A multi-touchpoint approach to intelligent customer service in digital systems based on artificial intelligence

Abstract

The present invention relates to the field of multi-channel intelligent customer service technology, specifically a multi-touch method for intelligent customer service of a digital system based on artificial intelligence, comprising the following steps: based on multi-channel user request entrances, analyzing operation time, contact identification and data structure, judging text and voice content features, identifying standardizable fields, matching keywords and knowledge tags, analyzing user portraits and behaviors, identifying permissions and sensitive items, and outputting desensitized permission results. In the present invention, by combining user behavior parameters under multi-channel entrances, feature classification and standardized organization can be performed for different entrance data, the accuracy of request identification can be improved through multi-dimensional parameter fusion, supporting the effective aggregation of cross-channel user identities and request sequences, promoting deep analysis of content semantics, achieving a high degree of matching between content structure and knowledge tags, automatically generating personalized service instructions and permission identification, and realizing sensitive content identification and secure output.

Description

Technical Field

[0001] The present invention relates to the technical field of multi-channel intelligent customer service, and in particular to a multi-touchpoint method for intelligent customer service of a digital system based on artificial intelligence. Background Art

[0002] The field of multi-channel intelligent customer service technology involves enterprises providing users with unified and efficient customer support services through multiple digital service channels. This field includes supporting users to initiate consultations and service requests through different entrances such as APP, official websites, social platforms, and telephones, processing various data forms such as text and voice generated by users in different channels, and aggregating and answering user questions from these channels to achieve the reception, diversion, answering and feedback of questions in the service process. Among them, the traditional digital system intelligent customer service multi-touchpoint method refers to the fact that enterprises usually provide service support to users in different channels by setting up telephone customer service, website online customer service, email support, social platform messages, etc., generally using keyword matching or FAQ search rules, and realizing the identification and automatic response of user questions by maintaining a question keyword library and preset answers to common questions. For problems that cannot be solved automatically, the user request information will be recorded and transferred to manual customer service for processing based on historical records and manual retrieval of relevant information.

[0003] Traditional multi-channel intelligent customer service lacks the ability to uniformly aggregate data from different entry points. The processing logic is limited to basic keyword retrieval or templated responses. Inconsistent data standards between channels make it difficult to effectively aggregate information. Session tracking is difficult when users switch identities or make multiple requests. Content analysis is difficult to go deep into the semantic level. The business diversion link does not respond promptly in terms of permission judgment and sensitive data control. In actual applications, this often leads to problems such as service chain breaks, data synchronization delays, and insufficient user feedback responses, affecting the overall processing efficiency in complex service environments. Summary of the Invention

[0004] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a multi-touchpoint method for intelligent customer service of a digital system based on artificial intelligence.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a multi-touchpoint method for intelligent customer service in a digital system based on artificial intelligence, comprising the following steps:

[0006] S1: Based on multi-channel user request entry points, analyze the operation time and touchpoint ID distribution, determine the differences in text structure of each entry point, perform consistency detection on voice content, compare device type mapping performance, identify data fields that can be standardized, and obtain unified request structure characteristics;

[0007] S2: Based on the unified request structure characteristics, analyze the mapping between identity tags and data numbers, determine the matching relationship between session fields and channel sources, compare the request time interval distribution, adjust the channel path processing order, prioritize requests according to their criticality, and obtain the channel master control sequence;

[0008] S3: Based on the channel master sequence, filter the keywords in the content field, analyze the semantic structure of the keywords and knowledge tags, determine the synergy between the subject-predicate structure and the tag content through semantic pairing of the keywords and tag indexes, locate the tag item with the best semantic association, and obtain the tag matching code;

[0009] S4: According to the label matching code, analyze the distribution of usage scenario keywords in the user portrait, screen the correlation between the channel identifier and the access trajectory, compare the adaptation degree of each channel portrait parameter, judge the performance of the request behavior in each channel, and obtain the channel portrait matching factor.

[0010] The improvements of the present invention are that the unified request structure features include contact type identification, data integration template, channel format type, the channel master control sequence includes request priority coding, session diversion mark, channel sequence index, the label matching code includes label number, semantic label index, content label parameter, and the channel portrait matching factor includes user feature code, scenario adaptation label, and behavior-oriented feature.

[0011] The present invention is improved in that the steps of obtaining the unified request structure feature are specifically as follows:

[0012] S111: Based on multi-channel user request entry points, analyze time records and touchpoint identifiers, aggregate user request times and categories across channels, compare the time concentration of user requests and touchpoint overlap within each channel, determine changes in active segments across channels, and obtain channel touchpoint activity distribution parameters;

[0013] S112: Based on the channel touchpoint activity distribution parameters, determine the format structure of the request data collected from each channel, analyze the data node hierarchy and naming differences, compare the differences in text structure between the app, official website, and social platforms, and screen key fields based on the consistency between the voice content and the text request to obtain a multi-channel structure matching coefficient;

[0014] S113: Based on the multi-channel structure matching coefficient, optimize the correspondence between the device type parameters and the standard fields, analyze the terminal operating system, device model and browser type, adjust the mapping path between the device parameters and the standard data, compare the correspondence of the fields generated by each device, and calculate the overlap ratio of the multi-channel fields to obtain a unified request structure feature.

[0015] The present invention is improved in that the steps of obtaining the channel master control sequence are specifically as follows:

[0016] S211: Based on the unified request structure characteristics, the channel format type and the contact type identifier are analyzed, the correspondence between the user identity tag and each data number is compared, number duplication and mapping overlap are determined, the data attribution mapping is optimized, and the data attribution mapping strength is obtained;

[0017] S212: Based on the data attribution mapping strength, analyze the session field and channel source field of each request, compare the position index in the sorting sequence, determine the consistency of the channel source field and the sorting, adjust the field combination, and obtain the sorting channel synergy coefficient;

[0018] S213: Based on the sorting channel synergy coefficient, calculate the joint performance of the request time interval parameter and the channel sequence index, determine the impact of the emergency flag state on the sorting, obtain the request priority sorting range, adjust the channel path sequence, and obtain the channel master control sequence.

[0019] The present invention is improved in that the steps of obtaining the tag matching code are specifically as follows:

[0020] S311: Based on the channel master control sequence, compare the associations of the subject-predicate structure field, the word order field, and the sentence structure field, determine the combination of subject-predicate keywords and semantic orientations, identify keyword groups with key semantic centers, and obtain semantic structure keyword groups;

[0021] S312: Based on the semantic structure keyword group, optimize its matching with the knowledge tag index field, analyze the semantic distance and word meaning offset of the keyword group in the tag index, calculate the pairing strength distribution between the keyword and the tag, and obtain the semantic pairing strength distribution;

[0022] S313: Based on the semantic pairing strength distribution, determine the degree of coordination between the subject-predicate structure and the label content, obtain a label coordination coupling factor, and arrange the candidate label items in order according to the subject-predicate matching according to the factor to obtain a label matching code.

