service maps log fragments to directory versions and writes them to version update records after deduplication. It also synchronously writes audit index records containing time slices and establishes primary and secondary indexes.

[0011] On the other hand, the present invention provides a cross-model customer profile collaborative update intelligent customer service system, including: Dialogue Access and Identity Authentication Module: Unifies access to dialogue requests from multiple channels, verifies credentials by channel and determines replay, parses customer identifiers, generates session numbers and link identifiers, and writes binding records to session storage; Session metadata and search key management module: Reads bound records to generate session metadata, generates search keys based on customer unique identifiers and enterprise dimensions, writes touchpoints and scenario identifiers and establishes index associations; Customer profile aggregation and item list generation module: Access customer information source based on search key, pull elements according to profile element catalog and complete field mapping, conflict merging and sensitivity level labeling, and generate profile item list associated with session number; Shared rule matching and definition module: Based on the business scenario identifier and target model identifier matching rule library, it generates a shared scope list and a de-identified definition table, forms a shared identifier and a version identifier, and writes them into session storage; Shared payload generation and model adaptation injection module: Extracts and desensitizes profile entries according to the shared scope, formats and trims them according to model constraints, generates input by splicing historical fragments according to templates, and records link identifiers and payload summaries; Interaction Log Collection and Session State Management Module: Records interaction logs and session state around the session number, solidifies message sequence number, role, timestamp, touch point and scenario definition, and maintains payload summary and sharing, version reference; The profile update and version chain management module receives tasks when a session ends or an update is triggered, locates the update item based on the interaction log and change marker, generates a version record, and writes it back to the profile storage. Audit Index and Archive Tracing Module: Summarizes session numbers, search keys, and shared related references and writes them into the audit index database, organized by primary and secondary indexes, for retrieval tracing and scope review; Scene Graph and Data Object Management Module: Maintains the mapping relationship between scenes and data objects, used to determine the available data objects and data retrieval range for a scene; Context key-value storage module: provides exact matching keys and model-independent universal keys for cross-model reuse and secondary compression conversion; The Measurement and Operation Iteration Module collects feedback metrics and drives the iteration of scenarios, data relationships, and template strategies to continuously optimize consistency and efficiency.

[0012] Compared with existing technologies, this invention provides a cross-model customer profile collaborative update intelligent customer service method and system, which has the following beneficial effects: 1. This invention, through unified entry point identity verification and idempotent binding, solidifies unique customer identifiers and session-level link identifiers, avoiding customer positioning deviations caused by repeated link building across channels; furthermore, it uses search keys to connect the retrieval and conflict merging of profile elements from multiple information sources, forming a list of standard field entries with sensitivity level annotations, reducing the consistency costs caused by customer tags being scattered across different systems and different model sides for maintenance; simultaneously, based on rule matching between business scenarios and target models, it outputs a shared scope list and a de-identified caliber table, and solidifies the shared caliber with shared identifiers and version identifiers, reducing the risk of leakage caused by inconsistent shared calibers. To address the issue of insufficient information due to risks or excessive anonymization, the system further combines model input constraints with scenario combination templates to sort fields, prohibit or remove certain fields, retain required fields, perform hierarchical pruning, and splice contexts on shared payloads. This alleviates the problems of model switching gaps and information overload caused by differences in context windows. Finally, the system drives the profile version chain update with interaction logs, change markers, and versioning trigger conditions, and simultaneously accumulates audit indexes to achieve traceable shared records and verifiable cross-model reuse. This enables controllable sharing, rapid adaptation, and continuous service consistency of cross-model customer profiles and session contexts while meeting latency and window constraints.

[0013] 2. This invention manages the configurations of profile directories, field mappings, sharing rules, anonymization criteria, and pruning rules in a versioned manner at both the enterprise and operational environment levels. It also writes rule and directory version references into the session metadata, ensuring that the data retrieval and sharing criteria for the same customer remain aligned and replayable across different entry points, scenarios, and models. When information sources are unavailable or fields are missing, the system switches according to data retrieval priority and integrity rules and records collection logs, providing a defined degradation path for the profile aggregation process. During the sharing phase, it prioritizes and records source tags for field-level criterion conflicts. During the update phase, it generates traceable version update records based on triggering conditions such as key field changes, tag increments, and cross-entry session counts, and writes these records into an audit index. This ensures version consistency, process verifiability, and accountability traceability in profile sharing and updates, even in situations with complex organizational structures, heterogeneous systems, and rapid model iteration. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the intelligent customer service method for collaborative updating of cross-model customer profiles according to the present invention; Figure 2 This is a schematic diagram of the structure of an intelligent customer service system for collaborative updating of cross-model customer profiles according to the present invention. Detailed Implementation

[0015] 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.

