An intelligent auxiliary question and answer processing method and device, a storage medium and an electronic device

Abstract

This application discloses an intelligent assisted question-and-answer processing method, device, storage medium, and electronic device, relating to the field of intelligent question-and-answer processing technology. All functions of this solution are customized around the scenario of supporting real-time question-and-answer for outbound calls. It integrates functions such as real-time voice acquisition and filtering, speech transcription, semantic analysis and core question extraction, hybrid knowledge base retrieval, answer ranking and push, question-and-answer log recording, and model self-optimization into a customized plug-in intelligent agent. Through the intelligent agent's exclusive architecture and scenario-based adaptation, it can identify the questions raised by customers in real time, automatically retrieve the enterprise knowledge base, quickly match and push the most likely relevant target answers to customer service personnel, without the need for manual knowledge base queries, thus improving the real-time performance, accuracy, and efficiency of question-and-answer support.

Description

Technical Field

[0001] This application relates to the field of intelligent question-answering technology, and more specifically, to an intelligent assisted question-answering processing method, apparatus, storage medium, and electronic device. Background Technology

[0002] Outbound customer service calls are an important question-and-answer process where customer service personnel proactively communicate with customers to handle inquiries, resolve issues, promote products, and provide notifications. Outbound customer service calls are widely used in industries such as finance, telecommunications, e-commerce, and education and training.

[0003] During the Q&A process, customers often raise various questions (such as product functions, pricing standards, application procedures, after-sales guarantees, etc.). Customer service personnel need to respond quickly and accurately to ensure communication efficiency and customer satisfaction.

[0004] However, customer service personnel often struggle to fully memorize all business knowledge, which can lead to delayed responses, inaccurate answers, and omissions of crucial information when faced with diverse customer inquiries. Furthermore, traditional question-and-answer processes require customer service representatives to manually consult knowledge bases, which not only prolongs customer wait times but also impacts outbound call progress and overall service experience.

[0005] Therefore, how to improve the real-time performance, accuracy, and efficiency of question-and-answer support is a problem that this application urgently needs to solve. Summary of the Invention

[0006] In view of this, this application discloses an intelligent assisted question-and-answer processing method, device, storage medium, and electronic device, which aims to identify customer questions in real time through the dedicated architecture and scenario-based adaptation of the intelligent agent, automatically search the enterprise knowledge base, quickly match and push the most likely relevant target answer to customer service personnel, without the need for manual knowledge base queries, thereby improving the real-time performance, accuracy, and efficiency of question-and-answer support.

[0007] To achieve the above objectives, the disclosed technical solution is as follows:

[0008] The first aspect of this application discloses an intelligent assisted question-answering processing method, which is applied to an external intelligent agent. The external intelligent agent is an end-to-end dedicated intelligent agent composed of an outbound call scenario customization module, an external architecture, a scenario-based adaptation module, and a closed-loop self-optimization module. The method includes:

[0009] While connected to the outbound calling system, the target voice data is transcribed in real time to obtain the text information corresponding to the target voice data.

[0010] The core issue in extracting text information based on natural language processing algorithm models;

[0011] Multiple relevant answers to the core question were retrieved using a proprietary hybrid retrieval rule.

[0012] The multiple related answers are sorted according to a preset sorting method, and a preset number of target answers are selected and recommended from the sorting results.

[0013] Determine the answer matching accuracy based on user questions and a preset number of target answers;

[0014] The natural language processing algorithm model and knowledge base index are updated based on the accuracy of the answer matching to complete closed-loop self-optimization.

[0015] The second aspect of this application discloses an intelligent assisted question-answering processing device, which is applied to an external intelligent agent. The external intelligent agent is an end-to-end dedicated intelligent agent composed of an outbound call scenario customization module, an external architecture, a scenario-based adaptation module, and a closed-loop self-optimization module. The method includes:

[0016] The real-time transcription unit is used to transcribe the target voice data in real time while connected to the outbound calling system, so as to obtain the text information corresponding to the target voice data.

[0017] The extraction unit addresses the core issue of extracting text information based on natural language processing algorithm models.

[0018] The retrieval unit is used to retrieve multiple relevant answers to the core question using specific hybrid retrieval rules.

[0019] The sorting and filtering unit is used to sort the multiple related answers according to a preset sorting method, filter out a preset number of target answers from the sorting results and recommend them;

[0020] The determination unit is used to determine the answer matching accuracy based on the user's question and a preset number of target answers;

[0021] The first update unit is used to update the natural language processing algorithm model and knowledge base index according to the answer matching accuracy, so as to complete the closed-loop self-optimization.