[0023] The present invention is improved in that the steps of obtaining the channel portrait matching factor are specifically as follows:

[0024] S411: Based on the tag matching code, the associated tag number and content tag parameter are analyzed, and the frequency of occurrence of each keyword in the user portrait is calculated in combination with the usage scenario keywords appearing in the user portrait, thereby generating a scenario keyword distribution feature;

[0025] S412: Based on the scene keyword distribution characteristics, filter the access trajectory fields related to the channel identifier, compare the correlation between the trajectory fields and the record fields in the channel access sequence, calculate the intersection ratio of each field, and obtain the channel trajectory field interaction structure;

[0026] S413: Based on the channel trajectory field interaction structure, compare the number of field categories, field expression structure characteristics and distribution density of user portrait parameters in each channel under the multi-scenario data structure, judge the joint distribution range of structurally consistent fields, determine the key data items of structural overlap, and obtain the channel portrait matching factor.

[0027] The present invention is improved in that the steps further include:

[0028] S5: Based on the channel profile matching factor, identify the permission level label in the content structure, analyze the association between user identity authentication parameters and content permissions, compare the permission level differences, identify the content that needs to be desensitized, adjust the processing node data, and obtain the desensitized permission output result;

[0029] The desensitized permission output result includes content desensitization label, permission discrimination parameters, and data feedback instructions.

[0030] The present invention is improved in that the steps for obtaining the desensitized permission output result are specifically as follows:

[0031] S511: Based on the channel profile matching factor, identify the location of the permission level label in the content structure, compare the hierarchy, order and association relationship of each permission label in the content field, and obtain the permission label hierarchy sequence;

[0032] S512: calling the permission label hierarchical sequence, comparing it with the user identity authentication parameters, determining the association between the permission label and the identity identifier, authentication method and authorization level in the content structure, and integrating the structure mapping to obtain the identity permission association structure;

[0033] S513: According to the identity authority association structure, the content fields involved are screened, the field access path and sensitive field identifiers are analyzed, the paths and identifiers of the fields whose authority labels are higher than the authentication level are adjusted, the node data is optimized and processed, and a desensitized authority output result is obtained.

[0034] Compared with the prior art, the advantages and positive effects of the present invention are:

[0035] In the present invention, by combining user behavior parameters under multiple channel entrances, feature classification and standardized organization can be performed for different entrance data, the accuracy of request identification can be improved through multi-dimensional parameter fusion, and the effective aggregation of cross-channel user identities and request sequences can be supported. Deep analysis of content semantics can be promoted, and a high degree of matching between content structure and knowledge tags can be achieved. Individualized service instructions and permission identification can be automatically generated to achieve sensitive content identification and secure output. The contextual consistency and response consistency of the service link can be guaranteed through multi-node data linkage throughout the entire process, meeting the data circulation and control requirements in complex business processes and targeted service scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a flow chart of the main steps of the present invention;

[0037] Figure 2 A flowchart for obtaining unified request structure features in the present invention;

[0038] Figure 3 This is a flow chart for obtaining the channel master control sequence in the present invention;

[0039] Figure 4 This is a flowchart for obtaining tag matching codes in the present invention;

[0040] Figure 5 This is a flow chart for obtaining the channel portrait matching factor in the present invention;

[0041] Figure 6 This is a flowchart for obtaining the desensitized permission output results in the present invention. DETAILED DESCRIPTION

[0042] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0043] In the description of the present invention, it should be understood that the terms "length", "width", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating directions or positional relationships, are based on the directions or positional relationships shown in the accompanying drawings and are only for the convenience of describing the present invention and simplifying the description. They do not indicate or imply that the devices or elements referred to must have a specific direction, be constructed and operate in a specific direction, and therefore should not be understood as limiting the present invention. In addition, in the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0044] Example

[0045] See also Figure 1 The present invention provides a technical solution: a multi-touch method for intelligent customer service in a digital system based on artificial intelligence, comprising the following steps:

[0046] S1: Based on multi-channel user request entry points, analyze the distribution patterns of operation time and touchpoint identifiers, determine the structural differences of text formats across multiple channels, perform content consistency checks on voice structures, compare the mapping of device type parameters in various data entries, identify data fields that can be standardized across multiple channels, and obtain unified request structure features;

[0047] S2: Based on the unified request structure characteristics, analyze the impact of identity tags on the data number mapping structure, determine the match between the channel source pointed to by the session field and the request sequence, compare the optimal distribution of the time interval parameter, optimize the Boolean expression of the emergency flag parameter in the sequence, adjust the processing order of each channel path, and prioritize the requests according to their criticality to obtain the channel master control sequence;

[0048] S3: Based on the channel master sequence, filter the keyword features in the content field, analyze the semantic structure of keywords and knowledge tags, and determine the degree of coordination between the subject-predicate structure and the tag content by comparing the semantic pairing relationship between keywords and tag indexes. Optimize the tag determination method and locate the tag item with the best semantic relevance to obtain the tag matching code.

[0049] S4: Based on the tag matching code, analyze the weight distribution of scenario keywords used in the user portrait, screen the correlation between the channel identifier and the access trajectory field, compare the adaptability of each channel portrait parameter in multiple scenarios, determine the parameter performance of user request behavior in each channel, determine the key data items of structural overlap, and obtain the channel portrait matching factor;

[0050] S5: Based on the channel portrait matching factor, identify the position of the permission level label in the content structure, analyze the correlation between user authentication parameters and content permissions, compare the structural differences between permission levels, optimize the identification process of sensitive fields in content data, identify content items that need desensitization protection, and adjust the data of each processing node to obtain the desensitized permission output result.

[0051] The unified request structure features include contact type identification, data integration template, and channel format type. The channel master control sequence includes request priority coding, session diversion tag, and channel sequence index. The label matching code includes label number, semantic label index, and content label parameter. The channel portrait matching factors include user feature code, scenario adaptation tag, and behavior pointing feature. The desensitized permission output results include content desensitization tag, permission judgment parameter, and data feedback instruction.