[0016] Example 1: Figure 1 A method for collaborative updating of customer profiles across models in intelligent customer service is presented, including: S1: After receiving a dialogue request at the unified enterprise portal, perform identity authentication, obtain the customer's unique identifier, generate a session number, and establish a binding record between the customer's unique identifier and the session number; S2: Write the binding record into the session metadata, generate a search key that can be reused across entry points, and write the touchpoint identifier and business scenario identifier into the session metadata; S3: Based on the search key, read the profile elements in the customer information source, classify and label the profile elements according to type and sensitivity level, and generate a list of profile items to be shared; S4: Based on the matching and sharing rules of scene identifiers and model identifiers, determine the sharing scope and desensitization criteria of the item list, generate the shared payload, and write the shared identifier and version identifier; S5: Format and trim the shared payload according to the input constraints of the target model, inject the session context into the target model, and record the call link identifier and payload summary; S6: When the session ends or an update condition is triggered, summarize the interaction logs and profile change information, update the profile entry version, and archive the shared records and audit index.

[0017] S1: After receiving a dialogue request at the enterprise unified portal, perform identity authentication, obtain the customer's unique identifier, generate a session number, and establish a binding record between the customer's unique identifier and the session number. The specific implementation is as follows: Enterprises deploy a dialogue access service on a unified entry point to receive dialogue requests from multiple channels and initiate the identity verification process. Upon receiving a request, the dialogue access service selects the corresponding credential parsing and verification rules based on the entry channel identifier, extracts identity credentials from the request header or body, and completes the verification. If the identity credential is a login token, the service verifies the token signature and validity period. If the identity credential is a single sign-on ticket, the service calls the unified authentication service to complete the ticket verification and obtain the authentication result. If the identity credential is a short-term access credential, the service verifies the issuer identifier, signature, and expiration time, and verifies that the application identifier bound to the access credential matches the entry channel identifier. The dialogue access service executes re-... The replay determination process takes the requested random number and request timestamp as input. Within the time window corresponding to the system parameter configuration item, it verifies the uniqueness of the random number under the same entry channel identifier and performs a threshold judgment on the deviation between the request timestamp and the system time. If an abnormal replay is determined, the binding process is terminated. Repeated submissions that hit the same idempotent key enter the idempotent return process. The time window and time deviation threshold are given by the system parameter configuration item, which is maintained and set with version identifiers according to the entry channel identifier and the operating environment. After the dialogue access service starts loading or publishes parameter changes, it refreshes the effective parameters and writes the parameter version identifier into the binding record for traceability. After identity verification, the dialogue access service parses the customer identifier in the request and normalizes it into an enterprise unified customer primary key as a unique customer identifier. When the customer identifier is a customer number or ID number, the dialogue access service calls the customer master data mapping service to complete the conversion and outputs a missing reason code based on the mapping return code table or transfers the request to the manual verification queue. Subsequently, the dialogue access service generates a session number, which is formed by combining a time slice and a random sequence. The time slice is used to represent the time period information of the generation time, and the random sequence remains unique within the same time slice. When concurrent requests from the same customer occur within the time window corresponding to the system parameter configuration item, the session number deduplication constraint is performed based on the unique customer identifier and the entry channel identifier. The dialogue access service synchronously generates a link identifier and establishes an association with the session number. The link identifier is written into the session link record as an association field of subsequent processing nodes. The dialogue access service generates binding records and writes them to the session storage. The binding record includes a unique customer identifier, session number, entry channel identifier, link identifier, authentication method identifier, request timestamp, and request random number, with a unique constraint set for the session number. The binding record writing uses an idempotent write mechanism; the idempotent key is generated by combining the entry channel identifier, request random number, and request timestamp. When an idempotent key is hit, the session number and link identifier already existing in the binding record are read and output. When concurrent writing triggers a conflict with the unique constraint of the session number, an idempotent key lookup is performed, and the corresponding session number and link identifier are output. After the binding record is written, the dialogue access service writes initial session context metadata. This initial metadata includes the session number, unique customer identifier, entry channel identifier, link identifier, and creation timestamp, and records a storage key reference. The initial metadata and the binding record are associated through the session number, used in subsequent steps to read the unique customer identifier based on the session number to generate a retrieval key, and to write the touchpoint identifier and business scenario identifier under the same session number.

