An intelligent customer service question and answer method and system, an electronic device, and a readable storage medium

Abstract

The application provides an intelligent customer service question and answer method and system, an electronic device and a readable storage medium. The method comprises the following steps: obtaining a consultation text input by a user and a current business node in a current employee business handling process; performing semantic error correction on the consultation text to obtain a corrected text; performing preliminary intent recognition on the corrected text by using a pre-trained intent recognition model to obtain one or more candidate user intents and a confidence of the candidate user intents; filtering the candidate user intents according to the current business node to retain a target intent within a range of execution allowed by the current business node; calling a business interface corresponding to the target intent based on the target intent; identifying a target entity in the corrected text; and generating a response result for the consultation text based on the target entity and the business interface. The application greatly improves the intent recognition and response accuracy of intelligent customer service in a human resource service scenario.

Description

Technical Field

[0001] This invention relates to the field of intent recognition technology, and more specifically, to an intelligent customer service question-and-answer method, system, electronic device, and readable storage medium. Background Technology

[0002] As enterprises deepen their digital transformation, employee self-service has become an important part of human resource management. In daily operations, it has become commonplace for employees to handle personnel, payroll, attendance, and benefits matters through online platforms. After completing these transactions, follow-up consultation and service support are crucial aspects of the employee experience, directly impacting employee satisfaction, organizational effectiveness, and the reputation of HR (Human Resources) services.

[0003] However, existing intelligent customer service systems generally lack robust mechanisms for handling ambiguous user input, spelling errors, or unclear intent when processing employee services and business transactions. Employees often use colloquial expressions, misspellings, omitted subjects, or unclear references when submitting inquiries. For example, statements like "I want to pay my social security contributions," "I need to make up for insufficient annual leave," or "My salary was too low last month" are easily misinterpreted by the system as "repayment" or "supplementary materials," or the system may fail to accurately determine the relevant pay cycle and details of the discrepancy. This can lead to failed intent recognition or incorrect guidance, ultimately impacting business processing efficiency and employee trust. Summary of the Invention

[0004] In view of this, the purpose of the present invention is to provide an intelligent customer service question-and-answer method, system, electronic device and readable storage medium to improve the accuracy of intent recognition and response.

[0005] Firstly, this application provides an intelligent customer service question-and-answer method, applied to employee service and business processing scenarios in human resources systems, including: Obtain the user's input consultation text and the current business node in the current employee's business processing flow; Perform semantic correction on the consultation text to obtain the corrected text; Based on the corrected text, a pre-trained intent recognition model is used to perform preliminary intent recognition, resulting in one or more candidate user intents and their confidence levels. Filter candidate user intents based on the current business node, and retain target intents that are within the execution range allowed by the current business node; Invoke the business interface corresponding to the target intent based on the target intent; Identify target entities in the corrected text; Generate response results for the consultation text based on the target entity and business interface.

[0006] Optionally, semantic correction is performed on the consultation text to obtain the corrected text, which includes: Identify the current language type of the consultation text; Based on language type, call the human resources domain dictionary corresponding to the language type; Spelling checks were performed on the words in the consultation text based on a human resources dictionary. If there are misspelled words, then the misspelled words will be matched with words in the dictionary based on their similarity. Replace misspelled words with the most similar match to obtain the corrected text.

[0007] Optionally, the target entities identified in the corrected text include: Extract the initial entities from the corrected text; Obtain employee business transaction records within a user-preset number of days, and the structured data of each record in the employee business transaction records; Obtain users' historical behavior logs in the human resources system; The initial entity is deambiguously processed based on the initial entity, the structured data of each record, and the historical behavior logs to obtain the target entity.

[0008] Optionally, after obtaining the target entity, the method further includes: Extract the target entity from the context of the user-input consultation text; The initial entities in the currently input consultation text are completed based on the target entities in the context.

[0009] Optionally, after obtaining the target intent, the method further includes: If multiple target intentions exist, generate a clarification query message based on the multiple target intentions; Retrieve clarification feedback results in response to clarification inquiry messages; The final target intent is determined based on the clarification feedback results.

[0010] Optionally, before generating a response to the consultation text based on the target entity and business interface, the method further includes: Determine the prerequisites for the business actions corresponding to the target intent based on the target intent; Check whether the prerequisite execution conditions have been met; If any conditions are not met, then an aggregated follow-up instruction containing all missing conditions is generated.

