An automatic interactive answering method and system of an after-sales consultation service platform

Abstract

The application relates to the technical field of after-sales intelligent service, in particular to an automatic interactive response method and system of an after-sales consultation service platform. The method collects initial interactive input information of the current round, historical business data corresponding to the type of industrial products, and pre-processes the input information to obtain an effective word sequence; the initial professional degree is determined according to the effective word sequence, the familiarity of each product is obtained based on the historical business data, and the initial level benchmark of the user is determined in combination with the historical level value; after the initial level benchmark is determined, a diagnosis instruction is issued to the user, the benchmark is corrected according to the user feedback to obtain a real level value; the response content is generated and integrated based on the real level to adaptively match the speech professional density and the number of auxiliary materials, and is uniformly output; after the current round ends, the interactive efficiency is calculated by counting the interactive step information, and the historical level value of the user is updated in combination with the real level, so that the user experience is improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent after-sales service technology, specifically to an automatic interactive response method and system for an after-sales consultation service platform. Background Technology

[0002] The Internet has driven the transformation of after-sales consultation for industrial products from manual service to intelligent and automated interaction. After-sales service for industrial equipment often requires multiple rounds of dialogue to guide users to complete professional operations such as troubleshooting and equipment calibration, which requires high interactivity.

[0003] Existing intelligent response solutions often employ technologies such as keyword matching, deep learning, and large language models to continuously improve semantic understanding and multi-turn dialogue capabilities. However, they generally adopt a uniform response strategy without differentiating users' professional abilities. This approach has significant drawbacks: on the one hand, for users with rich operating experience and solid professional knowledge, the overly basic or lengthy guidance provided by the system reduces communication efficiency and may even cause user dissatisfaction; on the other hand, for novice users lacking relevant experience, the system outputs too much technical jargon and insufficient supplementary materials, which can easily lead to user misunderstanding, operational errors, and thus increase the frequency of human intervention, affecting the user experience. Summary of the Invention

[0004] To address the technical problem that existing industrial after-sales automated interactive response systems employ a uniform guidance strategy, which fails to adapt to users with varying levels of operational experience, resulting in low interaction efficiency and a poor user experience, this invention provides an automated interactive response method and system for an after-sales consultation service platform. The specific technical solution adopted is as follows: This invention proposes an automatic interactive response method for an after-sales consultation service platform, the method comprising: Obtain the user's input information in the preset initial interaction round of the current consultation service and historical business data related to the industrial product type involved in the input information, and preprocess the input information to obtain an effective word sequence; The user's initial level of expertise is determined based on the effective word sequence, and the user's familiarity with each type of industrial product is determined based on historical business data. The user's initial level benchmark for each type of industrial product is determined based on the initial level of expertise, familiarity with each product, and the user's existing historical level values. After the initial level benchmark is determined, a diagnostic command is issued to the user; the user's response information to the diagnostic command is obtained, and the initial level benchmark is corrected based on the response information to obtain the user's true level value; Based on the actual performance level, determine the professional density and amount of supporting materials for the after-sales scripts for each type of industrial product in this round, generate corresponding response scripts and match supporting materials, integrate them and output them to users in a unified manner to complete this round of consultation service; Once completed, the interaction efficiency is calculated based on the number of interaction steps obtained from this consultation, and the user's historical level value is iteratively updated in combination with the actual level value.

[0005] Furthermore, the step of obtaining the user's input information in the current round of consultation service at the preset initial interaction round and historical business data related to the industrial product type involved in the input information, and preprocessing the input information to obtain a valid word sequence, includes: Obtain the input text entered by the user in the preset initial interaction rounds of this consultation service as input information, and simultaneously obtain the user account and the device tag provided by the user; Based on the user account, the system retrieves the user's purchase list and maintenance records within a preset historical period from the supply chain database as historical business data; it also performs text cleaning, Chinese word segmentation, and part-of-speech tagging on the input text to obtain a valid word sequence. Identify the type of industrial product corresponding to this round of consultation based on the equipment label and / or the entered text; For various industrial product types, the system uses industrial product type as the filtering condition to filter out purchase records of the same type from the purchase list; and uses user account and industrial product type as the combined filtering condition to filter out historical repair or consultation records of industrial product types from the maintenance records. The effective word sequence, the filtered purchase records, and the maintenance records are stored together using the industrial product type as the association key.

[0006] Furthermore, the initial professionalism determination process includes: Obtain a pre-built professional thesaurus, which includes at least a root word field, a synonym phrase field, and a part-of-speech tag field. The root word field is used to store benchmark technical terms; the synonym phrase field is associated with the root word field and is used to store at least one synonymous expression of the benchmark technical terms; the part-of-speech tag field is used to mark the part-of-speech category of the benchmark technical terms. Each word in the effective word sequence is matched sequentially with word roots and synonyms; words that match word roots or synonyms are recorded as professional words, and the number of professional words in the effective word sequence is counted as the number of professional words. Repeated words in the effective word sequence are deduplicated; the total number of words in the effective word sequence after deduplication is counted as the effective vocabulary size; If the effective vocabulary is zero, the user's initial professional level is set to zero; otherwise, the proportion of professional vocabulary to the effective vocabulary is calculated and used as the user's initial professional level.

[0007] Furthermore, the process for determining the user's familiarity with each type of industrial product includes: For various types of industrial products, obtain the filtered purchase records and maintenance records; calculate the purchase quantity of industrial product types involved in the filtered purchase records, and the number of times users reported or inquired about industrial product types in the filtered maintenance records. If the purchase quantity and the number of repair requests or inquiries are both zero, the user's familiarity with the type of industrial product will be set to zero. Otherwise, a weighted sum is calculated by using preset weighting coefficients to calculate the weighted sum of the purchase quantity and the number of repair requests or inquiries; where the weighting coefficient corresponding to the number of repair requests or inquiries is greater than the weighting coefficient corresponding to the purchase quantity. The weighted sum is normalized to obtain the user's familiarity with the type of industrial product.

[0008] Furthermore, determining the user's initial skill level benchmark for each industrial product type based on initial expertise, familiarity with various aspects, and the user's existing historical skill level includes: Retrieve the user's existing historical level values; If the user's consultation in this round only involves one type of industrial product, the initial level of expertise is multiplied by the user's familiarity with that type of industrial product to obtain the first reference value; When a user is involved in multiple industrial product types during this round of consultation, the arithmetic mean of the user's familiarity with all industrial product types is calculated as the average familiarity level; the initial level of expertise is multiplied by the average familiarity level to obtain the first reference value. If the user already has historical level values, the minimum value between the first reference value and the historical level values ​​will be used as the user's initial level benchmark for each industrial product type. If the user is consulting for the first time, or if the user's historical level values ​​are not available, the first reference value will be used directly as the user's initial level benchmark for each industrial product type.