[0018] S2: Write the binding record into the session metadata, generate a search key that can be reused across entry points, and write the touchpoint identifier and business scenario identifier into the session metadata. The specific implementation is as follows: After the unified enterprise portal completes the binding record writing, the session management service reads the corresponding binding record from the session storage using the session number as an index, obtains the customer unique identifier, entry channel identifier, link identifier, and creation time information, and generates session metadata entries. The session metadata entries are structured and encapsulated according to a preset field model, which specifies the field set and value caliber for the session number, customer unique identifier, entry channel identifier, link identifier, and creation timestamp. The session management service performs normalization processing on key fields, converts the time field into a unified time base expression, and assigns a metadata number to the session metadata entries, establishing a one-to-one association between the metadata number and the session number. The session management service sets a search key field in the session metadata entries. The search key is generated by combining the customer's unique identifier and the enterprise dimension identifier. The enterprise dimension identifier is used to distinguish tenant spaces or business entities and is determined by the tenant code, system code, or organization code issued by the unified entry point. The search key is concatenated according to a fixed field order and a fixed delimiter rule, and the enterprise's unified customer primary key is used as the value of the customer's unique identifier. If other identifiers such as customer number or ID number exist on the entry side, they are written into the alias extension field of the session metadata entries and do not participate in the search key generation. After the search key is written, the session management service establishes a two-way index of session number and search key in the session index table and records the enterprise dimension identifier simultaneously, which is used for searching the session key based on the session number and searching the session set based on the search key. The session management service writes touchpoint identifiers and business scenario identifiers into the session metadata entries. Touchpoint identifiers and business scenario identifiers preferentially use the entry point's input value; if the entry point is not provided, they are generated according to rule mapping. When both entry point input values ​​and mapping generated values ​​exist, the touchpoint identifier uses the entry point's input value, and the business scenario identifier uses the explicitly input value, with a source flag and conflict handling flag written in. The touchpoint identifier uses enumeration encoding and is associated with the touchpoint dictionary table, while the business scenario identifier uses scenario encoding and is associated with the scenario dictionary table. If the entry point does not provide a business scenario identifier and the rule mapping is not matched, a default business scenario identifier is determined according to the default rule, and a source flag is written to distinguish between explicitly input, mapping generated, and default completion. The touchpoint mapping table, scenario mapping table, default rules, and search key combination rules are maintained by the system rule configuration items. These rule configuration items are managed according to enterprise-level and runtime environment dimensions and have version identifiers. After loading the corresponding version rule, the session management service generates a search key and writes it to the touchpoint identifier and business scenario identifier. The session management service updates the session metadata entries using atomic writes, ensuring that the search key, touchpoint identifier, and business scenario identifier are written to the database in the same write operation, and writes the rule version identifier and metadata version number to the session metadata entries. The metadata version number is associated with the field model version, touchpoint dictionary version, and scenario dictionary version. The session management service writes a mapping record from the link identifier to the session number at the index layer, and subsequent processing nodes locate the corresponding session metadata entries through the link identifier. After writing is complete, subsequent steps access the customer information source and extract profile elements through the search key.

[0019] S3: Based on the search key, retrieve profile elements from the customer information source, categorize and label the profile elements according to type and sensitivity level, and generate a list of profile items to be shared. The specific implementation is as follows: After the session metadata entries are generated and stored, the profile aggregation service receives the session number as the input parameter and reads the search key, enterprise dimension identifier, and entry channel identifier corresponding to the session number from the session storage. The profile aggregation service determines the access scope and access order of customer information sources based on the enterprise dimension identifier and entry channel identifier. The profile aggregation service uses the search key as the basis for customer positioning to determine the unified customer primary key of the enterprise as the access criterion. When the search key does not contain the unified customer primary key of the enterprise, the identifier conversion is completed through the customer master data service, and the mapping source and mapping time are recorded. The profile aggregation service accesses customer information sources based on connection and field mapping configurations. These sources include one or more of the following: enterprise customer relationship management system, order contract system, service ticket system, and historical dialogue archive. The profile element catalog, field mapping configuration, field integrity rules, information source status code mapping table, and sensitivity level dictionary are maintained and version-identified by the configuration center according to enterprise and operating environment dimensions. After loading the corresponding version, the profile aggregation service performs data retrieval, merging, and annotation. The profile element catalog categorizes profile elements into identity information, relationship information, preferences and tags, historical service summary, and business status. The system identifies data types such as status and defines field sets, field data sources, data retrieval priorities, field validity periods, and conflict merging order for each type. The field validity period is defined in the field rules section of the profile element catalog. The profile aggregation service accesses each customer information source in the order of the profile element catalog, and the retrieval request carries the enterprise's unified customer primary key, enterprise dimension identifier, and permission scope marker. When an access failure or data loss occurs, the system determines the access result marker based on the information source status code mapping table and field integrity rules, switches to the next customer information source according to the data retrieval priority, and writes the access result marker, information source identifier, and time information into the profile collection log. The profile aggregation service merges and organizes the raw data retrieved according to the type definitions in the profile element catalog. Fields from different information sources under the same type are selected and discarded according to the conflict merging order, forming a correspondence between standard field keys and field value references. Field value references are saved using reference identifiers or summary values. The profile aggregation service annotates the standard field keys with sensitivity levels based on a sensitivity level dictionary. Sensitivity levels are divided into public, internal, sensitive, and strictly sensitive, and a sensitivity level tag and a level source tag are written for each field. When the same field hits multiple level rules at the same time, the rule with the higher level is used as the final annotation basis. The profile aggregation service generates a profile entry list and writes it to the session-side cache. Simultaneously, it writes a list summary and catalog version number to the session metadata extension fields. The profile entry list uses a record-by-record structure, with each record containing an entry identifier, feature type, standard field key, field value reference, sensitivity level flag, source flag, and update timestamp. The entry identifier remains unique within the same session and is generated by combining the feature type and standard field key. The profile aggregation service outputs a session number and a profile entry list reference identifier for the next step to read and match the shared scope list and the desensitization caliber table.