[0011] Optionally, the method further includes: An interaction log is generated based on the user's original input, the corrected text, the candidate user intent and the confidence level of the candidate user intent, and the final target intent, and the interaction log is labeled with log tags. The sample score result for each interaction log is determined based on the confidence of multiple candidate user intents in each interaction log and a pre-built sample scoring model. Compare the sample scoring results with a preset sample threshold; If the sample size exceeds the threshold, the interaction log and its tag will be used as training samples to update and train the intent recognition model.

[0012] Secondly, an intelligent customer service Q&A system is provided, applied to employee service and business processing scenarios in human resources systems, including: The acquisition unit is used to acquire the consultation text input by the user and the current business node in the current employee's business processing flow; The error correction unit is used to perform semantic error correction on the consultation text to obtain the corrected text. The intent recognition unit is used to perform preliminary intent recognition based on the error-corrected text using a pre-trained intent recognition model, and obtain one or more candidate user intents and the confidence scores of the candidate user intents. The filtering unit is used to filter candidate user intents based on the current business node, and retain target intents that are within the execution range allowed by the current business node; The invocation unit is used to invoke the business interface corresponding to the target intent based on the target intent. The recognition unit is used to identify target entities in the corrected text. The generation unit is used to generate response results for the consultation text based on the target entity and business interface.

[0013] Thirdly, an electronic device is provided, including a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; The memory is used to store computer programs; When the processor executes a program stored in the memory, it implements any of the methods described in the first aspect.

[0014] Fourthly, a computer-readable storage medium is provided, wherein a computer program is stored therein, and when executed by a processor, the computer program implements any of the methods described in the first aspect.

[0015] This invention provides an intelligent customer service question-and-answer method, system, electronic device, and readable storage medium. The method involves: acquiring user-inputted inquiry text and the current business node in the current employee's business processing flow; performing semantic correction on the inquiry text to obtain corrected text; using a pre-trained intent recognition model to perform preliminary intent recognition based on the corrected text, obtaining one or more candidate user intents and their confidence levels; filtering candidate user intents according to the current business node, retaining target intents within the execution range allowed by the current business node; calling the corresponding business interface based on the target intent; identifying target entities in the corrected text; and generating a response result for the inquiry text based on the target entity and the business interface. This invention significantly improves the tolerance for non-standard user expressions by introducing a semantic correction mechanism; filtering candidate intents based on the current business node ensures that only high-confidence operational intents are retained, enhancing the recognition efficiency and accuracy of the intelligent customer service server; and by integrating the identification of target entities with the corresponding business interfaces, it achieves end-to-end automated processing from raw input to accurate response, greatly improving the intent recognition and response accuracy of intelligent customer service in human resources scenarios.

[0016] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A flowchart of an intelligent customer service question-and-answer method provided by an embodiment of the present invention is shown; Figure 2 A flowchart of another intelligent customer service question-and-answer method provided by an embodiment of the present invention is shown; Figure 3 This diagram illustrates the structure of an intelligent customer service question-and-answer system provided in an embodiment of the present invention. Figure 4 A schematic diagram of the structure of an electronic device provided in an embodiment of the present invention is shown. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0020] This invention provides an intelligent customer service question-and-answer method, applied to employee service and business processing scenarios in a human resources system. The executing entity is an intelligent customer service server, such as... Figure 1 As shown, it includes the following steps: Step S101: Obtain the consultation text input by the user and the current business node in the current employee's business processing flow.

[0021] In this step, the employee self-service process of the human resources system includes several typical business processes, such as: Leave application process: application pending submission, under review, approved / rejected, leave completed; Salary inquiry process: monthly salary paid, payslip pending review, objection filed, completed; Social security and housing provident fund processing process: participation information pending confirmation, payment in progress, paid, exception handling in progress; Onboarding process: documents pending submission, background check in progress, onboarding training pending completion, onboarded. The current business node refers to the specific stage an employee is in within a particular business process.