[0022] As can be seen from the above technical solution, this application discloses an intelligent assisted question-answering processing method, device, storage medium, and electronic device. The method is applied to an external intelligent agent, which is an end-to-end dedicated intelligent agent composed of an outbound call scenario customization module, an external architecture, a scenario adaptation module, and a closed-loop self-optimization module. When the external intelligent agent is connected to the outbound call system, it performs real-time transcription of the target voice data to obtain the text information corresponding to the target voice data. Based on the natural language processing algorithm model, it extracts the core question of the text information, retrieves multiple related answers corresponding to the core question through a dedicated hybrid retrieval rule, sorts the multiple related answers according to a preset sorting method, selects a preset number of target answers from the sorting results and recommends them, determines the answer matching accuracy based on the user question and the preset number of target answers, and updates the natural language processing algorithm model and knowledge base index according to the answer matching accuracy to complete the closed-loop self-optimization.

[0023] This solution employs a customized, add-on intelligent agent tailored to outbound calling scenarios. It features an add-on architecture, scenario-specific adaptation, and closed-loop self-optimization, seamlessly integrating with existing enterprise outbound calling systems without requiring any modifications to the original system's architecture, functionality, or data storage model. The solution acquires voice data in real-time, including voice filtering that optimizes for customer verbal tics, call noise, and customer service interruptions. It employs a proprietary hybrid retrieval rule combining semantic matching and keyword search, performing semantic analysis and core question extraction for conversational and fragmented questions in outbound calling scenarios. Combined with comprehensive scoring and ranking, it accurately filters target answers to meet diverse customer service needs. Furthermore, real-time voice recognition and question extraction enable near-instantaneous response to customer inquiries, significantly reducing customer service query time and improving communication efficiency. All functions in this solution are customized around the scenario of real-time question and answer support for outbound calls. It integrates functions such as real-time voice acquisition and filtering, speech transcription, semantic analysis and core question extraction, hybrid knowledge base retrieval, answer ranking and push, question and answer log recording, and model self-optimization into a customized plug-in intelligent agent. Through the agent's exclusive architecture and scenario-based adaptation, it can identify the questions raised by customers in real time, automatically search the enterprise knowledge base, quickly match and push the most likely relevant target answers to customer service personnel, without the need for manual knowledge base queries, thus improving the real-time performance, accuracy, and efficiency of question and answer support. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0025] Figure 1 This is a framework diagram of an intelligent assisted question-answering processing system disclosed in an embodiment of this application;

[0026] Figure 2 This is a flowchart illustrating an intelligent assisted question-answering processing method disclosed in an embodiment of this application;

[0027] Figure 3 This is a schematic diagram of the structure of an intelligent auxiliary question-answering processing device disclosed in an embodiment of this application;

[0028] Figure 4 This is a schematic diagram of the structure of the electronic device disclosed in the embodiments of this application. Detailed Implementation

[0029] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0030] In this application, the terms "comprising," "including," or any other variations thereof are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0031] As the background technology indicates, in existing question-and-answer processing methods, customer service personnel often struggle to fully memorize all business knowledge. When faced with diverse customer questions, this can lead to delayed responses, inaccurate answers, and omissions of crucial information. Furthermore, traditional question-and-answer processing requires customer service representatives to manually consult knowledge bases, which not only prolongs customer waiting time but also impacts outbound call progress and service experience.

[0032] To address the aforementioned issues, this application discloses an intelligent assisted question-and-answer processing method, device, storage medium, and electronic device. This solution employs a dedicated plug-in intelligent agent tailored to outbound calling scenarios, featuring a plug-in architecture, scenario-based adaptation, and closed-loop self-optimization. It seamlessly integrates with existing enterprise outbound calling systems without requiring any modifications to the original system's architecture, functionality, or data storage mode. This solution acquires voice data in real-time through the plug-in intelligent agent. For example, voice filtering optimizes voice data specific to outbound calling scenarios, addressing customer verbal tics, call noise, and customer service interruptions. It designs a dedicated hybrid retrieval rule combining semantic matching and keyword search. Semantic analysis and core question extraction are performed on colloquial and fragmented questions in outbound calling scenarios. Combined with comprehensive scoring and ranking, target answers are accurately selected to meet diverse customer service needs. Furthermore, real-time voice recognition and question extraction enable second-level responses to customer questions, significantly reducing customer service query time and improving communication efficiency. All functions in this solution are customized around the scenario of real-time question and answer support for outbound calls. It integrates functions such as real-time voice acquisition and filtering, speech-to-text transcription, semantic analysis and core question extraction, hybrid knowledge base retrieval, answer ranking and push, question and answer logging, and model self-optimization into a customized plug-in intelligent agent. Through the agent's dedicated architecture and scenario-based adaptation, it can identify customer questions in real time, automatically search the enterprise knowledge base, quickly match and push the most likely relevant target answer to customer service personnel, eliminating the need for manual knowledge base queries and improving the real-time performance, accuracy, and efficiency of question and answer support. The specific implementation method is illustrated in the following examples.