[0020] S4: Based on the scene identifier and model identifier matching and sharing rules, determine the sharing scope and desensitization criteria of the item list, generate the shared payload and write it into the sharing identifier and version identifier. The specific implementation is as follows: After the profile item list and session number are associated and written, the rule matching service reads the session metadata using the session number as an index, obtains the business scenario identifier, touchpoint identifier, entry channel identifier, enterprise dimension identifier, and target model identifier, and reads the profile item list reference identifier to locate the set of profile items available for this session; the rule base and rule mapping table are maintained and versioned by the configuration center according to the enterprise dimension identifier and the runtime environment dimension, and the rule matching service loads the corresponding version and performs matching; the rule base maintains the scenario rule set according to the business scenario dimension and the model rule set according to the model dimension, and the rule mapping table records the association between the scenario rule set and the model rule set; the scenario rule set limits the types of profile elements that can be referenced and the range of standard field keys, and the model rule set provides input constraints and the set of available fields to limit the range of fields that the model side can accept and the field expression requirements; The rule matching service locates the scenario rule set based on the business scenario identifier and filters applicable rule entries according to the priority fields and matching conditions of the scenario rule entries. The matching conditions include touchpoint identifiers and entry channel identifiers. The rule matching service retrieves the associated model rule entries in the rule mapping table based on the target model identifier, obtains the model input constraint identifier and the model available field set identifier, and selects the rule entries that are compatible with the target model in the scenario rule set. The compatibility judgment adopts deterministic conditions to verify whether the field set declared by the scenario rule entry is covered by the model available field set, and to verify whether the sharing granularity declared by the scenario rule entry meets the field expression requirements of the model side. After the rule entries are determined, the rule matching service generates a sharing scope list. The sharing scope list records the set of profile element types that can be shared, the set of standard field keys that can be shared, and the sharing granularity information. The sharing granularity indicator field is expressed at the plaintext field level, summary level, or citation level. After the shared scope list is generated, the rule matching service reads the sensitivity level markers in the profile item list and combines them with the sensitivity level dictionary and scenario rule items to determine the de-identification criteria and generate a de-identification criteria table. The de-identification criteria table uses standard field keys as indexes and sets processing methods and processing parameters for each field. The processing methods cover one or more of replacement, masking, generalization, and segmented retention, and the corresponding processing strategies are configured according to the sensitivity level. The processing parameters and processing priorities are defined in the de-identification rule items of the scenario rule items. When the de-identification rule items and the sensitivity level dictionary provide criteria for the same field at the same time, the final criteria are determined according to the processing priority, and the criteria source marker and conflict handling marker are written. The rule matching service generates a shared identifier and a version identifier for this session and binds the shared identifier to the session number to identify the shared scope instance corresponding to this session. The version identifier is generated by combining the rule version and the field directory version. The rule version corresponds to the version of the scenario rule set and the model rule set used in this session, and the field directory version corresponds to the version scope of the portrait element directory and the standard field key set. The field directory version is obtained from the directory version number or the portrait element directory version identifier in the extended field of the session metadata. The rule matching service uses an atomic write method to write the shared scope list, the desensitized scope table, the shared identifier, and the version identifier to the session storage, and writes the shared identifier reference and the version identifier reference in the extended field of the session metadata for subsequent steps to read the shared scope list and the desensitized scope table and generate the shared payload.