[0022] When an employee submits an inquiry request through the enterprise's intelligent HR assistant (such as DingTalk, Lark, or the HR Bot in WeChat Work), the intelligent customer service server first obtains the natural language text entered by the employee, such as: "It seems that last month's salary was less than expected"; at the same time, it obtains the current business node of the employee in the "salary inquiry process" from the HR system, such as: "Monthly salary has been paid".

[0023] Step S102: Perform semantic correction on the consultation text to obtain the corrected text.

[0024] In this embodiment, since employee input often contains typos, colloquial expressions, or abbreviations, the system needs to perform semantic-level error correction. For example, "I want to request an annual salary" should be corrected to "I want to request annual leave".

[0025] Step S103: Based on the corrected text, perform preliminary intent recognition using a pre-trained intent recognition model to obtain one or more candidate user intents and the confidence level of the candidate user intents.

[0026] In this step, a deep learning-based intent recognition model (such as the BERT+Softmax architecture) can be used to encode the corrected text, converting the user's sentence into a contextual semantic vector to better capture semantic similarity and keyword relationships, thereby improving the generalization ability of intent classification. Then, the encoded semantic vector is used for prediction and classification. The model outputs one or more candidate intents and their confidence scores.

[0027] For example, for the consultation text "Last month's salary payment amount was too low", the model outputs the following candidate intents: Intent A: "Initiate a salary objection application" (confidence level: 0.88); Intent B: "Query salary details" (confidence level: 0.76); Intent C: "Inquire about the individual income tax deduction standard" (confidence level: 0.35).

[0028] Step S104: Filter candidate user intents based on the current business node, and retain target intents that are within the execution range allowed by the current business node.

[0029] Following the previous example, the intelligent customer service server determined that the employee is currently at the "Monthly Salary Paid" stage. Allowed operations at this stage include: "Query Salary Details," "Initiate Salary Objection," and "Download Payslip." Based on the confidence level of the candidate intents and business rules, the intent "Inquire about Individual Income Tax Deduction Standards" is filtered out. The system prioritizes retaining high-confidence intents directly related to the current stage.

[0030] Step S105: Invoke the business interface corresponding to the target intent based on the target intent.

[0031] Once the intelligent customer service server recognizes the target intent as "initiating a salary objection application," it calls the "Create Salary Objection Work Order" business interface in the HR system to prepare for subsequent processing. This interface will automatically associate contextual information such as employee ID, department, and salary cycle to prepare for the next step.

[0032] Step S106: Identify the target entities in the corrected text.

[0033] In this step, the system extracts key semantic entities from the corrected text "Last month's salary payment amount was too low": time entity: last month; business type: salary; problem type: amount too low.

[0034] It should be noted that this step can be performed after intent recognition or simultaneously with intent recognition.

[0035] Step S107: Generate a response result for the consultation text based on the target entity and business interface.

[0036] The system invokes the "Create Salary Discrepancy Work Order" interface, inputs information such as employee ID, salary period (July 2025), and problem description, automatically generates a work order and submits it to the compensation team for review.

[0037] Subsequently, a response result is returned to the employee: "Your salary discrepancy application for July 2025 has been submitted. The HR compensation team will verify it within 3 working days and feedback the result via email. You can also check the progress in 'My Work Orders'.".

[0038] The embodiments of the present invention significantly improve the fault tolerance for non-standard user expressions by introducing a semantic error correction mechanism; filtering candidate intents in combination with the current business node ensures that only operation intents with high confidence are retained, which enhances the recognition efficiency and accuracy of the intelligent customer service server; and through the integration of identified target entities and corresponding business interfaces, end-to-end automated processing from original input to accurate response is realized, which greatly improves the intent recognition and response accuracy of the intelligent customer service in human resource scenarios.

[0039] In multinational enterprises or global organizations, employees may use different languages (such as Chinese, English, Spanish, etc.) to submit service requests through the HR intelligent assistant. For example, a Chinese employee may input "I want to apply for annual leave", an English employee may input "I need to apply for annual leave", and a Spanish employee may input "Quiero solicitar vacaciones". However, the semantic error correction module of the existing HR intelligent customer service system is usually designed for only a single language (such as Chinese or English), and lacks the ability to automatically identify and process multilingual inputs.

[0040] Meanwhile, general spelling correction relies on standard language dictionaries, and fails to integrate human resources domain-specific terms (such as "annual leave", "probation period", "five social insurances and one housing fund", "probation period", "leave balance", "payroll cycle", "benefits enrollment", etc.), resulting in poor error correction performance for HR-specific terms.