[0009] Furthermore, the response information includes the time information and / or detection data information of the user's feedback operation results; The process of determining the user's true skill level includes: Obtain the first timestamp of the issued diagnostic command, and the second timestamp of the user's feedback operation result extracted based on the time information; obtain the preset standard consumption time and the preset tolerance time; calculate the sum of the preset standard consumption time and the preset tolerance time as the allowed consumption time; Calculate the difference between the second timestamp and the first timestamp as the actual time consumed by the user's operation; calculate the difference between the actual time consumed and the allowed time consumed as the timeout amount; if the timeout amount is not greater than zero, set the delay penalty value to zero; if the timeout amount is greater than zero, normalize the timeout amount to obtain the normalized timeout amount; multiply the normalized timeout amount by the preset penalty coefficient and take the negative value as the delay penalty value. The system calculates the number of data dimensions requested by the diagnostic command and the number of valid data dimensions that meet the preset validity conditions in the test data returned by the user; it then calculates the ratio of the number of valid data dimensions to the total number of data dimensions as the accuracy of the user's response. If the current round only has a delay penalty value, the delay penalty value is used as the observation level value for this round; if the current round only has feedback accuracy, the feedback accuracy is used as the observation level value for this round; if the current round has both a delay penalty value and feedback accuracy, the delay penalty value and feedback accuracy are added together to obtain the observation level value for this round. Calculate the deviation between the observed level value in this round and the corrected true level value in the previous round; multiply the deviation value by the preset step size to obtain the adjustment amount in this round; add the corrected true level value in the previous round to the adjustment amount in this round to obtain the corrected true level value in this round; wherein, in the first round of correction, the initial level benchmark is used as the corrected true level value in the previous round.

[0010] Furthermore, the process for determining the professional density and the amount of supporting data includes: For the various types of industrial products involved in this round of consultation, obtain the highest pre-defined professional density for each type of industrial product. Normalize the actual level value; multiply the normalized actual level value by the preset maximum professional density to obtain the professional density of the response script for this round. Calculate the difference between the value 1 and the normalized true level value as the true difference; multiply the true difference by the preset maximum extra amount, and round down the product to obtain the rounded result. Add the rounded-down result to the preset base quantity to obtain the number of supporting materials for this round of responses.

[0011] Furthermore, the process of generating corresponding response scripts, matching supplementary materials, integrating them, and then uniformly outputting them to the user includes: Obtain a pre-set auxiliary data library, which contains multiple auxiliary data for each type of industrial product. Each auxiliary data has a different complexity level and is sorted from low to high complexity level. The auxiliary data includes at least one of the following: text descriptions, images, 3D structural diagrams, and videos. Access the pre-set script template library, which contains multiple script templates for various industrial product types, each with different levels of professional density. For the various types of industrial products involved in this round of consultation, multiple script templates corresponding to each type of industrial product are obtained from the script template library; based on the number of supporting materials in this round of responses, the top number of supporting materials are selected from the sorted supporting materials in order of complexity from low to high. Based on the professional density of the response scripts in this round, the script template with the closest professional density level is selected from the multiple script templates obtained and used as the response script for this round. In this round of consultation involving multiple industrial product types, corresponding response scripts and supporting materials were generated for each industrial product type; the response scripts and supporting materials for each industrial product type were summarized to generate integrated response content; and the integrated response content was uniformly delivered to the user in the same round of responses.

[0012] Furthermore, the step of calculating the interaction efficiency based on the interaction step count information obtained from this consultation, and iteratively updating the user's historical level value in combination with the actual level value, includes: Obtain the user's historical level values ​​saved in the cloud; extract the total number of rounds of interaction with the user from the interaction step information of this consultation, as the actual number of steps; extract the ideal number of steps preset for each type of industrial product; and extract the number of steps in which the user repeatedly asked the same question during the interaction, as the repetition number. Calculate the ratio of the preset ideal number of steps to the actual number of steps as the step efficiency factor; calculate the ratio of repeated steps to the actual number of steps as the repetition factor; and calculate the difference between the value 1 and the repetition factor as the non-repetition question factor. The interaction efficiency is obtained by multiplying the step efficiency factor by the non-repeated question factor, and then by the preset transfer mark coefficient. If the consultation process involves a human agent, the transfer mark coefficient is recorded as 0; otherwise, it is recorded as 1. If this is the user's first consultation, the initial level benchmark of this consultation is used as the historical level value; otherwise, the difference between the final determined real level value at the end of this consultation and the historical level value is calculated as the level change amount; the level change amount, interaction efficiency and preset growth factor are multiplied together to obtain the effective progress amount; the historical level value is added to the effective progress amount to obtain the preliminary upgrade value. The minimum value between the actual level value and the initial upgrade value is taken as the level value after this upgrade; the upgraded level value is saved to the cloud database as the historical level value for subsequent consulting services.

[0013] An automatic interactive response system for an after-sales consultation service platform is disclosed. The system includes a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of an automatic interactive response method for an after-sales consultation service platform.

[0014] The present invention has the following beneficial effects: This invention integrates user input information from initial interactions with historical business data related to industrial products. It comprehensively calculates initial professionalism, product familiarity, and historical performance values ​​to determine the user's initial skill level benchmark from multiple dimensions, significantly improving the accuracy of initial assessments of user professional competence. By issuing diagnostic commands to obtain user response information, the initial skill level benchmark is dynamically adjusted to obtain a true skill level value that matches the user's actual situation. Based on the true skill level, the professional density of the communication script and the amount of supporting materials are adaptively matched to the user, pushing integrated response content adapted to their cognitive abilities. This caters to the reading needs of both novice and professional users, reducing ineffective communication and improving user experience. After each round of consultation, the interaction efficiency is calculated based on the number of interaction steps, and the historical skill level value in the cloud is iteratively updated in conjunction with the user's true skill level. This enables continuous dynamic adaptation of subsequent consultation services, improving the overall interaction efficiency and user consultation experience of industrial product after-sales consultation in the long term. Attached Figure Description

[0015] Figure 1 A flowchart illustrating an automatic interactive response method for an after-sales consultation service platform provided in one embodiment of the present invention; Figure 2 This is an example diagram illustrating the process of determining a user's true skill level as provided in one embodiment of the present invention. Detailed Implementation

[0016] The following description, in conjunction with the accompanying drawings, details the specific solution of the automatic interactive response method and system for an after-sales consultation service platform provided by this invention.

[0017] Please see Figure 1 The diagram illustrates an automatic interactive response method for an after-sales consultation service platform according to an embodiment of the present invention. The method includes: S101: Obtain the user's input information in the preset initial interaction round of the current consultation service and the historical business data related to the industrial product type involved in the input information, and preprocess the input information to obtain an effective word sequence.

[0018] The preset initial interaction rounds refer to the number of interaction rounds that the system pre-sets for collecting initial user information. Preferably, the preset initial interaction rounds can be set to 1 to 3 rounds.

[0019] It should be noted that preprocessing the input text to obtain a valid word sequence is a common existing technique, which will not be elaborated here. Specifically, the preprocessing includes text cleaning, Chinese word segmentation, and part-of-speech tagging, all of which are implemented using existing natural language processing techniques. After the above processing, a valid word sequence composed of words with actual semantic meaning is obtained.

[0020] Industrial product types are classifications of industrial equipment or consumables based on their function, purpose, or structure, such as "incinerator," "chemical protective suit," and "compressor."

[0021] Equipment is a specific instance of an industrial product type. An industrial product type can contain multiple specific models of equipment. For example, "Incinerator-Type X" and "Incinerator-Type Y" both belong to the industrial product type "Incinerator". Generally speaking, equipment within the same industrial product type has similar failure modes and operating procedures.

[0022] In this embodiment, the input text entered by the user in the current consultation service within a preset initial interaction round is obtained as input information, and the user account and the device tag provided by the user are obtained simultaneously. Based on the user account, the user's purchase list and maintenance records within a preset historical period are obtained from the supply chain database as historical business data. The input text is cleaned, Chinese word segmentation and part-of-speech tagging are performed to obtain a valid word sequence. Based on the device tag and / or the input text, the industrial product type corresponding to the current consultation is identified. For each industrial product type, the purchase record of the same type as the industrial product type is filtered from the purchase list. The user's historical repair or consultation records for the industrial product type are filtered from the maintenance records using the user account and industrial product type as joint filtering conditions. The valid word sequence, the filtered purchase record and maintenance record are stored together with the industrial product type as the association key.