[0052] In S1, distribution pattern refers to the statistical characteristics of user requests such as frequency, time concentration, and dispersion at different time nodes and different channel entrances (such as APP, official website, social platform, telephone, etc.), which are often used to analyze the time characteristics and channel preferences of user access behavior; structural difference refers to the different data structures of the content during the transmission of text format on different channels, such as JSON submitted in APP, form parameters on the website, and message format on the social platform. The system needs to identify and process the differences; device type parameters refer to the attribute information of the terminal device that initiates the user request, such as the device operating system (iOS, Android), hardware model, browser type, etc., which are used for subsequent data standardization and adaptation; mapping expression: refers to the specific expression of different device type parameters corresponding to the internal standard fields of the system in the data record. For example, the data field names transmitted by the same function on different devices are different, and the system needs to perform parameter mapping; standardized data fields refer to the standardized data field set that can be recognized and utilized by the subsequent intelligent customer service processing engine after unified format conversion and content extraction.

[0053] In S2, identity tags refer to information fields used to identify user identities or permission levels, such as user ID, membership level, and real-name status, to help distinguish different user requests and manage sessions; data number mapping structures refer to unique numbers assigned to different requests, as well as the correspondence structure between numbers and original request data and user information, used to track and manage multiple requests; matching refers to the degree of correspondence or adaptation between the request channel (such as APP, official website, etc.) reflected by the session field and the current request in the sorting, diversion, and aggregation processes; optimal distribution refers to the distribution status of multiple requests on the timeline, through which the best processing time or sorting order can be selected; emergency flag parameters refer to parameters or fields in the request content used to indicate whether they are urgently processed (such as labels such as "urgent" and "high priority"); Boolean expression: refers to the storage and judgment of the above-mentioned emergency flags or status in the system through Boolean (true / false, yes / no) fields, simplifying subsequent priority calculations.

[0054] In S3, keyword features refer to words and phrases extracted from user request content that can represent the core intent or theme, and are used to match knowledge base labels; semantic structure refers to the structural information of the request text in terms of grammar, part of speech, syntactic relations, etc., which is used to understand user intent and problem description; semantic pairing relationship refers to the correspondence between content and knowledge labels by comparing the semantic correlation between user request keywords and system knowledge labels; the degree of coordination refers to the logical and semantic matching or coupling between the subject-predicate structure, request intent and knowledge label content, reflecting the accuracy of automatic label classification; label determination method refers to the method process of determining the most appropriate knowledge label based on specific rules (such as semantic distance, keyword matching, etc.); the label item with the best semantic correlation refers to the knowledge label with the highest semantic correlation with the current request content among all optional labels.

[0055] In S4, weight distribution refers to the frequency and proportion of various usage scenario keywords in the user portrait in the overall portrait data, which is used to characterize user behavioral preferences; channel portrait parameters refer to the set of parameters such as user behavior data, historical interactions, preference configuration, etc. in different channels, which constitute the multi-dimensional portrait of the user on each channel; the degree of adaptation in multiple scenarios refers to the ability to match and adapt the portrait parameters displayed by users in different service scenarios or channels, and measures the effect of the fusion and utilization of portrait data from different channels; parameter performance refers to the specific behavioral characteristics or field changes reflected by user requests in each channel, such as request frequency, content type, response behavior, etc.; structural overlap refers to the degree to which different data fields can match and overlap with each other in structure under multiple channels, which is used to determine which data items are most suitable for joint processing and knowledge response.

[0056] In S5, the permission level label refers to the access permission level or security identifier (such as "visible only internally" or "VIP user") marked in the content data, which is used to control the access scope of the content; correlation refers to the degree of correlation between the user authentication parameters and the content permission label in terms of attributes, level, authorization scope, etc.; structural difference refers to the differences in data structure and expression between different permission level codes, such as digital codes, string identifiers, hierarchical levels, etc.; identification process refers to the specific operational steps for automatically identifying and marking fields that require special processing (such as desensitization, restricted access, etc.); processing node data refers to the intermediate information such as the processing status, change identification, operation records, etc. of the data content in each link of the system in the data processing chain.

[0057] S1 mainly collects and analyzes user requests from different channels, uniformly classifies operation time, entry logo, text and voice structure and device type, standardizes the format, and ensures that multi-source data can be processed uniformly; S2 mainly performs identity tag analysis, number mapping, channel and sequence comparison, and time and urgency sorting on the unified requests to improve the management efficiency and diversion accuracy of requests; S3 mainly extracts keywords from the request content, analyzes the semantic structure, and performs intelligent matching based on knowledge tags to automatically determine the most suitable knowledge tags to support efficient and accurate automatic responses; S4 extracts user scenarios and behavioral characteristics based on matching tags, combined with user portraits and their behavioral performance in various channels, to achieve targeted and detailed services; S5 identifies, desensitizes and securely processes the permission levels and sensitive fields of the response content to ensure data compliance and information security, and outputs service results that meet permission requirements.

[0058] See also Figure 2 ,The specific steps for obtaining the unified request structure features are:

[0059] S111: Based on multi-channel user request entry points, analyze time records and touchpoint identifiers, aggregate user request times and categories across channels, compare the time concentration of user requests and touchpoint overlap within each channel, determine changes in active segments across channels, and obtain channel touchpoint activity distribution parameters;

[0060] User request data is extracted from app logs, official website access records, social platform interaction data, and telephone customer service records. The timestamp information and touchpoint identification field corresponding to each request are obtained. The timestamp data is converted to standard time nodes. The 24-hour day is then divided into fixed time periods. The request frequency of different touchpoint types within each time period is counted, and the frequency peaks are classified by touchpoint. By comparing the concentration of app requests in the morning and social platform requests in the evening, the daily time periods of each touchpoint are identified. At the same time, it is detected whether each touchpoint frequently appears in the same time period. If two types of touchpoints appear frequently in the same time period, it can be determined that the time period is an overlapping active period. For example, if users frequently submit questions through the app between 9:00 and 11:00 and frequently consult on social platforms between 20:00 and 22:00, it indicates that these two time periods are typical active periods. The time periods are aggregated and identified to form a dataset containing time period tags, frequency indicators, and overlap status. This is used for subsequent channel time activity analysis and to obtain channel touchpoint activity distribution parameters.

[0061] S112: Based on the channel touchpoint activity distribution parameters, determine the format structure of the request data collected from each channel, analyze the data node hierarchy and naming differences, compare the differences in text structure between the app, official website, and social platforms, and combine the consistency between the voice content and text requests to screen key fields and obtain the multi-channel structure matching coefficient;

[0062] According to the high-activity time period defined by the channel touchpoint activity distribution parameters, extract the original request data of the three channels of APP, official website, and social platform during this time period, and analyze the performance of the structural format in each data to determine whether the APP request uses a nested structure such as JSON, whether the official website uses URL form encoding, and whether the social platform uses a simple key-value pair text structure. Then compare the field hierarchy depth in each structure to check whether there are any format differences in the field names. For example, a field is named in lowercase and underlined in the APP structure, but in camel case in the social platform structure. Then convert and normalize the naming performance of the same field in different formats, such as changing "user_id" to "user_id". It is mapped to a unified field name with "UserId", and then the speech-to-text content in each structure is checked for consistency with the text structure data to check whether the semantic intent matches the structure field. For example, when the user's voice description is "I want to check yesterday's logistics", can the key structure fields such as "type=query", "object=logistics", and "time=yesterday" be correctly identified? The key fields that are consistent in field hierarchy, naming rules, and semantic consistency across all channels are screened out, and a field structure comparison table is established on this basis. By summarizing the matching density and consistency of fields in different channels, a multi-channel structure matching coefficient that can represent cross-channel data consistency is obtained.