[0041] In addition, employees are prone to problems such as miswriting of phonetically similar words, case confusion, and misuse of abbreviations when expressing across languages. For example: an English user miswrites "annual leave" as "anual leave"; a Chinese user miswrites "social insurance" as "shebao" (similar-sounding wrong characters); a Spanish user miswrites "vacaciones" as "vacacines".

[0042] If the HR intelligent customer service system cannot accurately identify the current language type and call the human resources field dictionary of the corresponding language for semantic matching, the error correction result often deviates from the employee's real intention, thereby affecting the accuracy of subsequent intention recognition and business processing.

[0043] To this end, based on the aforementioned HR intelligent service process, the "semantic error correction" step is enhanced, which specifically includes the following sub-steps: Step S102A: Identify the current language type of the consultation text.

[0044] The intelligent customer service server can adopt a language recognition algorithm based on n-gram features or a pre-trained multilingual model (such as fastText, LangDetect, etc.) to analyze the linguistic features of the input text. For example, for the text "I need toanual leave next week", the intelligent customer service server identifies its language type as "English"; for "我想请年价", it is identified as "Chinese".

[0045] Step S102B: Call the human resources field dictionary corresponding to the language type based on the language type.

[0046] The system presets multiple sets of language-exclusive human resources field dictionaries, and each set of dictionaries includes common HR terms, process names, leave types, compensation items, etc. in that language. For example: the Chinese dictionary includes: annual leave, sick leave, probation period, five social insurances and one housing fund, payslip, onboarding materials, job transfer, resignation certificate, overtime pay, etc.; the English dictionary includes: annual leave, sick leave, probation period, payroll, payslip, onboarding, benefits, resignation, overtime, etc. The system dynamically loads the corresponding language dictionary according to the identified language type.

[0047] Step S102C: Perform spelling check on the words in the consultation text based on the human resources field dictionary.

[0048] In this step, after the input consultation text is segmented or tokenized, each word is compared one by one to check whether it exists in the HR field dictionary of the current language. If it is not hit, it is marked as a suspected spelling error. For example, "anual" is not in the English HR field dictionary and is marked as an erroneous word.

[0049] Step S102D: If there is a misspelled word, perform similarity matching between the misspelled word and the words in the dictionary.

[0050] In one example, algorithms such as edit distance or cosine similarity of word vectors can be used to calculate the similarity score between the incorrect word and all words in the dictionary. For example, "anual" and "annual" have an edit distance of 1, indicating the highest similarity.

[0051] Step S102E: Replace the misspelled words with the most similar matching results to obtain the corrected text.

[0052] For example, replacing "anual" with "annual" results in the corrected text: "I want to apply for annual leave". This correction will serve as the input for subsequent intent recognition and entity extraction.

[0053] In another implementation, a mapping relationship is established between incorrect and correct words, and stored in the corresponding HR domain dictionary to update the HR domain dictionary. For example, a mapping relationship is established between "anual" and "annual", and the error correction frequency of "anual" and "annual" is recorded; when a threshold is reached (e.g., more than 10 times), "anual" is automatically added to the synonyms or error correction mapping table of the English dictionary; the next time the user enters "anual", it is automatically recognized as "annual", without repeating the error correction process, which greatly saves the intent recognition time and significantly improves the long-term operating efficiency and adaptability of the system.

[0054] This invention, through an automatic language type recognition and domain dictionary binding mechanism, ensures that spelling errors in different languages ​​can be correctly corrected in a professional context, avoiding the miscorrection problems caused by general dictionaries. By constructing a three-in-one semantic error correction mechanism of "language recognition + domain dictionary + similarity matching," it effectively solves the comprehension difficulties caused by non-standard user input in multilingual environments, significantly improving the robustness and adaptability of intelligent customer service in complex after-sales scenarios.

[0055] In human resources service scenarios, employees may be involved in multiple business processes or have multiple similar matters recently. For example, they may submit annual leave and sick leave applications simultaneously, have multiple pending job transfer processes in different departments, have multiple social security payment locations, or switch between probation and formal employment periods. Relying solely on the natural language text entered by the employee (such as "I want to change my leave time") cannot determine which specific leave record or business process it refers to.