[0023] A user account is a unique identifier for a user in the system. It can be a mobile phone number, employee ID, or enterprise unified authentication ID, and is used to associate the user's historical business data and the information of the company to which they belong.

[0024] A device tag is a unique identifier for a target device; it can be a device serial number, nameplate number, or QR code label. Specifically, users can provide device tags through manual input, scanning a QR code, or near-field communication (NFC) reading.

[0025] It should be noted that supply chain databases are usually the infrastructure for enterprise information technology construction, and their data query and export functions are conventional technical means. This invention can directly utilize existing data interfaces to achieve this.

[0026] It should be noted that the preset historical period is used to limit the scope of historical business data retrieval. Its specific value can be set according to the type of device the user is interested in and actual needs. This embodiment does not impose a specific limitation. For example, the preset historical period can be set to one year.

[0027] In addition, it should be noted that users can initiate inquiries in various ways, including but not limited to directly entering text, voice input, and uploading images containing text descriptions.

[0028] Specifically, users can initiate inquiries via terminal devices (such as smartphones, tablets, industrial handheld terminals, or personal computers) using either text or voice input. When a user uses voice input, the system can call the existing speech recognition module to convert the speech into text, and then use the converted text as the input text.

[0029] The procurement list records information on various industrial products purchased by the user's company within a preset historical period, including at least fields such as industrial product name, model, quantity purchased, and purchase date.

[0030] The maintenance record records the equipment repair or technical consultation information initiated by the user's company within a preset historical period, including at least the repairer's account, equipment name, equipment model, repair date, and fault description.

[0031] S102: Determine the user's initial level of expertise based on the effective word sequence, and determine the user's familiarity with each type of industrial product based on historical business data; determine the user's initial level benchmark for each type of industrial product based on the initial level of expertise, each level of familiarity, and the user's existing historical level values.

[0032] It's important to understand that users' communication skills vary significantly in industrial equipment after-sales consultations. Experienced engineers typically use precise industry terminology to describe malfunctions (e.g., "the neoprene coating is cracking in a mesh pattern"), while novice users often only provide vague, colloquial descriptions (e.g., "This thing is broken"). Therefore, the density of technical terms used in user input directly reflects their mastery of the technical language in that field, and this mastery is positively correlated with the user's operational experience. Thus, analyzing the density of technical terms in the initial user input text allows for a quick, rough assessment of the user's theoretical knowledge level during the initial interaction phase.

[0033] In a preferred implementation, a pre-built professional thesaurus is obtained. This thesaurus includes at least a root word field, a synonym phrase field, and a part-of-speech tag field. The root word field stores baseline technical terms. The synonym phrase field, associated with the root word field, stores at least one synonymous expression of the baseline technical term. The part-of-speech tag field marks the part-of-speech category of the baseline technical term. Each word in the effective word sequence is sequentially matched with the root word and the synonym phrase. Words matching the root word or synonym phrase are recorded as professional terms, and the number of professional terms marked in the effective word sequence is counted as the number of professional terms. Duplicate words in the effective word sequence are deduplicated. The total number of words in the deduplicated effective word sequence is counted as the effective vocabulary size. If the effective vocabulary size is zero, the user's initial professional level is directly set to zero; otherwise, the proportion of the number of professional terms to the effective vocabulary size is calculated as the user's initial professional level.

[0034] It should be noted that the pre-built professional thesaurus can be implemented using conventional technologies in this field, without requiring any improvement to the thesaurus construction method itself. Specifically, domain experts can compile and input core terms and their synonyms for various industrial product types and common fault scenarios, forming a structured terminology mapping table; alternatively, existing industry thesaurus or equipment operation manuals can be imported and supplemented.

[0035] It is understandable that the root word field stores the benchmark technical term (such as "sealing failure"); the synonym phrase field stores the synonyms of the benchmark technical term (such as "air leakage" or "poor sealing"); and the part-of-speech tag field marks the part-of-speech category of the benchmark technical term (such as noun, verb, or adjective). These three are all conventional dictionary data structure designs and are common knowledge in this field.

[0036] For example, for the root word "seal failure", its synonyms can include synonyms such as "air leakage", "poor sealing", and "leakage", and the part-of-speech tag field can be marked as "fault-phenomenon" category; for the root word "bearing", its synonyms can include "bearing" and "bearing shell", and the part-of-speech tag field can be marked as "component-part".

[0037] It should be noted that the process of matching each word in the effective word sequence with the root word and synonyms in turn can be achieved using conventional string matching or synonym mapping techniques, which can be applied directly here.

[0038] It should be noted that deduplication can be implemented using a conventional set deduplication algorithm, which can be applied directly here. Specifically, each word in the valid word sequence is added to the set one by one, duplicate words are automatically merged, and finally, the total number of words in the set is counted to obtain the deduplicated valid word sequence.

[0039] Since the effective vocabulary is zero, it can be understood that the user has not provided any effective semantic information in the preset initial interaction rounds. Therefore, the system can directly set the initial professionalism to zero to avoid subsequent calculation errors.

[0040] For example, if the effective word sequence after deduplication is ["device", "broken", "how", "repair"], the effective vocabulary is 4. Among them, the professional vocabulary database matches "device" as a professional term, while "broken", "how", and "repair" are not matched, so the number of professional terms is 1. The initial professional score is 1÷4=0.25, indicating that the user's theoretical knowledge level is low.

[0041] Initial professionalism is used to quantify a user's level of theoretical knowledge in the initial consultation phase (i.e., the preset initial interaction round). A higher initial professionalism indicates a higher proportion of technical terms in the user's initial input text, reflecting that the user has good professional expression skills and theoretical understanding; a lower initial professionalism indicates that the user uses more colloquial or non-professional expressions, reflecting that their theoretical knowledge level is relatively limited.

[0042] It's important to understand that in industrial after-sales scenarios, users' operational experience with different types of industrial products is often uneven. For example, a user who is very familiar with incinerators, which are their daily work objects, might have little experience with chemical protective suits. In this case, assessing a user's overall operational ability solely based on initial expertise would incorrectly generalize the user's proficiency in one type of equipment to all equipment types, leading to oversimplification or overcomplication in troubleshooting new equipment types. Therefore, the system can utilize two types of objective data: the user's company's historical procurement records (reflecting which equipment the company owns) and the user's personal maintenance records (reflecting which equipment malfunctions the user has personally handled). These data can be used to calculate the user's operational familiarity with different types of industrial products, providing a precise basis for determining the initial skill benchmark at the equipment type level.

[0043] As a preferred implementation, for various industrial product types, filtered purchase records and maintenance records are obtained; the purchase quantities of industrial product types involved in the filtered purchase records and the number of times users reported or inquired about industrial product types in the filtered maintenance records are statistically analyzed; if both the purchase quantity and the number of reports or inquiries are zero, the user's familiarity with the industrial product type is directly set to zero; otherwise, a weighted sum is obtained by using a preset weighting coefficient to calculate the weighted sum of the purchase quantity and the number of reports or inquiries; wherein, the weighting coefficient corresponding to the number of reports or inquiries is greater than the weighting coefficient corresponding to the purchase quantity; the weighted sum is then normalized to obtain the user's familiarity with the industrial product type.