[0063] S113: Based on the multi-channel structure matching coefficient, the correspondence between the device type parameters and the standard fields is optimized. The terminal operating system, device model, and browser type are analyzed, and the mapping path between the device parameters and the standard data is adjusted. The correspondence between the fields generated by each device is compared, and the overlap ratio of the multi-channel fields is calculated to obtain a unified request structure feature.

[0064] Perform standard field mapping on the device type parameters in user requests. First, extract device parameters such as operating system type, device model, and browser type from the request log, and then classify them into unified standard fields through the preset field mapping table. For example, classify "Android" as "Mobile System" and "Chrome" as "Desktop Browser". Then compare the standard fields with the predefined field set in the system to check whether there are any missing or duplicate fields, and analyze whether the number of fields and field structure are consistent between the APP and official website requests. Then, perform field content consistency comparison on the device parameters extracted from different sources such as the APP and official website to check whether the same standard field can be correctly extracted from different devices, and judge the structural uniformity of the fields generated by multiple devices. If the same field can be matched on multiple devices, the mapping is considered accurate. Finally, all accurately mapped field sets are constructed into a standard field template. At the same time, the fields that cannot be classified in different device structures are marked, and the standard field name, original field name, and source device are recorded in the field integration template as part of the unified request structure feature.

[0065] See also Figure 3 , the specific steps for obtaining the channel master sequence are:

[0066] S211: Based on the unified request structure characteristics, the channel format type and contact type identifier are analyzed, the correspondence between the user identity tag and each data number is compared, number duplication and mapping overlap are determined, the data attribution mapping is optimized, and the data attribution mapping strength is obtained;

[0067] Extract channel format type fields and contact type identification fields from various types of request data, and classify the channel format type fields according to the data structure performance, such as APP in JSON format, official website in form structure, and social platform in natural language text. Then extract the user request source tag from the contact type identification field, such as APP entrance, customer service entrance, social platform entrance, etc., and then establish a correspondence table between the channel and contact identification, analyze the many-to-one or one-to-one relationship between the two, and on this basis extract user identity tag fields, such as "user_id", "vip_level", "auth_token", etc., and extract the request number fields "request_id" and "trace_id", and build a correspondence table for the two types of fields. Count the number of numbers matched by each identity tag and the number duplication, mark the situation where multiple numbers correspond to one identity tag as overlapping items, and then determine whether the same number appears repeatedly under multiple identity tags. If the same number is repeated under two or more different Identity tag references are marked as number conflict items. For example, in the actual sample, "user_id=A123" corresponds to "request_id=REQ001, REQ002, REQ003", and "user_id=B456" also points to "REQ002", which means that "REQ002" has an attribution conflict. By counting the number of attributions and the number of conflicts, the number attribution confusion ratio is obtained. The attribution mapping rules are further adjusted to bind each number to the identity tag with which it first appears. Subsequent duplicate numbers are compared for identity tag consistency. If they are consistent, the mapping is retained. If they are inconsistent, the number is forcibly split or marked as an abnormal number. The updated number attribution relationship is then analyzed to calculate the proportion of numbers with clear attribution. If the attribution ratio is above 90%, the attribution mapping strength is considered to be high. If it is between 70% and 90%, it is considered to be medium strength. If it is below 70%, it is considered to be weak mapping. The attribution label, conflict record, and mapping strength score of each number are determined to form the data attribution mapping strength.

[0068] S212: Based on the data attribution mapping strength, analyze the session field and channel source field of each request, compare the position index in the sorting sequence, determine the consistency of the channel source field and the sorting, adjust the field combination, and obtain the sorting channel synergy coefficient;

[0069] Select the request data with the attribution mapping strength score higher than the set threshold, extract the session fields, such as "session_id", "conversation_token", etc., and the corresponding channel source fields, such as "source_channel", "entry_point", etc., count the number of unique identifiers corresponding to the session fields in each request, and bind them to the channel source, build a channel distribution list of the request data, and then sort all requests in ascending order by the submission time field, record the position index value of each request after sorting, and then analyze the relationship between the channel source fields of adjacent requests in the sorting sequence and the previous and next requests one by one. If the same channel source appears continuously in the sequence at a high ratio, it is judged that it is highly consistent with the sorting order. If there are frequent crossovers and source jumps, the consistency is low. Then combine the session field and the channel field to form a structured identification string, such as "session001_APP", "session001_Web", and observe whether the channel source appears multiple times in the same session. If the channel type switches multiple times within a session, the combination is unstable and needs to be marked as a channel jump item. By analyzing indicators such as the number of channel changes in each session, the channel differences between adjacent requests, and the number of consecutive requests from the same source, the field combination is adjusted, prioritizing highly consistent field pairs and merging duplicate fields or removing redundant fields. For example, "source_channel=Web" and "entry_point=WebLogin" can be merged into a single field "web_login_source". All combined fields are then re-positioned in the sorted list, and channel continuity and request aggregation before and after the adjustment are compared. Finally, by statistically analyzing the synergy data between the channel and the sorting order, the proportion of the continuous segment length occupied by each field combination in the sorted sequence is calculated. A ratio exceeding 80% is considered high synergy, a ratio between 50% and 80% is moderate synergy, and a ratio below 50% is weak synergy. The final result forms the sorting channel synergy coefficient, which serves as a reference for subsequent session management and request routing.

[0070] S213: Based on the sorting channel synergy coefficient, calculate the combined performance of the request time interval parameter and the channel sequence index to determine the impact of the emergency flag status on the sorting, using the formula:

[0071] ;

[0072] Get request prioritization margin , adjust the channel path sequence to obtain the channel master sequence, where Indicates the The time interval parameter of a request reflects the time distribution characteristics of the request in the sequence. Indicates the average of all request time interval parameters, used to measure the overall average level of all request time intervals. Indicates the The corresponding coordination level of the request in the sorting channel synergy coefficient is used to quantify the matching between the channel source field and the sorting order. Indicates the The urgent flag Boolean flag of the request is used to distinguish whether the request is urgent or not. Indicates the The corresponding number of the request in the channel sequence index, which is used to identify the specific order of the request in the channel processing path. Indicates the number of requests.