[0033] It should be noted that the intelligent assisted question-answering processing method, device, storage medium and electronic device provided in this application can be used in the technical field of intelligent question-answering processing, etc. The above is only an example and does not limit the application field of the intelligent assisted question-answering processing method, device, storage medium and electronic device provided in this application.

[0034] refer to Figure 1 The diagram shown is a framework diagram of an intelligent assisted question-answering processing system disclosed in an embodiment of this application. The intelligent assisted question-answering processing system includes an external intelligent agent (independent server), an existing outbound call system (existing enterprise server), an enterprise knowledge base (database server), and a customer service terminal.

[0035] The specific functions of the plug-in intelligent agent, existing outbound call system, enterprise knowledge base, and customer service terminal are as follows:

[0036] External intelligent agents:

[0037] The plug-in intelligent agent is developed based on Python. An intelligent agent is a digital entity with autonomous perception, decision-making, and execution capabilities. It can simulate the interaction logic and business processing flow of humans in specific scenarios, autonomously completing a series of operations such as information retrieval, communication interaction, and task advancement. The plug-in intelligent agent integrates a speech recognition engine and a Natural Language Processing (NLP) algorithm model. It includes submodules for speech acquisition, speech transcription, question extraction, knowledge base retrieval, and answer push.

[0038] The architecture independence of the plug-in intelligent agent: It adopts a pure plug-in design, which integrates with the enterprise's existing outbound calling system in a non-intrusive manner. It only obtains voice data through the audio acquisition interface, without requiring any modification to the original system's architecture, functions, or data storage mode.

[0039] The plug-in intelligent agent integrates functions such as real-time voice acquisition and filtering, voice transcription, semantic analysis and core question extraction, knowledge base hybrid retrieval, answer sorting and push, question and answer log recording, and model self-optimization. It is an end-to-end dedicated intelligent agent for customer service outbound call question and answer support scenarios.

[0040] Strong scenario adaptability of plug-in intelligent agents: All functions are customized around the scenario of "real-time Q&A support for customer service outbound calls". For example, voice filtering is optimized for customer verbal tics, call noise and customer service interruptions in outbound call scenarios, and semantic analysis extracts core questions for conversational questions and fragmented questions in outbound call scenarios.

[0041] Personalized human-machine collaboration with plug-in intelligent agents: It can assist human customer service and provide answer references for customer service representatives. Customer service representatives can choose / modify the answers independently, and the target answers pushed are adapted to the different communication styles of customer service representatives and the personalized needs of customers.

[0042] The functions and implementation details of each submodule are as follows:

[0043] (1) Voice Acquisition Submodule: By calling the audio acquisition interface of the existing outbound calling system, the two-way voice stream data between customer service and customer is acquired in real time. The two-way voice stream data is pushed to the voice transcription submodule in a streaming mode with a transmission delay of no more than 100ms to ensure the real-time performance of voice processing. At the same time, voice acquisition is dynamically started and stopped according to the call status synchronized with the outbound calling system (started during the call and stopped when the call ends).

[0044] (2) Speech-to-text submodule: Receives bidirectional speech stream data transmitted from the speech acquisition submodule, calls the speech recognition engine to perform real-time transcription of the bidirectional speech stream data, converting it into text information (transcription accuracy not less than 96%). Simultaneously, a built-in speech filtering algorithm removes noise, customer service interruptions, customer filler words (such as "um," "ah," "this"), repetitive sentences, and other invalid speech data, retaining the text content corresponding to the core speech. For example, if a customer says, "Um... I want to ask, how is the penalty interest calculated if a credit card payment is overdue?", the filtered text becomes "How is the penalty interest calculated for overdue credit card payments?".