[0044] It should be noted that the operations of counting the number of purchases, repair requests, or inquiries are essentially routine data filtering and counting processes, which belong to basic database query functions, and will not be elaborated on here.

[0045] Purchase quantity refers to the total number of industrial products of the same type as those in this round of consultation purchased by a user's company within a preset historical period.

[0046] The number of repair requests or inquiries refers to the total number of times a user account initiates repair requests or inquiries for this type of industrial product within a preset historical period.

[0047] It should be noted that the preset weighting coefficients can be determined based on the type of industrial product or the actual needs of the scenario; this embodiment does not impose specific limitations. For example, the weighting coefficient for the purchase quantity can be set to 0.4, and the weighting coefficient for the number of repair requests or inquiries can be set to 0.6.

[0048] Because, generally speaking, for the same type of industrial product, there is no strict correlation between the quantity purchased by a company and the actual operating experience of users. For example, a large purchase quantity may mean that the company owns a lot of this type of equipment, but the user may not have operated it personally. However, repair requests or inquiries directly record the operation requests initiated by the user, which better reflects their actual operating experience. Therefore, the weighting coefficient corresponding to the number of repair requests or inquiries is set to be higher than the weighting coefficient corresponding to the purchase quantity.

[0049] It is understandable that if a user's purchase quantity and number of repair requests or inquiries are both zero, it means that the user's company has never purchased this type of equipment, and the user has never requested repairs or inquired about this type of equipment. In other words, the user has neither company-level contact with this type of equipment nor personal-level operational experience. In this case, the system directly sets the user's familiarity with this type of industrial product to zero.

[0050] A specific example of normalizing a weighted sum is as follows: Assume the weighted sum is represented by x, and its normalized value y can be calculated using a sigmoid function, i.e. , where e represents the natural constant; this sigmoid function maps x to the range of 0 to 1, the larger x is, the closer y is to 1, the smaller x is, the closer y is to 0, and when x=0, y=0.5.

[0051] Familiarity is used to quantify a user's level of operational experience with a certain type of industrial product. A higher level of familiarity indicates that the user has more experience purchasing and / or operating that type of equipment within a preset historical period, reflecting a higher level of familiarity with troubleshooting, operating procedures, and maintenance points for that type of equipment. Conversely, a lower level of familiarity indicates that the user has limited opportunities and practical experience with that type of equipment, reflecting relatively insufficient operational experience.

[0052] In this embodiment, the user's existing historical skill level is obtained. If the user's current consultation involves only one industrial product type, the initial expertise level is multiplied by the user's familiarity with that industrial product type to obtain a first reference value. If the user's current consultation involves multiple industrial product types, the arithmetic mean of the user's familiarity with all industrial product types is calculated as the familiarity average. The initial expertise level is multiplied by the familiarity average to obtain the first reference value. If the user already has a historical skill level, the minimum value between the first reference value and the historical skill level is used as the user's initial skill level benchmark for each industrial product type. If this is the user's first consultation or the user's historical skill level does not exist, the first reference value is directly used as the user's initial skill level benchmark for each industrial product type.

[0053] Since initial professionalism reflects a user's theoretical knowledge level demonstrated through verbal expression, while familiarity reflects a user's operational experience accumulated through historical business data, and these two aspects characterize a user's ability from different dimensions—the former reflecting "how much they know" and the latter reflecting "how much they have done"—neither dimension alone is sufficient to comprehensively assess a user's operational level. Therefore, multiplying the two allows for a balance between the user's theoretical knowledge and practical experience in the initial stage.

[0054] The first reference value reflects the combined effect of the user's theoretical knowledge level and operational experience with a certain type of industrial product.

[0055] Generally speaking, it's better to underestimate a user's initial operational ability than to overestimate it and lead to inappropriate subsequent guidance strategies. Therefore, by taking the minimum value between the first reference value and the historical average value, the system can ensure that subsequent interactions start at a more conservative level, avoiding providing users with guidance beyond their capabilities due to a single assessment bias.

[0056] S103: After the initial level benchmark is determined, a diagnostic command is issued to the user; the user's response information to the diagnostic command is obtained, and the initial level benchmark is corrected according to the response information to obtain the user's true level value.

[0057] Since the system has not yet made a preliminary estimate of the user's operational capabilities before the initial skill level is determined, issuing diagnostic commands prematurely would make it impossible to select appropriate command difficulty and delivery style based on the user's skill level. Therefore, it is necessary to first estimate the initial skill level and then issue appropriate diagnostic commands based on that initial skill level.

[0058] Diagnostic instructions include standard operating procedures to instruct users to perform specified tests or operations on industrial products. For example, "Please measure the bearing temperature and report the value" or "Please check if the fan is operating normally."

[0059] It should be noted that the issuance of diagnostic instructions can be achieved using conventional push notification technology, which is common knowledge in the field. Specifically, the system sends the instruction text to the user's terminal device (such as a smartphone, industrial handheld terminal, or personal computer), and the user receives and views the instruction content through the terminal device.

[0060] It should be noted that after the user completes the operation according to the diagnostic instructions, the response information is sent back through the terminal device. The technologies involved are all conventional human-computer interaction methods, which will not be elaborated here. Specifically, the user can directly enter text or values ​​in the input box of the terminal device's interactive interface, or the system can automatically transcribe the voice input into text. The user can also complete the data transmission by selecting preset options (such as drop-down menus, radio buttons, etc.). After receiving the transmitted information, the system extracts its timestamp and detection data content to obtain structured response information.

[0061] As an example, fault feature information is extracted from the effective word sequence; based on the industrial product type and fault feature information, the corresponding diagnostic instructions are matched from the preset guidance library; then, it is determined whether a diagnostic instruction has been successfully matched from the preset guidance library; if no effective fault feature information is extracted, or no corresponding diagnostic instruction is matched, it is determined that the user input cannot guide the user into the standardized troubleshooting process, the system terminates the automatic response process, and transfers the inquiry to human customer service for intervention.

[0062] If a matching diagnostic command is successfully found, a diagnostic command will be issued to the user to instruct the user to perform the specified test or operation.

[0063] It should be noted that extracting fault feature information from the effective word sequence can be achieved through keyword matching using a pre-set fault feature word list, which is consistent with the principle of conventional keyword retrieval and is common knowledge in this field. Alternatively, text classification or sequence labeling models based on natural language processing can be used as alternative methods. Specifically, based on common fault descriptions in industrial after-sales systems, the system can pre-set a "fault feature word list" (e.g., containing words such as "stuck," "abnormal noise," "leakage," "not rotating," "high temperature," and "vibration"), and directly match it from the effective word sequence; a match indicates the extraction of fault feature information.

[0064] The pre-defined guidance library is a structured knowledge base pre-stored in the system to record standardized diagnostic processes for various industrial product types under different fault characteristics. Specifically, for common fault modes of various industrial product types, domain experts pre-define standardized troubleshooting steps, and convert each step into a diagnostic instruction, which is then stored in the pre-defined guidance library in the order of execution.

[0065] For example, for the index key "Incinerator + stuck", the preset guidance library can store a series of sequentially arranged diagnostic instructions such as "Please check if the grate drive motor is running normally" and "Please measure the tension value of the grate transmission chain".

[0066] The response information includes the time information and / or detection data information of the user's returned operation results.

[0067] The time information of the operation result refers to the timestamp recorded when the user submits the operation result through the terminal device after completing the operation required by the diagnostic command.