[0073] The request prioritization range refers to the value reflecting the comprehensive priority ranking performance of each request in the overall sorting optimization link, after jointly analyzing and calculating factors such as the time interval parameters, channel sequence, channel matching performance and urgency status of each request for all collected user requests in the multi-channel intelligent customer service system. It is used to quantify the priority strength of each user request in the comprehensive sorting under multi-channel paths, and provide a sorting reference for the channel master control sequence.

[0074] Get the requested time interval parameters , emergency signs , channel cooperation and channel order , and the dimensions are unified to eliminate the influence of different units. The time interval parameter after normalization is , the channel cooperation degree is , the emergency sign is , the channel sequence index is , number of requests , the time interval parameter mean , substitute various data and calculate item by item as follows:

[0075] Item 1: ;

[0076] Item 2: ;

[0077] Item 3: ;

[0078] Item 4: ;

[0079] The sum of the above four items is:

[0080] ;

[0081] This numerical result is the request prioritization amplitude , indicating that under the combination of multiple factors based on time interval difference, channel sorting index and emergency status, the overall sorting amplitude of the current request set is close to 1, belonging to the medium-high amplitude segment under the sorting master control condition. This result can be directly used as the sorting benchmark for the subsequent channel master control sequence and as a reference for the priority adjustment of each channel path. The formula comprehensively introduces participating factors under different dimensions, and through structured normalization and unified scale conversion, it controls the unequal influence between indicators while maintaining the sorting sensitivity, making the sorting result more adaptable to overall scheduling.

[0082] See also Figure 4 , the specific steps for obtaining the tag matching code are:

[0083] S311: Based on the channel master control sequence, the association between the subject-predicate structure field, the word order field, and the sentence structure field is compared to determine the combination of subject-predicate keywords and semantic orientations, identify keyword groups with key semantic centers, and obtain semantic structure keyword groups;

[0084] Extract the subject-predicate structure field from the user request text. This process is completed by identifying the dependency relationship between verbs and nouns. For example, when the user requests "I want to cancel the order", "I" is extracted as the subject, "cancel" as the predicate, and "order" as the object. Then extract the word order field, that is, determine whether the order of keyword appearance conforms to the word order norms of Chinese or the target language, such as "subject-predicate-object" or "subject-object-predicate", etc., and then extract the sentence structure field to check whether the sentence type is a declarative sentence, interrogative sentence, imperative sentence, etc., and combine punctuation and modal particles to assist in the judgment. Then, combine the collocation relationship between the subject and predicate with the semantic direction field for judgment. For example, "cancel + order" indicates a negative action, "query + logistics" indicates a neutral information action, and the identified combination is Whether a directed semantic orientation is formed. If the subject, predicate and object form a clear directional relationship, they are marked as valid semantic units. Further semantic action sets are extracted from multiple requests to construct keyword groups. For example, combinations such as "refund application", "order cancellation" and "express inquiry" are extracted from multiple user requests as semantic centers. By combining and judging the co-occurrence frequency of keywords, the length of the dependency path and the distance between words, combinations with high correlation, stable dependency and clear semantic direction are screened out. For example, the combination of "apply for refund" appears stably in more than 95% of the requests and the distance between the words is less than 3 word positions. It can be retained as a high-quality semantic structure keyword group, forming a group of semantic structure keyword groups based on the main control sequence extraction and in line with the grammatical dependency rules.

[0085] S312: Based on the semantic structure keyword phrase, optimize its matching with the knowledge tag index field, analyze the semantic distance and word meaning offset of the keyword phrase in the tag index, calculate the pairing strength distribution between the keyword and the tag, and obtain the semantic pairing strength distribution;

[0086] It is aligned and matched with the preset knowledge tag index field library, and the core field set used for index matching in the knowledge tag is extracted, such as label words such as "return", "payment failure", and "logistics delay". Then, each group of keyword combinations is compared with the label index field one by one to determine its semantic distance. By calculating the position difference of the word vectors between the keywords and label words in the keyword group in the semantic space, the similarity of the word meaning is measured. At the same time, the direction and degree of the word meaning offset between the keyword's original meaning and the label dictionary meaning are recorded. For example, if the offset angle between the keyword group "cancel order" and the label "return" is within 10° and the distance value is within the range of 0.3, it is determined to be a semantically close item. Then, the pairing relationship between all semantic structure keyword groups and multiple knowledge tags is constructed into a pairing matrix. The pairing strength between each keyword group and each label is compared line by line. The pairing strength is calculated based on a semantic distance less than 0.4 and a word meaning offset less than 0.2 as a high pairing benchmark. When the number of labels that meet the above conditions exceeds 3, the keyword group matching strength is classified as high strength. If the number of labels that meet the conditions is 2, it is classified as medium strength. If only 1 is matched, it is classified as low strength. By classifying and counting the matching distribution of all keyword groups, the proportion of the three types of segments with high pairing strength, medium pairing strength, and low pairing strength is obtained. For example, among 100 keyword groups, 45 fall in the high strength area, 30 are medium strength, and 25 are low strength. The corresponding semantic pairing strength distribution is the matching ability indicator of the semantic structure of this batch, which serves as the basis for the next step of label encoding optimization.

[0087] S313: Based on the distribution of semantic pairing strength, the degree of coordination between the subject-predicate structure and the label content is determined using the formula:

[0088] ;

[0089] Get the tag cooperative coupling factor , according to this factor, the candidate label items are arranged in order according to the subject-verb matching, and the label matching code is obtained, where, Indicates the The semantic matching coefficient between keywords and tag index items, Indicates the The semantic vector of keywords, Represents the corresponding label item semantic vector, Indicates the The coupling amount between the subject-verb structure and the semantic label, represents the sentence structure correction factor, represents the semantic category adjustment coefficient, Indicates the The amount of sentence fit between the class label item and the keyword group, is the total number of keyword and tag pairing groups, The number of categories for the label.

[0090] The tag collaborative coupling factor is a quantitative parameter used to measure the degree of semantic collaboration and coupling between a specific subject-predicate structure (the subject-predicate grammatical features of a user request) and candidate tag items (knowledge tags, service tags, etc.) in multi-touchpoint intelligent customer service. It can be used to sort all candidate tag items and select the tag item that best matches the semantics of the current user request.