[0021] S5: Format and trim the shared payloads according to the input constraints of the target model, inject the session context into the target model, and record the call link identifier and payload summary. The specific implementation is as follows: After the shared scope list and the de-identification caliber table are written to the session storage, the shared payload generation service reads the shared identifier, version identifier, business scenario identifier, and target model identifier from the session storage using the session number as the entry point, and obtains the corresponding shared scope list and de-identification caliber table based on the shared identifier reference. The shared payload generation service also reads the profile item list reference identifier and loads the profile item set. The caliber priority table, pruning rules, reason code mapping table, business scenario combination template, and payload template configuration items are managed hierarchically by the configuration center according to the enterprise dimension identifier and the operating environment dimension, and optional items are configured at the business scenario dimension. The configuration items have version identifiers. The shared payload generation service and the model call gateway load the configuration items matching the version for processing in this step. The shared payload generation service determines the extraction scope based on the shared scope list and locates the corresponding entries in the portrait entry set according to the standard field key set defined by the shared scope list. It extracts the entry identifier, feature type, standard field key, field value reference, and sensitivity level marker. For missing field entries, it performs placeholder or skip processing according to the shared granularity rules and writes a missing marker. After completing the entry extraction, the shared payload generation service performs calibration processing on the field value references according to the desensitization calibration table. Calibration processing is performed item by item using the standard field key as the index: when a replacement rule is hit, the field value reference is replaced with a template value or category value; when a masking rule is hit, a masking expression is generated according to the masking rule; when a generalization rule is hit, the field value reference is converted to a higher-level category according to the hierarchical dictionary; when a segment retention rule is hit, the specified segment is retained according to the segment position, and the remaining segments are masked. When the same standard field key hits multiple calibration rules simultaneously, the final processing method is determined according to the calibration priority table, and the calibration entry identifier is written in the processing record. The shared load generation service organizes the field set that has undergone caliber processing into shared loads. Shared loads are expressed in the form of structured text or key-value pair sequences. The structured text is organized into segments according to feature type, and within each segment, fixed-format entries are generated based on the field order and splicing rules determined by the load template configuration items. Key-value pair sequences are arranged by field key and field value pairs, and include source tags and update timestamp references. The shared load generation service reads the model input constraint configuration based on the target model identifier. The model input constraint configuration includes the maximum input length, field order requirements, a list of prohibited fields, and a list of required fields for the scenario. The maximum input length is specified in the measurement parameters. The length of the shared load is determined by the model input constraint configuration during the load processing stage. During load processing, the retention priority of the fields is first determined based on the list of required fields for the scenario, and then the field keys are sorted according to the field order requirements. Field keys that hit the prohibited field list are removed. When the shared load exceeds the maximum input length, the pruning rules exclude field keys in the list of required fields for the scenario and exclude field keys marked as non-pruning in the rule base. Then, the remaining field keys are removed in order of retention priority from low to high until the maximum input length constraint is met. The set of removed field keys and the pruning reason code are written in the pruning record. During the payload preparation phase, context fragment boundaries are set based on business scenario identifiers and business scenario combination templates. The insertion positions and combination order of shared payloads and session history message fragments are determined, and boundary markers and segment numbers are written for each segment. After the shared payload preparation is completed, the model invocation gateway receives the session number and shared payload, reads session history message fragment references from the session storage, merges the shared payload and history message fragments into model input content according to the segment number and combination order, and generates a call link identifier. The call link identifier is bound to the session number and written with a processing node marker. The processing node marker is used to mark the node information of the unified entry point, profile aggregation, rule matching, payload generation, and model invocation stages. The model invocation gateway writes the call link identifier to the session storage and also writes it to the payload summary field. The payload summary field writes the structure summary of the shared payload, the field key set summary, and the pruning record reference information, and writes the version identifier reference and the shared identifier reference for subsequent session archiving and profile update processes.