[0068] Test data refers to the specific data returned by the user after performing tests or operations according to the diagnostic instructions. For example, if the diagnostic instruction requires "Please measure the bearing temperature and report the value", the test data is the temperature value returned by the user (e.g., 85℃); if the diagnostic instruction requires "Please confirm whether the fan is operating normally", the test data is the confirmation result returned by the user (e.g., operating normally).

[0069] The process of determining a user's true skill level is as follows: Figure 2 As shown, it includes: S201: Obtain the first timestamp of the issued diagnostic command, and the second timestamp of the user's feedback operation result extracted based on the time information; and obtain the preset standard consumption time and the preset tolerance duration; calculate the sum of the preset standard consumption time and the preset tolerance duration as the allowed consumption time.

[0070] The preset standard time refers to the reasonable time required for a skilled operator to complete the actions required by the same diagnostic instruction under standard operating procedures.

[0071] The preset tolerance period is a fixed tolerance used to offset the conventional physical time consumption caused by network communication transmission, user reading and understanding instructions, and terminal input.

[0072] It should be noted that the specific values ​​of the preset standard time and preset tolerance time can be determined based on engineering experience, and this embodiment does not impose specific limitations. For example, the preset standard time for the diagnostic command "measure bearing temperature" can be set to 30 seconds, and the preset standard time for the diagnostic command "replace filter element" can be set to 120 seconds; the preset tolerance time can be set to 30 to 60 seconds.

[0073] In addition, during the process of obtaining the second timestamp, the system sets a maximum waiting time. If no operation result is received from the user after the maximum waiting time since the first timestamp of the diagnostic command was issued, the waiting is terminated and the second timestamp acquisition is determined to have failed. At this time, the actual time taken by the user in this operation is set to a preset maximum timeout, which is greater than the allowed timeout, to ensure that the interaction in this round generates a positive timeout in subsequent calculations, thereby obtaining the corresponding delay penalty value.

[0074] It should be noted that the specific value of the maximum waiting time can be determined based on engineering experience, and this embodiment does not impose a specific limitation. For example, the maximum waiting time can be between 120 seconds and 300 seconds.

[0075] It should be noted that the specific value of the preset maximum time consumption should be ensured to be greater than the allowable time consumption under normal operation, and can be adjusted according to the actual needs of the business scenario. For example, the preset maximum time consumption can be set to 3600 seconds.

[0076] S202: Calculate the difference between the second timestamp and the first timestamp as the actual time consumed by the user in this operation; calculate the difference between the actual time consumed and the allowed time consumed as the timeout amount; if the timeout amount is not greater than zero, set the delay penalty value to zero; if the timeout amount is greater than zero, normalize the timeout amount to obtain the normalized timeout amount; multiply the normalized timeout amount by the preset penalty coefficient and take the negative value as the delay penalty value.

[0077] It should be noted that if the timeout is not greater than zero, it means that the user has completed the operation within the allowed time, and the delay penalty value can be zero.

[0078] A specific example of normalizing the timeout is as follows: divide the timeout by (the sum of the actual time and the preset standard time) to compress the timeout to a controllable ratio range, thus obtaining the normalized timeout, where the normalized timeout is between 0 and 1, and the more severe the timeout, the closer it is to 1.

[0079] The preset penalty coefficient is used to control the overall sensitivity of the delay penalty. The larger the value, the heavier the penalty for the same amount of timeout; the smaller the value, the lighter the penalty.

[0080] It should be noted that the specific value of the preset penalty coefficient can be determined based on the type of industrial product or the urgency of the failure scenario; this embodiment does not impose a specific limitation. For example, the preset penalty coefficient can be set between 0.3 and 0.5.

[0081] Since the delay penalty is a negative evaluation of the user's timed-out operation, the negative value reflects the punishment for the timed-out operation. The more timed out, the heavier the punishment. However, the normalization process limits the penalty value to a controllable range, avoiding an excessive drop in the level value due to a single serious timeout.

[0082] S203: Calculate the number of data dimensions requested by the diagnostic command and the number of valid data dimensions that meet the preset validity conditions in the test data information returned by the user; calculate the ratio of the number of valid data dimensions to the number of data dimensions as the feedback accuracy of the user's response.

[0083] The number of data dimensions refers to the total number of test data items that a diagnostic command requires the user to return. For example, if the command "Please measure the bearing temperature and record the vibration amplitude" requests two data items, namely temperature and vibration amplitude, then the number of data dimensions is 2.

[0084] The number of valid data dimensions refers to the number of dimensions in the detection data returned by the user that meet the preset validity conditions. For example, if the instruction requests two dimensions, temperature and vibration amplitude, and the user returns both the temperature value (within the valid range) and the vibration amplitude value (within the valid range), then the number of valid data dimensions is 2; if the user only returns the temperature value and not the vibration amplitude value, then the number of valid data dimensions is 1.

[0085] Preset validity conditions refer to the legal range and format requirements for each data dimension preset by the system. These can be set according to industry standards, and this embodiment does not impose specific limitations. For example, the valid range for temperature is -20℃ to 200℃. If the temperature value returned by the user falls within this range, it is counted as a valid dimension; if it exceeds this range, it is not counted.

[0086] It should be noted that the statistics on the number of data dimensions are essentially reading preset structured data fields; the statistics on the number of valid data dimensions are essentially checking whether the data returned by the user meets the preset validity conditions (such as whether the field exists or whether the value is within the threshold range). Both are conventional data statistics and condition judgment processes in existing technologies and can be directly applied here.

[0087] Furthermore, based on this, it is determined whether the number of data dimensions is zero. If the number of data dimensions is zero, it means that the diagnostic command did not request any test data. In this case, the feedback accuracy has no distinguishing significance for the evaluation of the user's operation quality. Therefore, there is no need to perform division. The feedback accuracy of this response is directly set to the preset full value, that is, the feedback accuracy is recorded as 1. If the number of data dimensions is not zero, the ratio of the number of valid data dimensions to the number of data dimensions is calculated as the feedback accuracy of the user's response.

[0088] Feedback accuracy is used to quantify the user's understanding and execution quality of diagnostic commands. Higher feedback accuracy indicates that the user's returned data covers more request dimensions, reflecting a more complete and accurate understanding and execution of the diagnostic commands. Conversely, lower feedback accuracy indicates that the user has missed more request dimensions or returned more invalid data, reflecting lower execution quality of the commands.

[0089] S204: If the current round only has a delay penalty value, use the delay penalty value as the observation level value for the current round; if the current round only has feedback accuracy, use the feedback accuracy as the observation level value for the current round; if the current round has both a delay penalty value and feedback accuracy, add the delay penalty value and the feedback accuracy to obtain the observation level value for the current round.

[0090] Because both delay penalty and feedback accuracy exist, a user's performance in this round of operations can be evaluated from two dimensions: feedback accuracy reflects the completeness of the user's actions and the accuracy of data feedback, and is a positive evaluation indicator; delay penalty reflects the timeliness of the user's actions, and is a negative evaluation indicator (usually zero or negative). Therefore, feedback accuracy provides a positive score, and the delay penalty is deducted from it, comprehensively reflecting both "whether it was done correctly" and "how quickly it was done" on the same numerical basis.

[0091] S205: Calculate the deviation between the observed level value in this round and the corrected true level value in the previous round; multiply the deviation value by the preset step size to obtain the adjustment amount in this round; add the corrected true level value in the previous round to the adjustment amount in this round to obtain the corrected true level value in this round; wherein, in the first round of correction, the initial level benchmark is used as the corrected true level value in the previous round.