[0091] Extract the structural fit relationship between each keyword group and the label item in the subject-predicate structure field, obtain the subject-predicate position information and sentence structure sequence of the keyword item, call the total semantic pairing strength of 0.086, and the structural coupling factor Together they form the numerator of the formula, and then set the sentence structure correction factor , this factor is used to represent the normalized weight of sentence errors caused by word order or subject-verb inversion, and further introduces the semantic category adjustment coefficient , which is used to adjust the sentence structure differences in cross-semantic label scenarios, build the denominator structure, and set the label category fitting interval in normalized form as 、 , represents the logical extension relationship between the two types of label items relative to the keyword word order and semantic content. Combining the above parameters, substituting them into the formula, the paired semantic items are:

[0092] The first group of keywords: "bill", corresponding matching coefficient , keyword semantic vector , label vector , normalized distance , the corresponding product is ;

[0093] The second group of keywords: "query", corresponding matching coefficient , keyword semantic vector , label vector , normalized distance , the corresponding product is ;

[0094] The third group of keywords: "I", corresponding matching coefficient , keyword semantic vector , label vector , normalized distance , the corresponding product is ;

[0095] Add up all the paired terms:

[0096] ;

[0097] Combine numerator terms:

[0098] ;

[0099] Combine the denominator terms:

[0100] ;

[0101] Substitute into the calculation:

[0102] ;

[0103] The results show that there is a stable coupling relationship between the current keyword group and the candidate tags in multiple dimensions such as semantic structure, subject-verb matching, and label fitting interval. It indicates that the degree of coordination between the subject-predicate structure and the label content is in a high adaptation range, indicating that the integrity of the semantic expression and the directionality of the grammatical structure of the label item can meet the unified label mapping requirements. The numerical result and the label matching code of the step S313 in this formula form a direct correspondence, that is, by calculating all candidate label items By sorting, the label number, label semantic index and content marking parameters corresponding to the maximum collaborative coupling factor can be extracted, thereby generating a unique coding structure and outputting it as a label matching code.

[0104] See also Figure 5 , the specific steps for obtaining the channel portrait matching factor are:

[0105] S411: Based on the tag matching code, the associated tag number and content tag parameter are analyzed, and the frequency of occurrence of each keyword in the user profile is calculated based on the usage scenario keywords appearing in the user profile, thereby generating a scenario keyword distribution feature.

[0106] Extract the tag number field from the structure. This field is used to identify the correspondence between the semantic tag and the user request. For example, the tag number "TAG002" indicates that it is related to refunds. Then extract the content tag parameters corresponding to each tag, such as "scenario type = after-sales processing", "keywords = return and refund", and group them according to the tag number. Then call the usage scenario keyword data stored in the user portrait. The field format is "shopping", "payment", "logistics", "after-sales", etc., and perform a keyword cross-comparison with each content tag parameter to determine whether there is a keyword consistency or inclusion relationship. Then perform frequency statistics on the number of times each type of keyword in the user portrait appears in all records. For example, the keyword "logistics" appears 12 times, "after-sales" appears 8 times, and "payment" appears 15 times in the user portrait, based on the total number of user portrait records of 100. , calculate the occurrence frequency of each keyword as 12%, 8%, and 15% respectively, construct a word frequency vector table with all keyword frequency data, and then sort them by frequency, mark the frequency range, for example, keywords with an occurrence frequency greater than 10% are high-frequency keywords, 5% to 10% are medium-frequency keywords, and less than 5% are low-frequency keywords. By screening the occurrence intensity of each keyword under the tag number, determine whether it appears cross-wise in multiple tags. If "payment" appears in both "TAG001" and "TAG003" tags at the same time, and the frequency is greater than 10%, it is marked as a key scenario keyword. All keywords that meet high frequency and multi-tag attribution are classified as the main distribution keyword group. Combined with the above sorting and tag attribution range, the scene keyword distribution feature is formed. This feature is used to express the frequency trend, attribution diversity and frequency concentration of keywords in different tags.

[0107] S412: Based on the distribution characteristics of the scene keywords, filter the access trajectory fields related to the channel identifier, compare the correlation between the trajectory fields and the record fields in the channel access sequence, calculate the intersection ratio of each field, and obtain the channel trajectory field interaction structure;

[0108] The main distribution keyword group is selected as the matching benchmark, and the access trajectory fields attached to the original request data are traversed to extract the page jump information, click nodes, and entry and exit fields in each user behavior path. For example, the fields are "enter_page=Home", "click_button=Buy Now", "exit_page=Payment Completed", etc., and then the channel identification fields are extracted, such as "channel=APP", "channel=Web", "channel=WeChat", and a trajectory field set with the channel identification as the main index is constructed. Then, the duplication between the access trajectory fields and the record fields in the channel access sequence is analyzed. For example, in the user trajectory field of the APP channel access record, "Homepage-Product Page-Shopping Cart-Payment Page" appears at the same time, while in the Web channel, only "Homepage-Shopping Cart-Payment Page" appears. Determine if the field "product page" exists only in the app trajectory. Calculate the number of intersections and unions of the two field sets and derive the intersection ratio. For example, if the app and web trajectory fields each have 5 items, including 3 common fields, the intersection ratio is 60%. An intersection ratio above 75% is considered high structural consistency, 50% to 75% is moderate structural consistency, and below 50% is weak structural consistency. Then, traverse all channel pairs one by one to construct a trajectory field intersection ratio matrix between multiple channels. Screen for field pairs with a repetition rate above a threshold and record whether the field pairs have a directional structure in the main channel, such as whether there is an "entry-jump-exit" time series combination. If there is structural continuity and the field content consistency is above 75%, mark it as an interactive structure field. Then, organize the field combination sets with characteristics such as channel directionality, field repetition, and structural sequence consistency, and define them as the channel trajectory field interaction structure.

[0109] S413: Based on the channel trajectory field interaction structure, compare the number of field categories, field expression structure characteristics and distribution density of user portrait parameters in each channel under the multi-scenario data structure, determine the joint distribution range of structurally consistent fields, and determine the key data items of structural overlap using the formula:

[0110] ;

[0111] Get channel profile matching factor ,in, Representative The intersection ratio of the trajectory fields of the channels, Representative The number of field categories contained in the channel, Representative The field structure characteristic encoding of each channel, Represents the total number of all channel field categories. Representative The distribution density of user portrait fields in each scenario, Representative The field distribution baseline density under each scenario, is the number of channels, is the number of service scenarios.

[0112] The channel portrait matching factor refers to a quantitative characteristic parameter that comprehensively measures the similarity between user portrait structures, field structure overlap, and data intersection characteristics under different channels in multi-channel intelligent customer service based on data such as label matching coding, scenario keyword distribution characteristics, and channel trajectory field interaction structure. It is used to characterize and determine the degree of fusion and matching of user portrait data structures in multiple dimensions such as scenario adaptation, field distribution, and structural characteristics between multiple service channels, providing a unified and standard structured basis for subsequent intelligent diversion, precise response, and authority control.