[0056] Traditional Named Entity Recognition (NER) models are mostly trained on general corpora and do not integrate contextual data such as employees' historical business processing records, approval process status, organizational structure information, and job change logs. This results in entity recognition being detached from the real organizational context, limiting accuracy and easily leading to service mismatch, process misoperation, or compliance risks.

[0057] Therefore, based on the aforementioned intelligent HR customer service process, an entity disambiguation mechanism based on an employee-business-behavior heterogeneity graph is introduced, such as... Figure 2 As shown, the specific steps include the following: Step S106A: Extract the initial entities from the corrected text.

[0058] In this step, a context-enhanced named entity recognition model (such as BERT-BiLSTM-CRF) can be used to extract the initial entities from the corrected text.

[0059] For example, given the text "I want to change my annual leave from next week to the end of the month", the initial entities can be extracted as "next week", "end of the month", and "annual leave".

[0060] Step S106B: Obtain employee business processing records within the user's preset number of days and the structured data of each record in the employee business processing records.

[0061] The HR intelligent customer service server retrieves all relevant business processing records for the employee within the past 30 days from the enterprise's HRIS system and extracts structured information from each record. For example, business types include: leave application, job transfer approval, salary adjustment, and social security enrollment.

[0062] For example, the system found that this employee currently has two annual leave records: Record A: LEAVE-202508-001, Time: 2025-08-11 to 2025-08-15, Status: Approved; Record B: LEAVE-202507-005, Time: 2025-07-20 to 2025-07-22, Status: Completed; Therefore, "next week's annual leave" is more likely to refer to record A.

[0063] Step S106C: Obtain the user's historical behavior logs in the human resources system.

[0064] The system retrieves the employee's recent behavioral patterns from the HR platform logs, including, for example: Browsing behavior: Viewed the details page of "LEAVE-202508-001" 3 times; Action taken: Attempted to modify the leave application 3 days ago but did not submit it; Search behavior: Searching for "annual leave adjustment process"; Message interaction: I consulted the HR robot multiple times about the policy related to "change of leave time".

[0065] These behavioral data reflect the key business processes that employees are currently focusing on.

[0066] Step S106D: Based on the initial entity, the structured data of each record, and the historical behavior log, the initial entity is deambigued to obtain the target entity.

[0067] In one feasible implementation, the deambiguity processing method includes: Step 1: Using a single user inquiry as a unit, construct a local heterogeneous graph based on the initial entity, the structured data of each record, and historical behavior logs. ,in: It represents a set of nodes, containing six types of entities: users, consultation text, initial entities, business process records, business nodes, and behavioral events (browsing, searching, etc.).

[0068] The set represents the relationship between entities: for example, a user initiates a consultation text; the consultation text contains an initial entity; the initial entity matches a business process record; the user submits a business process record; the business process record belongs to an organization node; the user executes a behavior event; the behavior event acts on the business process record; the business process record is similar to other processes.

[0069] The type labels for nodes and edges are used for type-aware processing in graph neural networks.

[0070] Step 2: Encode each node in the graph using vectors to obtain the initialized feature vector; Step 3: Perform multi-layer graph attention propagation:

[0071] in, Represents a node i Neighbors; Indicates the first Attention weights of layers; Represents a node Initialize the feature vector; This represents the parameter matrix.

[0072] Through 2-3 layers of GAT propagation, information flows in the graph, eventually causing the representation of the "consultation text" node to gradually integrate semantics from paths such as historical processes, behavioral trajectories, and organizational rules.

[0073] Step 4: Entity Path Reasoning and Target Selection After GAT encoding is completed, the initial entity nodes are calculated. Record with each candidate business process semantic affinity between :

[0074] in: This represents the final embedded representation of the initial entity node after being updated by the L-layer GAT. This indicates the Lth generation after GAT update. Candidate SKU nodes The final embedded representation; This represents the final embedded representation of the consultation text node after being updated by the L-layer GAT. This represents vector concatenation; MLP stands for Multilayer Perceptron, used for nonlinear fusion.

[0075] Finally, the business process record with the highest score is selected as the target entity:

[0076] in, Represents the target entity.