[0092] It should be noted that the preset step size is used to control the magnitude of a single correction, and its specific value can be adjusted according to actual needs. For example, for scenarios that need to quickly adapt to changes in user skill level, a larger step size value, such as 0.4, can be selected; for scenarios that need to maintain evaluation stability, a smaller step size value, such as 0.2, can be selected.

[0093] In addition, upper and lower limits need to be protected for the corrected true level value: if the corrected true level value is less than zero, it should be set to zero; if the corrected true level value is greater than one, it should be set to one.

[0094] S104: Based on the actual level values, determine the professional density and amount of supporting materials for the after-sales scripts for each type of industrial product in this round, generate corresponding response scripts and match supporting materials, integrate them and output them to the user in a unified manner to complete this round of consultation service.

[0095] It's important to understand that users have significantly different skill levels. Expert users typically only need a few technical terms to understand the key points, while novice users require detailed, non-technical guidance, along with multimodal supplementary materials such as images and 3D diagrams to complete the operation. If a uniform response script (i.e., the same level of technical detail) and the same amount of supplementary materials are used for all users, the user experience will be inconsistent: "experts find it verbose and time-consuming," while "novices find it obscure and lacking in guidance." Therefore, the system can dynamically adjust the technical detail of the response script and the amount of supplementary materials based on the actual skill level. This allows high-level users to receive concise, professional, and efficient guidance, while low-level users receive detailed, easy-to-understand, and multi-dimensional support, thus achieving differentiated service.

[0096] As a preferred implementation, for the various industrial product types involved in this round of consultation, the highest preset professional density for each industrial product type is obtained; the actual level value is normalized; the normalized actual level value is multiplied by the preset highest professional density to obtain the professional density of the response script for this round; the difference between the value 1 and the normalized actual level value is calculated as the actual difference; the actual difference is multiplied by the preset maximum additional quantity, and the product result is rounded down to obtain the rounded result; the rounded result is added to the preset base quantity to obtain the number of auxiliary materials for this round of response.

[0097] It should be noted that the specific value of the preset maximum professional density can be pre-set by domain experts based on the complexity of professional knowledge in different fault scenarios, or it can be determined by statistical analysis of the density of professional terms in the response scripts of experts and engineers in the field. This embodiment does not impose a specific limitation. For example, the preset maximum professional density in a specific scenario can be set to 0.9, which indicates that the proportion of professional terms in the most professional version of the script template library is 90% in that scenario.

[0098] It is understandable that the maximum professional density can be set separately for different types of industrial products and different fault characteristics. For example, the maximum professional density preset for the "incinerator + jamming" scenario can be set to 0.9, and the maximum professional density preset for the "incinerator + abnormal noise" scenario can be set to 0.85.

[0099] Since the true level value has been limited to between 0 and 1 in the previous correction steps, the normalization process here can directly use the true level value as the normalized value without additional mapping. If the rules for taking the true level value change (for example, adopting a rating system of 0 to 100), then a linear scaling method can be used to map it to the 0 to 1 range, that is, the true level value is divided by 100.

[0100] The preset maximum additional quantity is used to control the adjustment range of the amount of supplementary data as the user's actual level changes.

[0101] It should be noted that the specific value of the preset maximum additional quantity can be determined based on the total number of data in the auxiliary database corresponding to the industrial product type and fault characteristics. It usually does not exceed the total number of data in the auxiliary database for the scenario minus the preset base quantity. For example, if there are 5 auxiliary data for a certain scenario in the auxiliary database and the preset base quantity is 1, then the preset maximum additional quantity can be set to 4.

[0102] It is understandable that the higher the actual skill level value, the closer the professional density is to the preset maximum professional density, and the more professional the language; the lower the actual skill level value, the lower the professional density, and the more colloquial the language.

[0103] The preset minimum amount of information is to ensure that even users with the highest actual skill levels receive basic support materials, avoiding situations where there is "zero support materials." For example, at least one image of the device or a basic explanatory text.

[0104] It should be noted that the specific value of the preset base quantity can be determined based on actual needs, and this embodiment does not impose a specific limitation. For example, the preset base quantity can be 1.

[0105] In this embodiment, a pre-set auxiliary data library is obtained, which contains multiple auxiliary materials for each type of industrial product. Each auxiliary material has a preset complexity level and is sorted from low to high complexity. The auxiliary materials include at least one of text descriptions, images, 3D structural diagrams, and videos. A pre-set script template library is also obtained, which contains multiple script templates for each type of industrial product. Each script template has a preset professional density level. For each type of industrial product involved in this round of consultation, multiple script templates corresponding to that industrial product type are obtained from the script template library. Based on this round... The number of supplementary materials is determined by selecting the top number of materials from the sorted supplementary materials in ascending order of complexity. Based on the professional density of the response script in this round, the script template with the closest professional density level is selected from the multiple script templates and used as the response script for this round. If the consultation involves multiple industrial product types, corresponding response scripts and supplementary materials are generated for each industrial product type. The response scripts and supplementary materials for each industrial product type are then summarized to generate an integrated response. This integrated response is then uniformly output to the user in the same round of responses.

[0106] It should be noted that the construction of the auxiliary database can be achieved using conventional data management techniques. Specifically, for each type of industrial product and its fault characteristics, domain experts pre-compile corresponding operational guidance materials (including at least text descriptions, images, 3D structural diagrams, and videos). These materials are labeled according to their complexity level (e.g., text, images, 3D structural diagrams, and videos, from low to high), and stored categorized by industrial product type and fault characteristics. When each material is added to the database, the system adds type tags (i.e., text / image / 3D diagram / video) and association tags (related industrial product type or fault characteristics), forming a searchable auxiliary database. This construction method is consistent with the database construction method of conventional multimedia data management systems and is a standard technical approach in this field.

[0107] It should be noted that the construction of the script template library can be achieved using conventional template management techniques. Specifically, for each type of industrial product and its fault characteristics, domain experts pre-write multiple response script templates with different levels of professional density, ranging from completely colloquial to highly specialized. Each template is labeled with its corresponding level of professional density and stored according to industrial product type or fault characteristics. The construction method of the above-mentioned script template library is consistent with the database construction method of conventional script template management systems and is a conventional technical means in this field.

[0108] It is understandable that the higher the professional density level, the more professional terms are used in the script template; the lower the professional density level, the more the script template tends to use colloquial expressions.

[0109] S105: After completion, calculate the interaction efficiency based on the interaction step information obtained from this consultation, and iteratively update the user's historical level value in combination with the actual level value.

[0110] As a preferred implementation, the historical level values ​​saved by the user in the cloud are obtained; the total number of interactions with the user is extracted from the interaction step information of this consultation as the actual number of steps; the preset ideal number of steps for each industrial product type is extracted; and the number of steps in which the user repeatedly asked the same question during the interaction is extracted as the repetition step number; the ratio of the preset ideal number of steps to the actual number of steps is calculated as the step efficiency factor; the ratio of the repetition step number to the actual number of steps is calculated as the repetition factor; and the difference between the value 1 and the repetition factor is calculated as the non-repetition question factor; the step efficiency factor is multiplied by the non-repetition question factor, and then multiplied by the preset transition marker coefficient to obtain the interaction efficiency; wherein, if this consultation... If the consultation is transferred to a human agent, the transfer marker coefficient is recorded as 0; otherwise, it is recorded as 1. If this is the user's first consultation, the initial level benchmark of this consultation is used as the historical level value. Otherwise, the difference between the final determined real level value and the historical level value at the end of this consultation is calculated as the level change. The level change, interaction efficiency, and preset growth factor are multiplied to obtain the effective progress. The historical level value and the effective progress are added to obtain the initial upgrade value. The minimum value between the real level value and the initial upgrade value is taken as the post-upgrade level value. The upgraded level value is saved as the historical level value for subsequent consultation services in the cloud database.