[0045] (3) Problem extraction submodule: Based on the NLP algorithm model, semantic analysis is performed on the filtered target speech data.

[0046] First, the sentence structure (such as subject, predicate, and object) is identified through part-of-speech tagging and syntactic analysis.

[0047] Then, an intent recognition algorithm is used to extract the customer's core questions and generate standardized question descriptions (such as extracting "How is the penalty interest for overdue credit card payments calculated?" as the core question "How is the penalty interest for overdue credit card payments calculated?").

[0048] Finally, add keyword tags (such as "credit card", "overdue payment", "penalty interest", "calculation method") to the core questions for subsequent knowledge base retrieval.

[0049] (4) Knowledge base retrieval submodule: Receives the core questions and keyword tags output by the question extraction submodule, and adopts a unique hybrid retrieval strategy of "semantic matching + keyword retrieval".

[0050] The first step is to use the open-source search and data analysis engine (Elasticsearch) to perform keyword retrieval and match knowledge data in the knowledge base that contain the same keyword tags;

[0051] The second step is to calculate the semantic similarity between the core question and the "problem description" field in the retrieved knowledge data, based on a pre-trained semantic similarity model (such as the BERT model).

[0052] The third step involves combining keyword matching and semantic similarity to comprehensively score the search results (out of 100 points), sorting them from highest to lowest score, and selecting the standard answers corresponding to the highest-scoring knowledge data from a predetermined number (e.g., the top 10). The predetermined number is set according to actual circumstances, and this application does not impose a specific limit.

[0053] (5) Answer Push Submodule: The Top 10 target answers after filtering are organized according to the scoring order and pushed to the display interface of the customer service terminal through an independent communication interface. At the same time, the push time, answer content, corresponding core questions and other information are recorded to provide data support for subsequent Q&A log statistics.

[0054] (6) Log recording and model optimization submodule: After the outbound call is completed, a complete question and answer log is recorded (including but not limited to the customer's original voice, transcribed text, core questions, the top 10 answers pushed, the answers selected by customer service, and subsequent customer feedback). The answer matching accuracy (the proportion of selected answers in the top 10 push results) and customer satisfaction data are regularly statistically analyzed. For cases with large semantic matching deviations (such as incorrect extraction of core questions or retrieval results that are not related to the questions), they are automatically marked and used for incremental training of the NLP algorithm model to continuously improve the accuracy of question extraction and retrieval matching.

[0055] Existing outbound calling system:

[0056] The core functions and implementation details of the existing outbound calling system are as follows:

[0057] (1) Basic communication: Complete the establishment and transmission of voice calls between customer service and customers, ensure the normal operation of outbound communication, and support multiple call methods such as landline and mobile phone.

[0058] (2) Voice data interface: Provides a standardized audio acquisition interface (including but not limited to supporting Real-Time Messaging Protocol (RTMP), Hypertext Transfer Protocol (HTTP) and other protocols), without modifying the core functions of the original system. Only the voice data reading permission is opened, allowing the plug-in intelligent agent to obtain the call voice stream data in real time, ensuring that the original outbound call business is not affected.

[0059] (3) Synchronize call status: Synchronize call status (such as in call, call ended, call interrupted) with the external intelligent agent to facilitate the external intelligent agent to start or stop the voice processing process.

[0060] Enterprise Knowledge Base:

[0061] An enterprise knowledge base is a structured database that stores various business-related knowledge, such as product information, service rules, frequently asked questions, and business processes. It supports query methods such as keyword search and semantic matching. The technical architecture of the enterprise knowledge base includes a hybrid storage architecture using MySQL and Elasticsearch. MySQL is used to store structured knowledge data (such as knowledge ID, knowledge category, question description, standard answer, and update time); Elasticsearch is used to achieve efficient full-text search and semantic matching.

[0062] The core functions and implementation details of the enterprise knowledge base are as follows:

[0063] (1) Knowledge entry and update: It supports managers to enter new business knowledge (such as new product rules and new business processes) through the back-end management interface, and can also import knowledge data in Excel format in batches; it supports operations such as modifying, deleting and removing knowledge content to ensure the timeliness and accuracy of knowledge base data.

[0064] (2) Knowledge structuring: The entered knowledge data is automatically structuring, extracting keywords (such as "credit card", "overdue payment", "penalty interest" etc.) and semantic tags (such as "billing rules", "repayment related" etc.) to build a knowledge index, laying the foundation for rapid retrieval.