[0113] Extract the number of user profile field categories constructed in each channel in turn , field structure feature coding and the intersection ratio of trajectory fields , and normalized to the same dimension and set as:

[0114] For channel 1, let , , , the normalized result is , ;

[0115] For channel 2, let , , , the normalized result is , ;

[0116] For channel 3, , , , the normalized result is , ,

[0117] Substituting each term into the numerator and calculating item by item, we get:

[0118] ;

[0119] ;

[0120] ;

[0121] The numerator of the total is:

[0122] ;

[0123] At the same time, the total number of all channel fields is counted as , in the service scenario, the field usage density of scenario 1 is set to , whose base density is , the field usage density of scenario 2 is , whose base density is , then:

[0124] ;

[0125] ;

[0126] The denominator is calculated as:

[0127] ;

[0128] Substituting into the formula we get:

[0129] ;

[0130] This numerical result represents that under the background of the density difference between the current channel trajectory interaction structure and the multi-scene portrait structure, the channel portrait matching factor formed is 0.0336, indicating that the consistency between the current multi-channel portrait structures is in a low range based on the normalized reference. This factor can be used as a structural indicator for sorting and selection when judging the key fields of structural overlap in the future. The formula is based on the intersection ratio of the trajectory fields. and normalized structural parameters 、 The product operation enhances the expressiveness of structural coupling and combines the density deviation term A fitness adjustment term is formed to effectively reflect the matching distribution trend of structural fusion.

[0131] See also Figure 6 The specific steps for obtaining the desensitized permission output results are as follows:

[0132] S511: Based on the channel profile matching factor, identify the location of the permission level label in the content structure, compare the hierarchy, order and association relationship of each permission label in the content field, and obtain the permission label hierarchy sequence;

[0133] Extract the field information of the permission level label with the field name "access_level", "auth_scope", "permission_tag" in the content structure, locate the position of the field in the original content structure, and record the hierarchical structure and parent node path to which it belongs. For example, in a set of nested data, there is "access_level=VIP" under "user_info". This label is located in the second layer. Then extract the path structure of all permission-related fields from the entire content field structure. For each permission label, record its node level number, the order number of the field in the node, and the logical relationship between the previous and next fields, such as whether the permission field is adjacent to the identity field or whether it is nested in the user behavior field. Then group the same labels, such as "access_level=Admin", "access_level=Guest", "ac If "access_level=Internal" is found, the field positions of all tags in the group in different documents are compared. If a tag appears multiple times at different levels, it is recorded as a hierarchical drift item. Then, it is determined whether the tag has a fixed order relationship with a specific field, such as whether "access_level" always comes after "user_id". If it appears more than three times in different order positions, it is recorded as an inconsistent order item. Then, the nested or parallel relationship between each permission tag is counted. If it is found that the "auth_scope" field is always parallel to "access_level" and appears in the same structural layer, it is marked as a strong parallel association. All record items are organized into a permission tag hierarchical sequence, which contains dimensional information such as the tag name, the level number, the path index in the structure, the adjacent field number, whether it is parallel to other tags, and whether there is sequence drift. This structure serves as the basis for subsequent permission and identity comparison.

[0134] S512: calling the permission label hierarchy sequence, comparing it with the user identity authentication parameters, determining the association between the permission label and the identity identifier, authentication method, and authorization level in the content structure, and integrating the structure mapping to obtain the identity permission association structure;

[0135] Perform correlation analysis on each permission tag structure and user authentication parameters in turn. The authentication parameters are composed of fields such as "user_id", "auth_type", and "cert_level". First, extract the authentication parameters from the structure and compare their positions in the content structure with the structural paths of the permission tag locations to see if they coincide or are under the same parent node. If both the "user_id" field and the "access_level" field are nested under the "user_info" node, it is considered that the two have a structural adjacency relationship. Then determine whether there is hierarchical consistency in field values ​​between the permission tag and the authentication method. For example, whether "auth_type=real_name" and "cert_level=3" match "access_level=internal". If the authentication method is "real_name" and the permission level is "internal", confirm whether they correspond by setting the mapping table. If "real_name" corresponds to the lowest level, "internal" permission is matched. If the permission field level is higher than the current authentication parameter authorization level, the match fails. Combining the above-mentioned field path adjacency, field value level matching, label-to-field coverage and other information, a structural mapping relationship record table is constructed between each authentication field and the permission label to further determine whether there is cross-node field reference behavior. For example, the "auth_scope" field appears in the "account_section" node, and the "cert_level" field appears in the "user_info" node. It is necessary to determine whether the structural association is achieved through logical reference. If there is no reference path, the permission field is regarded as an isolated label and cannot participate in the match. Finally, all combinations of permission fields and authentication fields with path connection, field level matching, and field position alignment are marked as valid structural mapping combinations. All valid combinations are organized into an identity permission association structure, which is used to identify which identity parameters are effectively bound to which permission fields, and the binding method is path binding, level matching, or equal field values.

[0136] S513: Based on the identity and permission association structure, the involved content fields are screened, the field access paths and sensitive field identifiers are analyzed, the paths and identifiers of fields with permission labels higher than the authentication level are adjusted, and the node data is optimized and processed to obtain a desensitized permission output result;

[0137] First, extract the content fields corresponding to the permission field combinations marked as valid matches in the structure, filter all content items directly or indirectly bound to the permission fields into a candidate set, and then perform access path analysis on the fields in the candidate set, tracking the jump nodes, control parameters, and authorization control bits in the field access path. For example, if a content field "internal_comment" can only be accessed through "access_level=Admin", then its access path must include a permission judgment node, record the node structure at each level in the path and judge the integrity of the path, then extract whether the field is marked as a sensitive field. Sensitive fields are defined by labels such as "sensitive=true" and "confidential_flag=yes", and then combined with the user's current identity level parameters, judge whether the field permission label level is higher than the current identity level, for example, "internal_comment" The "nal_comment" is identified as "access_level=Admin", while the current authentication is only "auth_type=login" and "cert_level=1". The authentication level is lower than the access permission level. The path of the field needs to be adjusted and its original access path is marked as disabled. At the same time, the desensitized field content is added instead, such as replacing the original field content with "***". Then, the processing node data of the field in the data processing link is recorded, including field status change records, path replacement operations, field visibility marks, etc., and all affected nodes are updated. For example, the "response_data.comment" field is marked as "masked". Finally, the processing path, original permission label, user authentication parameters, and adjusted marking status corresponding to each desensitized field are output to construct the desensitized permission output result.

[0138] The above are merely preferred embodiments of the present invention and do not limit the present invention in any other form. Any technician familiar with the profession may use the technical content disclosed above to change or modify it into an equivalent embodiment with equivalent changes and apply it to other fields. However, any simple modification, equivalent change and modification made to the above embodiment based on the technical essence of the present invention without departing from the content of the technical solution of the present invention shall still fall within the scope of protection of the technical solution of the present invention.