[0022] S6: When a session ends or an update condition is triggered, summarize the interaction logs and profile change information, update the profile entry version, and archive shared records and audit indexes. The specific implementation is as follows: After the model calls the gateway to combine the shared payload and session history message fragments and enters the dialogue processing, the session management service establishes an interaction log stream with the session number and continuously writes log records in the order of message arrival. The log records include message sequence number, message role tag, timestamp, touchpoint identifier, business scenario identifier, link identifier, and entry channel identifier. The message sequence number adopts a monotonically increasing sequence, the message role tag distinguishes between user-side input and system-side output, the timestamp adopts the time base agreed upon by the session metadata, and the touchpoint identifier and business scenario identifier reference the session metadata entries. The session management service maintains the session status field, which records the session start, in progress, and end status and the corresponding time point. The session extension field also writes a payload summary reference, a sharing identifier reference, and a version identifier reference, which are used to locate the association information between the shared scope and the model input payload by the session number. When a session ends or the session management service detects that the profile update conditions are met, the session management service triggers the profile update process. The profile update conditions are configured as a deterministic set of conditions according to the business scenario. This set includes one or more of the following: key field change markers from the customer information source, markers for newly added tags or preference elements within the session, and markers for repeated sessions across entry points. Key field change markers are generated by the customer information source change notification or by the session management service performing a pull comparison based on field rule items. The pull comparison uses the current profile entry version corresponding to the search key in the profile storage as the baseline and the current field value returned by the customer information source as the comparison, generating a set of changed field keys and writing a change timestamp. Newly added tags or preference element markers within the session are written to the interaction log by the dialogue access or model output according to the agreed-upon tag format. The session management service parses the interaction log based on the tag field mapping items to generate a set of new tag field keys. The tag field mapping item is issued by the configuration center, which limits the field key and value scope of the tag fragment in the interaction log; the cross-entry duplicate session trigger flag is obtained by the session management service by counting the session count corresponding to the search key within the time window. The statistical object is the session metadata record associated with the search key. The counting range is limited by the time window, and session records that have not established session metadata or have been marked as invalid are excluded based on the session status field; the time window and trigger threshold are maintained by the configuration center according to the business scenario, enterprise dimension identifier and operating environment dimension and have version identifiers. The session management service loads the corresponding version and performs the trigger judgment; when any flag meets the trigger rule, the session management service sends an update task to the profile update service. The update task carries the session number, search key, business scenario identifier, shared identifier reference, version identifier reference, payload summary reference, link identifier, shared scope list reference and desensitization table reference; After receiving the update task, the profile update service reads the interaction log index and payload summary from the session storage, and determines the profile element directory version based on the directory version number or version identifier reference in the extended fields of the session metadata. The profile update service uses the standard field key set corresponding to the directory version as the matching criterion to perform field key mapping on the change fragments, tag fragments and business status fragments in the interaction log, generates candidate change items, and merges them with the field key set corresponding to the key field change marker to remove duplicates. The profile update service generates a profile entry version update record for the merged candidate change items. The version update record writes the entry identifier, standard field key, change type, change timestamp, source marker and directory version marker, and writes the version update record back to the profile storage, completing the database association of the profile entry version update record. After the profile update service completes the write-back, it writes audit index records to the audit index database. The audit index records include session number, search key, sharing identifier, version identifier, shared scope list reference, de-identification table reference, payload summary reference, link identifier, and interaction log index, and also include a time slice field for segmented retrieval. The audit index database establishes a primary index with session number and search key, and a secondary index with time slice and version identifier, supporting retrieval and tracing by session number, search key, time slice, and version identifier.

[0023] In this embodiment, the enterprise first deploys a dialogue access service at a unified entry point to receive dialogue requests from multiple channels. It verifies identity credentials and performs replay checks according to the entry channel, parses the customer's unique identifier, generates a session number and link identifier, and establishes a binding record before writing it to the session storage using idempotent rules. The session management service reads the binding record to generate session metadata entries, generates search keys based on the customer's unique identifier and the enterprise dimension identifier, and simultaneously writes touchpoint identifiers and business scenario identifiers, establishing an index relationship. The profile aggregation service, based on the search keys, pulls profile elements from the customer information source according to the profile element catalog, completes field mapping, conflict merging, and sensitivity level labeling, forming a profile entry list. The matching service performs the prescribed matching according to the business scenario and target model, generates a shared scope list and a desensitization caliber table, and writes the shared identifier and version identifier into the session storage; the shared payload generation service extracts portrait entries according to the shared scope and processes them according to the desensitization caliber, and completes the payload formatting and trimming by combining the model input constraints. The model calls the gateway and session history fragments to combine them, and records the payload summary and call chain; when the session ends or the update conditions are met, the session management service triggers the portrait update process. The portrait update service generates portrait entry version update records based on the interaction log and change markers and writes them back to the portrait storage, and writes them into the audit index for subsequent retrieval and traceability.