[0077] For example, the system determines that "next week's annual leave" most likely points to annual leave record A: LEAVE-202508-001, and uses it as the target leave application.

[0078] This disambiguation method breaks away from the static, linear framework of traditional weighted scoring. By constructing a heterogeneous semantic graph and utilizing GAT for end-to-end reasoning, it achieves a deep understanding and accurate capture of user intent. This method not only improves disambiguation accuracy but also possesses strong interpretability and self-evolutionary capabilities.

[0079] In human resources service scenarios, the interaction between users and intelligent customer service is often not a single-round dialogue, but a continuous multi-round exchange. The user input contains pronouns such as "this" and "that," which can be ambiguous, for example, "Never mind, I'll cancel that application first."

[0080] Therefore, based on the above embodiments, after obtaining the target entity, the method further includes: Step S106E: Extract the target entity from the context of the user-input consultation text.

[0081] Step S106F: Complete the initial entities in the currently input consultation text based on the target entities in the context.

[0082] For example, in the first round of dialogue, “I want to change the time of this annual leave application,” the system identifies the target entity as annual leave record A: LEAVE-202508-001 and stores it in the context cache.

[0083] In the second round of dialogue, if the message is "Never mind, I'll cancel that application first," then the initial entity is identified as "that application." The system automatically completes the initial entity, resulting in the target entity as annual leave record A: LEAVE-202508-001. This eliminates the need for further disambiguation processing and improves entity recognition efficiency.

[0084] Based on the above embodiments, after obtaining the target intent, the method further includes: Step S108: If there are multiple target intents, generate a clarification query message based on the multiple target intents.

[0085] For example, if the current objective is to "check salary details" or "salary is incorrect", you can further ask clarification questions, such as: "Do you want to check salary details, or do you believe the amount is incorrect and need to appeal?"

[0086] Step S109: Obtain the clarification feedback result in response to the clarification inquiry message.

[0087] The user replied: "The amount is incorrect and an appeal is required."

[0088] Step S110: Determine the final target intent based on the clarification feedback results.

[0089] Based on the clarification feedback that "the amount is incorrect and an appeal is needed," the final determination of the target intent was: "to initiate a wage dispute application," and the corresponding business processing procedure was initiated.

[0090] In human resources service scenarios, employees may initiate business applications (such as "I want to cancel my annual leave" or "I want to modify my job transfer information") through intelligent customer service or self-service platforms. However, not all requests can be executed immediately. Each HR business has its prerequisites for execution. If these prerequisites are not met, employees need to provide supplementary materials, complete the approval process, or correct the status.

[0091] Based on the above embodiments, before generating a response result for the consultation text based on the target entity and business interface, the method further includes: Step S111: Determine the prerequisites for the execution of the business action corresponding to the target intent based on the target intent.

[0092] In this step, for example, the prerequisite for canceling annual leave is "approval has not yet been approved"; the prerequisite for applying for resignation is "completion of the work handover checklist".

[0093] In one example, a mapping database between business nodes and pre-execution conditions can be pre-built; the corresponding business node can be determined by the final target intent, and then the corresponding pre-execution conditions can be determined.

[0094] Step S112: Check whether the prerequisite execution conditions have been met; if there are unmet conditions, generate an aggregated follow-up instruction containing all missing conditions.

[0095] The server sequentially calls the validation interfaces corresponding to each precondition, obtains the validation results, and summarizes the results into a condition satisfaction status table. If any condition is not satisfied, an aggregated follow-up question is generated. The aggregated follow-up question combines multiple missing conditions into a single natural language question, used to retrieve multiple missing entity information at once. For example: "To submit your resignation application, you still need to complete the following: 1. Submit approval from your immediate supervisor; 2. Complete the exit interview questionnaire; 3. Return your company laptop. Do you wish to proceed now?"

[0096] This invention, through pre-verification, identifies and prompts missing items in advance, avoiding business interface call failures. It compresses what might have been 3-4 rounds of item-by-item questioning into a single round of aggregated questioning, allowing users to complete multiple information entries in a single response, improving interaction efficiency by over 60%.

[0097] Based on the above embodiments, the method further includes: Step S113: Generate an interaction log based on the user's original input, the corrected text, the candidate user intent and the confidence level of the candidate user intent, and the final target intent, and label the interaction log with log tags.