Claims

1. A multi-touchpoint method for intelligent customer service in a digital system based on artificial intelligence, characterized in that: The following steps are involved: S1: Based on multi-channel user request entry points, analyze the operation time and touchpoint ID distribution, determine the differences in text structure of each entry point, perform consistency detection on voice content, compare device type mapping performance, identify data fields that can be standardized, and obtain unified request structure characteristics; S2: Based on the unified request structure characteristics, analyze the mapping between identity tags and data numbers, determine the matching relationship between session fields and channel sources, compare the request time interval distribution, adjust the channel path processing order, prioritize requests according to their criticality, and obtain the channel master control sequence; The steps for obtaining the channel master control sequence are specifically as follows: S211: Based on the unified request structure characteristics, the channel format type and the contact type identifier are analyzed, the correspondence between the user identity tag and each data number is compared, number duplication and mapping overlap are determined, the data attribution mapping is optimized, and the data attribution mapping strength is obtained; S212: Based on the data attribution mapping strength, analyze the session field and channel source field of each request, compare the position index in the sorting sequence, determine the consistency of the channel source field and the sorting, adjust the field combination, and obtain the sorting channel synergy coefficient; S213: Based on the sorting channel synergy coefficient, calculate the combined performance of the request time interval parameter and the channel sequence index, determine the impact of the emergency flag state on the sorting, obtain the request priority sorting range, adjust the channel path sequence, and obtain the channel master control sequence; S3: Based on the channel master sequence, filter the keywords in the content field, analyze the semantic structure of the keywords and knowledge tags, determine the synergy between the subject-predicate structure and the tag content through semantic pairing of the keywords and tag indexes, locate the tag item with the best semantic association, and obtain the tag matching code; S4: Analyze the distribution of usage scenario keywords in the user profile based on the tag matching code, screen the correlation between the channel identifier and the access trajectory, compare the adaptability of the channel profile parameters, determine the performance of the request behavior in each channel, and obtain the channel profile matching factor; The specific steps for obtaining the channel portrait matching factor are as follows: S411: Based on the tag matching code, the associated tag number and content tag parameter are analyzed, and the frequency of occurrence of each keyword in the user portrait is calculated in combination with the usage scenario keywords appearing in the user portrait, thereby generating a scenario keyword distribution feature; S412: Based on the scene keyword distribution characteristics, filter the access trajectory fields related to the channel identifier, compare the correlation between the trajectory fields and the record fields in the channel access sequence, calculate the intersection ratio of each field, and obtain the channel trajectory field interaction structure; S413: Based on the channel trajectory field interaction structure, compare the number of field categories, field expression structure characteristics and distribution density of user portrait parameters in each channel under the multi-scenario data structure, judge the joint distribution range of structurally consistent fields, determine the key data items of structural overlap, and obtain the channel portrait matching factor.

2. The multi-touchpoint method for intelligent customer service in a digital system based on artificial intelligence according to claim 1, characterized in that: The unified request structure features include contact type identification, data integration template, and channel format type; the channel master control sequence includes request priority coding, session diversion tag, and channel sequence index; the label matching code includes label number, semantic label index, and content label parameter; the channel portrait matching factor includes user feature code, scenario adaptation tag, and behavior-oriented feature.

3. The multi-touchpoint method for intelligent customer service in a digital system based on artificial intelligence according to claim 1, characterized in that: The steps for obtaining the unified request structure feature are specifically as follows: S111: Based on multi-channel user request entry points, analyze time records and touchpoint identifiers, aggregate user request times and categories across channels, compare the time concentration of user requests and touchpoint overlap within each channel, determine changes in active segments across channels, and obtain channel touchpoint activity distribution parameters; S112: Based on the channel touchpoint activity distribution parameters, determine the format structure of the request data collected from each channel, analyze the data node hierarchy and naming differences, compare the differences in text structure between the app, official website, and social platforms, and screen key fields based on the consistency between the voice content and the text request to obtain a multi-channel structure matching coefficient; S113: Based on the multi-channel structure matching coefficient, optimize the correspondence between the device type parameters and the standard fields, analyze the terminal operating system, device model and browser type, adjust the mapping path between the device parameters and the standard data, compare the correspondence of the fields generated by each device, and calculate the overlap ratio of the multi-channel fields to obtain a unified request structure feature.

4. The multi-touchpoint method for intelligent customer service in a digital system based on artificial intelligence according to claim 1, characterized in that: The steps for obtaining the tag matching code are specifically as follows: S311: Based on the channel master control sequence, compare the associations of the subject-predicate structure field, the word order field, and the sentence structure field, determine the combination of subject-predicate keywords and semantic orientations, identify keyword groups with key semantic centers, and obtain semantic structure keyword groups; S312: Based on the semantic structure keyword group, optimize its matching with the knowledge tag index field, analyze the semantic distance and word meaning offset of the keyword group in the tag index, calculate the pairing strength distribution between the keyword and the tag, and obtain the semantic pairing strength distribution; S313: Based on the semantic pairing strength distribution, determine the degree of coordination between the subject-predicate structure and the label content, obtain a label coordination coupling factor, and arrange the candidate label items in order according to the subject-predicate matching according to the factor to obtain a label matching code.

5. The multi-touchpoint method for intelligent customer service in a digital system based on artificial intelligence according to claim 1, characterized in that: The steps also include: S5: Based on the channel profile matching factor, identify the permission level label in the content structure, analyze the association between user identity authentication parameters and content permissions, compare the permission level differences, identify the content that needs to be desensitized, adjust the processing node data, and obtain the desensitized permission output result; The desensitized permission output result includes content desensitization label, permission discrimination parameters, and data feedback instructions.

6. The multi-touchpoint method for intelligent customer service in a digital system based on artificial intelligence according to claim 5, characterized in that: The steps for obtaining the desensitized permission output result are specifically as follows: S511: Based on the channel profile matching factor, identify the location of the permission level label in the content structure, compare the hierarchy, order and association relationship of each permission label in the content field, and obtain the permission label hierarchy sequence; S512: calling the permission label hierarchical sequence, comparing it with the user identity authentication parameters, determining the association between the permission label and the identity identifier, authentication method and authorization level in the content structure, and integrating the structure mapping to obtain the identity permission association structure; S513: According to the identity authority association structure, the content fields involved are screened, the field access path and sensitive field identifiers are analyzed, the paths and identifiers of the fields whose authority labels are higher than the authentication level are adjusted, the node data is optimized and processed, and a desensitized authority output result is obtained.

Images