[0024] Example 2: Figure 2 A cross-model customer profile collaborative update intelligent customer service system is presented, including: Dialogue Access and Identity Authentication Module: Unifies access to dialogue requests from multiple channels, verifies credentials by channel and determines replay, parses customer identifiers, generates session numbers and link identifiers, and writes binding records to session storage; Session metadata and search key management module: Reads bound records to generate session metadata, generates search keys based on customer unique identifiers and enterprise dimensions, writes touchpoints and scenario identifiers and establishes index associations; Customer profile aggregation and item list generation module: Access customer information source based on search key, pull elements according to profile element catalog and complete field mapping, conflict merging and sensitivity level labeling, and generate profile item list associated with session number; Shared rule matching and definition module: Based on the business scenario identifier and target model identifier matching rule library, it generates a shared scope list and a de-identified definition table, forms a shared identifier and a version identifier, and writes them into session storage; Shared payload generation and model adaptation injection module: Extracts and desensitizes profile entries according to the shared scope, formats and trims them according to model constraints, generates input by splicing historical fragments according to templates, and records link identifiers and payload summaries; Interaction Log Collection and Session State Management Module: Records interaction logs and session state around the session number, solidifies message sequence number, role, timestamp, touch point and scenario definition, and maintains payload summary and sharing, version reference; The profile update and version chain management module receives tasks when a session ends or an update is triggered, locates the update item based on the interaction log and change marker, generates a version record, and writes it back to the profile storage. Audit Index and Archive Tracing Module: Summarizes session numbers, search keys, and shared related references and writes them into the audit index database, organized by primary and secondary indexes, for retrieval tracing and scope review; Scene Graph and Data Object Management Module: Maintains the mapping relationship between scenes and data objects, used to determine the available data objects and data retrieval range for a scene; Context key-value storage module: provides exact matching keys and model-independent universal keys for cross-model reuse and secondary compression conversion; The Measurement and Operation Iteration Module collects feedback metrics and drives the iteration of scenarios, data relationships, and template strategies to continuously optimize consistency and efficiency.

[0025] It should be noted that this invention can be deployed on the device itself to realize embedded applications, or it can run on a PC or other terminal with a user interface, thereby meeting various hardware environments and usage requirements.

[0026] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wireless or wired transmission; wired transmission methods include optical fiber, twisted pair, coaxial cable, etc.; wireless transmission includes infrared, microwave, etc. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center containing one or more sets of available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium. A semiconductor medium can be a solid-state drive.

[0027] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and modules described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0028] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or modules may be electrical, mechanical, or other forms.

[0029] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple network modules. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0030] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0031] If the aforementioned functions are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0032] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0033] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for collaboratively updating intelligent customer profiles across models, characterized in that, include: S1: After receiving a dialogue request at the unified enterprise portal, perform identity authentication, obtain the customer's unique identifier, generate a session number, and establish a binding record between the customer's unique identifier and the session number; S2: Write the binding record into the session metadata, generate a search key that can be reused across entry points, and write the touchpoint identifier and business scenario identifier into the session metadata; S3: Based on the search key, read the profile elements in the customer information source, classify and label the profile elements according to type and sensitivity level, and generate a list of profile items to be shared; S4: Based on the matching and sharing rules of scene identifiers and model identifiers, determine the sharing scope and desensitization criteria of the item list, generate the shared payload, and write the shared identifier and version identifier; S5: Format and trim the shared payload according to the input constraints of the target model, inject the session context into the target model, and record the call link identifier and payload summary; S6: When the session ends or an update condition is triggered, summarize the interaction logs and profile change information, update the profile entry version, and archive the shared records and audit index.

2. The intelligent customer service method for cross-model collaborative updating of customer profiles according to claim 1, characterized in that, After receiving a dialogue request through the unified enterprise portal, the system performs identity authentication, obtains the customer's unique identifier, generates a session number, and establishes a binding record between the customer's unique identifier and the session number, including: The login token, single sign-on ticket, or short-term access credential are verified separately according to the entry channel, and the replay is determined based on the request random number and request timestamp. The determination parameters are given by the versioning configuration item. The customer identifier is mapped to a unified customer primary key, a session number and a link identifier are generated and deduplication is performed, the binding record and initial metadata are written by an idempotent key, abnormal replay terminates, and conflict callback is returned.

3. The intelligent customer service method for cross-model collaborative updating of customer profiles according to claim 1, characterized in that, Write the binding record to the session metadata, generate a search key that can be reused across entry points, and write the touchpoint identifier and business scenario identifier into the session metadata, including: The session management service reads the bound records by session number, generates session metadata entries by field model, and unifies the time field to the same time base. A search key is generated using the customer's unique identifier and the enterprise's dimension identifier, and a two-way index is established. Other identifiers are written into the alias extension field and do not participate in the search key generation. Touchpoint identifiers and business scenario identifiers are determined and recorded based on entry priority, mapping completion, or default completion, and source conflicts are recorded. Link mapping and version number are atomically written and registered according to versioning rules.

4. The intelligent customer service method for cross-model collaborative updating of customer profiles according to claim 1, characterized in that, Based on the search key, profile elements are retrieved from customer information sources. These elements are then categorized and labeled according to type and sensitivity level, generating a list of profile items to be shared, including: The profile aggregation service reads the search key, enterprise dimension identifier, and entry channel identifier by session number, determines the access order of information sources, and uses the customer's primary key as the access criterion. When the primary key is missing in the search key, perform master data mapping and record the source time; Load the versioned directory and mapping configuration, and pull and merge fields according to the validity period. When a failure or missing item occurs, it is marked according to the status code mapping table and integrity rules, switched according to priority, and the collection log is recorded. Generate standard field keys and field value references in the order of conflict, mark the sensitivity level, and write them into the list summary, directory version number, and reference identifier.