[0098] In this embodiment of the invention, after each user consultation is processed, a structured interaction log record is automatically generated and stored.

[0099] For example, the log labels are semantic error correction - success / failure; multiple intents - exist / clarified, etc.

[0100] Step S114: Determine the sample score result for each interaction log based on the confidence of multiple candidate user intents in each interaction log and the pre-built sample scoring model.

[0101] In one feasible implementation, the sample scoring model is as follows:

[0102] in, The results are the sample scoring results; Input for the user; Candidate intents predicted by the intent recognition model; For the ultimate goal and intention.

[0103] Step S115: Compare the sample scoring results with the preset sample threshold.

[0104] The higher the score, the more ambiguous the model's judgment, and the more valuable the sample is for learning.

[0105] Step S116: If the sample threshold is exceeded, the interaction log and its log tag are used as training samples to update and train the intent recognition model.

[0106] These samples are categorized into a high-value sample pool and are periodically (e.g., daily) triggered for incremental training or online fine-tuning of the intent recognition model without human intervention.

[0107] As real user input and feedback are continuously accumulated, the above process will continue to cycle. Each time the "error correction → clarification → confirmation → learning" closed loop is completed, the intent recognition model will become more intelligent.

[0108] This invention constructs a "self-learning engine" for an intelligent customer service system. By designing a scientific sample scoring model and log labeling system, the system can automatically identify high-value training samples from massive amounts of real dialogues, enabling the intent recognition model to autonomously evolve and learn.

[0109] Based on the same inventive concept, embodiments of the present invention provide an intelligent customer service question-and-answer system, applied to employee service and business processing scenarios in human resources systems, such as... Figure 3 As shown, it includes: The acquisition unit 301 is used to acquire the consultation text input by the user and the current business node in the current employee's business processing flow.

[0110] The error correction unit 302 is used to perform semantic error correction on the consultation text to obtain the corrected text.

[0111] The intent recognition unit 303 is used to perform preliminary intent recognition based on the error-corrected text using a pre-trained intent recognition model, and obtain one or more candidate user intents and the confidence level of the candidate user intents.

[0112] The filtering unit 304 is used to filter candidate user intents based on the current business node, and retain target intents that are within the execution range allowed by the current business node.

[0113] Calling unit 305 is used to call the business interface corresponding to the target intent based on the target intent.

[0114] The recognition unit 306 is used to identify target entities in the corrected text.

[0115] The generation unit 307 is used to generate a response result for the consultation text based on the target entity and the business interface.

[0116] Based on the same technical concept, embodiments of the present invention also provide an electronic device, such as... Figure 4As shown, it includes a processor 401, a communication interface 402, a memory 403, and a communication bus 404, wherein the processor 401, the communication interface 402, and the memory 403 communicate with each other through the communication bus 404.

[0117] Memory 403 is used to store computer programs; The processor 401 is used to execute the program stored in the memory 403 to implement the steps of the intelligent customer service question and answer method.

[0118] The communication bus mentioned in the above electronic devices can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus can be divided into address bus, data bus, control bus, etc. For ease of illustration, only one thick line is used to represent it in the diagram, but this does not mean that there is only one bus or one type of bus.

[0119] The communication interface is used for communication between the aforementioned electronic devices and other devices.

[0120] The memory may include random access memory (RAM) or non-volatile memory (NVM), such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0121] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0122] The computer program product for the intelligent customer service question-and-answer method provided in this embodiment of the invention includes a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the methods described in the preceding method embodiments. For specific implementation details, please refer to the method embodiments, which will not be repeated here.

[0123] The intelligent customer service Q&A system provided in this embodiment of the invention can be specific hardware on a device or software or firmware installed on the device. The system provided in this embodiment of the invention has the same implementation principle and technical effects as the aforementioned method embodiments. For the sake of brevity, any parts not mentioned in the system embodiments can be referred to the corresponding content in the aforementioned method embodiments. Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems, units, and processes described above can all be referred to the corresponding processes in the above method embodiments, and will not be repeated here.

[0124] In the embodiments provided by this invention, it should be understood that the disclosed systems and methods can be implemented in other ways. The system embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. Furthermore, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the shown or discussed mutual couplings, direct couplings, or communication connections may be through some communication interfaces. Indirect couplings or communication connections between systems or units may be electrical, mechanical, or other forms.