[0111] In addition, if the actual number of steps is zero, or the number of repeated steps is greater than or equal to the actual number of steps, the interaction efficiency is set to zero.

[0112] Actual steps: The total number of rounds of interaction between the system and the user, i.e., the sum of the number of question-and-answer rounds in this consultation, where the system issues instructions or replies and the user responds or asks questions. This can be counted in real-time by the system during the service process and read directly after the service ends.

[0113] The preset ideal number of steps: This is the minimum number of interaction rounds required to complete the consultation for this scenario under ideal conditions, based on the type of industrial product and its fault characteristics. Its specific value can be set according to the type of industrial product; this embodiment does not impose any specific limitations.

[0114] Repeat steps: The number of steps a user takes to repeatedly ask the same question during the interaction. The system can determine whether the user's current question is highly similar to a previous question based on semantic similarity, and whether the system has already responded to the same question before. If the above conditions are met, the question is counted as a repeat step.

[0115] It should be noted that the preset ideal number of steps refers to the minimum number of interaction rounds required to complete the consultation under ideal conditions for each type of industrial product and fault characteristics. The specific value is determined by domain experts according to standard operating procedures, and this embodiment does not impose a specific limitation. For example, for troubleshooting a "clogged incinerator", the preset ideal number of steps is 3 steps, such as step 1: confirm the fault phenomenon; step 2: check the drive motor; step 3: check the transmission chain. For "leaking chemical protective suits", the preset ideal number of steps is 2 steps, such as step 1: check for external damage; step 2: perform an airtightness test.

[0116] The step efficiency factor reflects the difference between the actual number of steps and the ideal number of steps in the communication process. Specifically, the closer the step efficiency factor is to 1, the closer the actual number of steps is to the ideal number, and the higher the communication efficiency; conversely, the smaller the step efficiency factor, the more actual steps are taken, and the lower the communication efficiency.

[0117] The non-repeating question factor reflects the proportion of effective questions asked by users during the interaction process. A higher non-repeating question factor indicates more effective questions and higher communication efficiency; conversely, a lower non-repeating question factor indicates lower communication efficiency.

[0118] The change in level reflects the degree of improvement or regression in the user's performance relative to the past, with positive values ​​indicating improvement and negative values ​​indicating regression.

[0119] It should be noted that the specific value of the preset growth factor can be adjusted according to the actual scenario, and this embodiment does not impose a specific limitation. For example, for scenarios where skills improve rapidly (such as the new employee training period), a larger preset growth factor, such as 0.4, can be used to enable historical levels to track the user's actual growth more quickly; for scenarios where skills are relatively stable (such as senior engineers), a smaller growth factor, such as 0.2, can be used.

[0120] It's important to understand that the growth factor's role is to prevent drastic fluctuations in historical performance values ​​caused by a single abnormal performance (such as a user's exceptionally good or bad performance at one time), making the updates of historical performance values ​​smoother and more stable. Interaction efficiency, on the other hand, serves as a weight for the credibility of the current level change. That is, the higher the interaction efficiency, the more credible the current level change is, and the closer the current level change is to the actual amount of progress. The lower the interaction efficiency, the less credible the current level change is, and the current level change is compressed accordingly.

[0121] It's important to note that taking the minimum of the actual performance value and the initial upgrade value is a conservative strategy. Its core purpose is to prevent historical performance from being artificially inflated to exceed the actual performance value observed in the current instance. Specifically, if a user's current performance is better than historical performance (i.e., the actual performance value is greater than historical performance), taking the smaller value ensures that the upgraded performance value does not exceed the current actual performance value, avoiding overestimation of the user due to a single good performance. Conversely, if a user's current performance is worse than historical performance (i.e., the actual performance value is less than historical performance), the initial upgrade value is already lower than historical performance due to the negative difference, and taking the smaller value naturally reverts to a lower actual performance value.

[0122] In addition, based on the conservative minimum value, it is determined whether the current actual level value is greater than the historical level value. If the current actual level value is greater than the historical level value, it indicates that the user's performance is better than the historical level. To prevent over-upgrading caused by a single improvement, the minimum value between the actual level value and the initial upgrade value is taken as the level value after this upgrade. If the current actual level value is less than or equal to the historical level value, it indicates that the user's performance is flat or regresses. In this case, no truncation is performed, and the initial upgrade value is directly taken as the level value after this upgrade, so as to retain the damping buffer effect of the preset growth factor on the magnitude of regression.

Claims

1. An automatic interactive response method for an after-sales consultation service platform, characterized in that, The method includes: Obtain the user's input information in the preset initial interaction round of the current consultation service and historical business data related to the industrial product type involved in the input information, and preprocess the input information to obtain an effective word sequence; The user's initial level of expertise is determined based on the effective word sequence, and the user's familiarity with each type of industrial product is determined based on historical business data. The user's initial level benchmark for each type of industrial product is determined based on the initial level of expertise, familiarity with each product, and the user's existing historical level values. After the initial level benchmark is determined, a diagnostic command is issued to the user; the user's response information to the diagnostic command is obtained, and the initial level benchmark is corrected based on the response information to obtain the user's true level value; Based on the actual performance level, determine the professional density and amount of supporting materials for the after-sales scripts for each type of industrial product in this round, generate corresponding response scripts and match supporting materials, integrate them and output them to users in a unified manner to complete this round of consultation service; Once completed, the interaction efficiency is calculated based on the number of interaction steps obtained from this consultation, and the user's historical level value is iteratively updated in combination with the actual level value.

2. The automatic interactive response method for an after-sales consultation service platform according to claim 1, characterized in that, The process of obtaining user input information from the current consultation service in a preset initial interaction round and historical business data related to the industrial product type involved in the input information, and preprocessing the input information to obtain a valid word sequence, includes: Obtain the input text entered by the user in the preset initial interaction rounds of this consultation service as input information, and simultaneously obtain the user account and the device tag provided by the user; Based on the user account, the system retrieves the user's purchase list and maintenance records within a preset historical period from the supply chain database as historical business data; it also performs text cleaning, Chinese word segmentation, and part-of-speech tagging on the input text to obtain a valid word sequence. Identify the type of industrial product corresponding to this round of consultation based on the equipment label and / or the entered text; For various industrial product types, the system uses industrial product type as the filtering condition to filter out purchase records of the same type from the purchase list; and uses user account and industrial product type as the combined filtering condition to filter out historical repair or consultation records of industrial product types from the maintenance records. The effective word sequence, the filtered purchase records, and the maintenance records are stored together using the industrial product type as the association key.