5. The intelligent customer service method for cross-model collaborative updating of customer profiles according to claim 1, characterized in that, Based on the scene identifier and model identifier matching and sharing rules, the sharing scope and desensitization criteria of the item list are determined, a shared payload is generated, and the shared identifier and version identifier are written into it, including: The rule matching service reads the scene identifier, touchpoint identifier, entry channel identifier, target model identifier, and profile entry set by session number, and loads the versioned rule base and rule mapping table; Filter scenario rule entries according to the priority conditions corresponding to the touch point identifier and the entry channel identifier, perform compatibility verification with the model rule entries corresponding to the target model, generate a shared scope list and a desensitization caliber table, and record the source of conflict. Generate a shared identifier and combine the rule version with the directory version to form a version identifier. The directory version is taken from the session extension field and atomically written to the session storage.

6. The intelligent customer service method for cross-model collaborative updating of customer profiles according to claim 1, characterized in that, The shared payload is formatted and trimmed according to the input constraints of the target model, the session context is injected into the target model, and the call chain identifier and payload summary are recorded, including: The payload generation service reads the shared identifier, version identifier, scene identifier, and model identifier by session number, and loads the versioned template, caliber priority table, and pruning rules. Entries are extracted based on the shared scope. If a missing entry is found, it is either placed in place or skipped according to the shared granularity and a missing tag is written. Field value references are processed according to the desensitization criteria. When multiple rules are matched, the final criteria are determined according to the criteria priority and the criteria entry identifier is recorded. First, exclude required fields and fields that cannot be clipped according to model constraints, then clip them in stages and write the reason codes. The fragment boundaries and paragraph numbers are set according to the scene template. The model calls the gateway to splice historical fragments and writes the link node tags and payload summary references.

7. The intelligent customer service method for cross-model collaborative updating of customer profiles according to claim 1, characterized in that, When a session ends or an update condition is triggered, the interaction logs and profile change information are aggregated, profile entry versions are updated, and shared records and audit indexes are archived, including: The session management service writes interaction logs by session number and maintains session status, writing payload summary references, shared identifier references, and version identifier references; When a session ends or triggers a key field change flag, tag flag, or cross-entry session count flag, an update task is generated according to the versioned time window and threshold, carrying the retrieval key, link identifier, shared scope list reference, and desensitization caliber table reference. The profile update service maps log fragments to directory versions and writes them to version update records after deduplication. It also synchronously writes audit index records containing time slices and establishes primary and secondary indexes.

8. A cross-model customer profile collaborative update intelligent customer service system, used to implement the cross-model customer profile collaborative update intelligent customer service method according to any one of claims 1-7, characterized in that, include: Dialogue Access and Identity Authentication Module: Unifies access to dialogue requests from multiple channels, verifies credentials by channel and determines replay, parses customer identifiers, generates session numbers and link identifiers, and writes binding records to session storage; Session metadata and search key management module: Reads bound records to generate session metadata, generates search keys based on customer unique identifiers and enterprise dimensions, writes touchpoints and scenario identifiers and establishes index associations; Customer profile aggregation and item list generation module: Access customer information source based on search key, pull elements according to profile element catalog and complete field mapping, conflict merging and sensitivity level labeling, and generate profile item list associated with session number; Shared rule matching and definition module: Based on the business scenario identifier and target model identifier matching rule library, it generates a shared scope list and a de-identified definition table, forms a shared identifier and a version identifier, and writes them into session storage; Shared payload generation and model adaptation injection module: Extracts and desensitizes profile entries according to the shared scope, formats and trims them according to model constraints, generates input by splicing historical fragments according to templates, and records link identifiers and payload summaries; Interaction Log Collection and Session State Management Module: Records interaction logs and session state around the session number, solidifies message sequence number, role, timestamp, touch point and scenario definition, and maintains payload summary and sharing, version reference; The profile update and version chain management module receives tasks when a session ends or an update is triggered, locates the update item based on the interaction log and change marker, generates a version record, and writes it back to the profile storage. Audit Index and Archive Tracing Module: Summarizes session numbers, search keys, and shared related references and writes them into the audit index database, organized by primary and secondary indexes, for retrieval tracing and scope review; Scene Graph and Data Object Management Module: Maintains the mapping relationship between scenes and data objects, used to determine the available data objects and data retrieval range for a scene; Context key-value storage module: provides exact matching keys and model-independent universal keys for cross-model reuse and secondary compression conversion; The Measurement and Operation Iteration Module collects feedback metrics and drives the iteration of scenarios, data relationships, and template strategies to continuously optimize consistency and efficiency.

Images