[0125] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0126] In addition, the functional units in the embodiments provided by the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0127] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, essentially, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0128] It should be noted that similar labels and letters in the following figures indicate similar items. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. In addition, the terms "first", "second", "third", etc. are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0129] Finally, it should be noted that the above-described embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. All should be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. An intelligent customer service question-and-answer method, characterized in that, Employee service and business processing scenarios applied to human resources systems include: Obtain the user's input consultation text and the current business node in the current employee's business processing flow; Perform semantic correction on the consultation text to obtain the corrected text; Based on the corrected text, a pre-trained intent recognition model is used to perform preliminary intent recognition, resulting in one or more candidate user intents and the confidence level of the candidate user intents. The candidate user intents are filtered based on the current business node, and target intents that are within the execution range allowed by the current business node are retained; Based on the target intent, the corresponding business interface is invoked; Identify the target entities in the corrected text; A response result is generated based on the target entity and the business interface for the consultation text.

2. The method according to claim 1, characterized in that, The semantic correction of the consultation text, resulting in the corrected text, includes: Identify the current language type of the consultation text; Based on the current language type, call the human resources domain dictionary corresponding to the language type; The spelling of words in the consultation text is checked based on the aforementioned human resources dictionary. If there are misspelled words, then the misspelled words will be matched with words in the human resources field dictionary based on their similarity. The misspelled word is replaced with the most similar match to obtain the corrected text.

3. The method according to claim 1, characterized in that, The target entities identified in the corrected text include: Extract the initial entities from the corrected text; Obtain employee business processing records within a user-preset number of days and the structured data of each record in the employee business processing records; Obtain users' historical behavior logs in the human resources system; Based on the initial entity, the structured data of each record, and the historical behavior log, the initial entity is deambiguously processed to obtain the target entity.

4. The method according to claim 3, characterized in that, After obtaining the target entity, the method further includes: Extract the target entity from the context of the user-input consultation text; The initial entities in the currently input consultation text are completed based on the target entities in the context.

5. The method according to claim 1, characterized in that, After obtaining the target intent, the method further includes: If there are multiple target intentions, a clarification query message is generated based on the multiple target intentions; Obtain the clarification feedback result in response to the clarification inquiry message; The final target intent is determined based on the clarification feedback results.

6. The method according to claim 1, characterized in that, Before generating a response result for the consultation text based on the target entity and the business interface, the method further includes: Based on the target intent, determine the prerequisites for the execution of the business action corresponding to the target intent; Check whether the prerequisite execution conditions have been met; If any conditions are not met, then an aggregated follow-up instruction containing all missing conditions is generated.

7. The method according to claim 1, characterized in that, The method further includes: An interaction log is generated based on the user's original input, the corrected text, the candidate user intent and the confidence level of the candidate user intent, and the final target intent, and the interaction log is labeled with log tags. The sample score result of each interaction log is determined based on the confidence level of the multiple candidate user intentions in each interaction log and the pre-built sample scoring model. The sample scoring results are compared with a preset sample threshold; If the sample size exceeds the threshold, the interaction log and its tag are used as training samples to update and train the intent recognition model.

8. An intelligent customer service question-and-answer system, characterized in that, Employee service and business processing scenarios applied to human resources systems include: The acquisition unit is used to acquire the consultation text input by the user and the current business node in the current employee's business processing flow; The error correction unit is used to perform semantic error correction on the consultation text to obtain the corrected text. The intent recognition unit is used to perform preliminary intent recognition based on the error-corrected text using a pre-trained intent recognition model, and obtain one or more candidate user intents and the confidence level of the candidate user intents. The filtering unit is used to filter the candidate user intents according to the current business node, and retain the target intents that are within the execution range allowed by the current business node; The invocation unit is used to invoke the business interface corresponding to the target intent based on the target intent; The recognition unit is used to identify target entities in the corrected text; The generation unit is used to generate a response result for the consultation text based on the target entity and the business interface.

9. An electronic device, characterized in that, It includes a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus; The memory is used to store computer programs; When the processor executes the program stored in the memory, it implements the method described in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method described in any one of claims 1-7.

Images