3. The automatic interactive response method for an after-sales consultation service platform according to claim 1, characterized in that, The initial professionalism determination process includes: Obtain a pre-built professional thesaurus, which includes at least a root word field, a synonym phrase field, and a part-of-speech tag field. The root word field is used to store benchmark technical terms; the synonym phrase field is associated with the root word field and is used to store at least one synonymous expression of the benchmark technical terms; the part-of-speech tag field is used to mark the part-of-speech category of the benchmark technical terms. Each word in the effective word sequence is matched sequentially with word roots and synonyms; words that match word roots or synonyms are recorded as professional words, and the number of professional words in the effective word sequence is counted as the number of professional words. Repeated words in the effective word sequence are deduplicated; the total number of words in the effective word sequence after deduplication is counted as the effective vocabulary size; If the effective vocabulary is zero, the user's initial professional level is set to zero; otherwise, the proportion of professional vocabulary to the effective vocabulary is calculated and used as the user's initial professional level.

4. The automatic interactive response method for an after-sales consultation service platform according to claim 2, characterized in that, The process for determining the user's familiarity with each type of industrial product includes: For various types of industrial products, obtain the filtered purchase records and maintenance records; calculate the purchase quantity of industrial product types involved in the filtered purchase records, and the number of times users reported or inquired about industrial product types in the filtered maintenance records. If the purchase quantity and the number of repair requests or inquiries are both zero, the user's familiarity with the type of industrial product will be set to zero. Otherwise, a weighted sum is calculated by using preset weighting coefficients to calculate the weighted sum of the purchase quantity and the number of repair requests or inquiries; where the weighting coefficient corresponding to the number of repair requests or inquiries is greater than the weighting coefficient corresponding to the purchase quantity. The weighted sum is normalized to obtain the user's familiarity with the type of industrial product.

5. The automatic interactive response method for an after-sales consultation service platform according to claim 1, characterized in that, The process of determining the user's initial skill level benchmark for each industrial product type based on initial expertise, familiarity with various aspects, and the user's existing historical skill level includes: Retrieve the user's existing historical level values; If the user's consultation in this round only involves one type of industrial product, the initial level of expertise is multiplied by the user's familiarity with that type of industrial product to obtain the first reference value; When a user is involved in multiple industrial product types during this round of consultation, the arithmetic mean of the user's familiarity with all industrial product types is calculated as the average familiarity level; the initial level of expertise is multiplied by the average familiarity level to obtain the first reference value. If the user already has historical level values, the minimum value between the first reference value and the historical level values ​​will be used as the user's initial level benchmark for each industrial product type. If the user is consulting for the first time, or if the user's historical level values ​​are not available, the first reference value will be used directly as the user's initial level benchmark for each industrial product type.

6. The automatic interactive response method for an after-sales consultation service platform according to claim 1, characterized in that, The response information includes the time information and / or detection data information of the user's returned operation results; The process of determining the user's true skill level includes: Obtain the first timestamp of the issued diagnostic command, and the second timestamp of the user's feedback operation result extracted based on the time information; obtain the preset standard consumption time and the preset tolerance time; calculate the sum of the preset standard consumption time and the preset tolerance time as the allowed consumption time; Calculate the difference between the second timestamp and the first timestamp as the actual time consumed by the user's operation; calculate the difference between the actual time consumed and the allowed time consumed as the timeout amount; if the timeout amount is not greater than zero, set the delay penalty value to zero; if the timeout amount is greater than zero, normalize the timeout amount to obtain the normalized timeout amount; multiply the normalized timeout amount by the preset penalty coefficient and take the negative value as the delay penalty value. The system calculates the number of data dimensions requested by the diagnostic command and the number of valid data dimensions that meet the preset validity conditions in the test data returned by the user; it then calculates the ratio of the number of valid data dimensions to the total number of data dimensions as the accuracy of the user's response. If the current round only has a delay penalty value, the delay penalty value is used as the observation level value for this round; if the current round only has feedback accuracy, the feedback accuracy is used as the observation level value for this round; if the current round has both a delay penalty value and feedback accuracy, the delay penalty value and feedback accuracy are added together to obtain the observation level value for this round. Calculate the deviation between the observed level value in this round and the corrected true level value in the previous round; multiply the deviation value by the preset step size to obtain the adjustment amount in this round; add the corrected true level value in the previous round to the adjustment amount in this round to obtain the corrected true level value in this round; wherein, in the first round of correction, the initial level benchmark is used as the corrected true level value in the previous round.

7. The automatic interactive response method for an after-sales consultation service platform according to claim 1, characterized in that, The process for determining the professional density and the amount of supporting data includes: For the various types of industrial products involved in this round of consultation, obtain the highest pre-defined professional density for each type of industrial product. Normalize the actual level value; multiply the normalized actual level value by the preset maximum professional density to obtain the professional density of the response script for this round. Calculate the difference between the value 1 and the normalized true level value as the true difference; multiply the true difference by the preset maximum extra amount, and round down the product to obtain the rounded result. Add the rounded-down result to the preset base quantity to obtain the number of supporting materials for this round of responses.

8. The automatic interactive response method for an after-sales consultation service platform according to claim 7, characterized in that, The process of generating corresponding response scripts, matching them with supporting materials, and then integrating them for unified output to the user includes: Obtain a pre-set auxiliary data library, which contains multiple auxiliary data for each type of industrial product. Each auxiliary data has a different complexity level and is sorted from low to high complexity level. The auxiliary data includes at least one of the following: text descriptions, images, 3D structural diagrams, and videos. Access the pre-set script template library, which contains multiple script templates for various industrial product types, each with different levels of professional density. For the various types of industrial products involved in this round of consultation, multiple script templates corresponding to each type of industrial product are obtained from the script template library; based on the number of supporting materials in this round of responses, the top number of supporting materials are selected from the sorted supporting materials in order of complexity from low to high. Based on the professional density of the response scripts in this round, the script template with the closest professional density level is selected from the multiple script templates obtained and used as the response script for this round. In this round of consultation involving multiple industrial product types, corresponding response scripts and supporting materials were generated for each industrial product type; the response scripts and supporting materials for each industrial product type were summarized to generate integrated response content; and the integrated response content was uniformly delivered to the user in the same round of responses.

9. The automatic interactive response method for an after-sales consultation service platform according to claim 1, characterized in that, The process of calculating interaction efficiency based on the interaction step count information obtained from this consultation, and iteratively updating the user's historical level value in conjunction with the actual level value, includes: Obtain the user's historical level values ​​saved in the cloud; extract the total number of rounds of interaction with the user from the interaction step information of this consultation, as the actual number of steps; extract the ideal number of steps preset for each type of industrial product; and extract the number of steps in which the user repeatedly asked the same question during the interaction, as the repetition number. Calculate the ratio of the preset ideal number of steps to the actual number of steps as the step efficiency factor; calculate the ratio of repeated steps to the actual number of steps as the repetition factor; and calculate the difference between the value 1 and the repetition factor as the non-repetition question factor. The interaction efficiency is obtained by multiplying the step efficiency factor by the non-repeated question factor, and then by the preset transfer mark coefficient. If the consultation process involves a human agent, the transfer mark coefficient is recorded as 0; otherwise, it is recorded as 1. If this is the user's first consultation, the initial level benchmark of this consultation is used as the historical level value; otherwise, the difference between the final determined real level value at the end of this consultation and the historical level value is calculated as the level change amount; the level change amount, interaction efficiency and preset growth factor are multiplied together to obtain the effective progress amount; the historical level value is added to the effective progress amount to obtain the preliminary upgrade value. The minimum value between the actual level value and the initial upgrade value is taken as the level value after this upgrade; the upgraded level value is saved to the cloud database as the historical level value for subsequent consulting services.

10. An automatic interactive response system for an after-sales consultation service platform, characterized in that, The system includes a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement the steps of the method as described in any one of claims 1 to 9.

Images