[0065] (3) Search interface provision: Provides a search API interface to support keyword search and semantic matching requests sent by plug-in intelligent agents, and returns relevant knowledge data and matching score.

[0066] Customer service terminal:

[0067] Technical architecture of customer service terminal: Supports personal computer (PC) terminals (such as Windows / Mac systems) and mobile terminals (such as Android / iOS systems). The PC terminal is a standalone desktop application (which can be developed based on the open-source Electron framework), and the mobile terminal is an application (APP) plugin that can run in parallel with existing customer service terminal software.

[0068] The core functions and implementation details of the customer service terminal are as follows:

[0069] (1) Answer Display: The target number of answers pushed by the plug-in intelligent agent (e.g., 10 answers) are displayed in a separate interface in order of score from high to low or from low to high. Each answer is marked with a matching score (e.g., "98 points", "95 points", etc.) to facilitate customer service to quickly identify the best answer. At the same time, the function of copying and editing the answer content is supported. Customer service can directly copy the answer and send it to the customer, or modify it according to the actual communication scenario before responding.

[0070] (2) Feedback submission: Customer service can mark the target number of pushes with feedback (such as "accurate", "inaccurate", "irrelevant" etc.). The feedback information is synchronized to the plug-in intelligent agent in real time for model optimization and knowledge base update.

[0071] (3) Historical record query: Supports customer service to query the Q&A records of the current call or the historical call, including customer questions, pushed answers, and answers selected by themselves, which facilitates subsequent business review and problem tracing.

[0072] The complete business process steps of the intelligent assisted question-answering system framework are as follows:

[0073] System deployment and integration: The plug-in intelligent agent is deployed on a separate server and establishes a communication connection with the enterprise's existing outbound calling system and enterprise knowledge base through a non-intrusive interface; customer service personnel install the corresponding client application or plugin on their terminals to complete account login and permission verification.

[0074] Outbound call initiation and voice acquisition: Customer service representatives dial customer numbers through the existing outbound call system. After the call is established, the existing outbound call system synchronizes the "calling in progress" status to the external intelligent agent, and the voice acquisition submodule is activated to acquire two-way voice stream data between customer service representatives and customers in real time.

[0075] Speech transcription and question extraction: The speech transcription submodule transcribes the customer's speech into text in real time and filters out invalid information. The question extraction submodule performs semantic analysis on the text information to extract the core questions and keyword tags.

[0076] Knowledge base retrieval and answer selection: The knowledge base retrieval submodule performs a mixed search based on core questions and keyword tags, and selects the top 10 target answers by comprehensive score.

[0077] Answer push and customer service selection: The answer push submodule pushes 10 target answers to the customer service terminal in real time. Customer service representatives can view the answer list and select, modify or refer to relevant answers to respond to customers based on their questions.

[0078] Call End and Log Recording: After the customer service representative ends the call with the customer, the existing outbound calling system will synchronize the "call ended" status to the external intelligent agent, and the voice processing flow will stop; the log recording and model optimization submodule generates question and answer logs and statistically analyzes relevant data.

[0079] System optimization: The plug-in intelligent agent regularly updates the NLP algorithm model and knowledge base index based on question-and-answer logs and customer service feedback to improve system performance.

[0080] The intelligent agent in this solution is not a generic AI module / robot found in existing technologies, but rather a dedicated intelligent agent customized for customer service outbound call scenarios. It possesses core characteristics such as an add-on architecture, scenario-based adaptation, and closed-loop self-optimization. The add-on intelligent agent eliminates the need to modify the existing outbound call system, reducing deployment costs and implementation risks. Real-time speech-to-text transcription and question extraction enable rapid identification of customer questions. Semantic matching combined with keyword retrieval improves the accuracy of knowledge base searches, and targeted answer pushes meet the diverse selection needs of customer service personnel. Question-and-answer logs provide data support for system optimization, continuously improving answer matching accuracy, ultimately achieving a dual improvement in the efficiency and accuracy of customer service outbound call question-and-answer processes.

[0081] This application provides a plug-in intelligent agent that can be quickly deployed without modifying the existing outbound calling system. By identifying customer questions in real time, automatically searching the knowledge base and pushing relevant answers, it helps customer service personnel respond to customers quickly and accurately, solving the pain points of low response efficiency, insufficient accuracy and high system modification costs in existing solutions, while reducing the workload and training costs of customer service personnel.

[0082] In this embodiment, the solution employs a dedicated plug-in intelligent agent customized for outbound calling scenarios, featuring an external architecture, scenario-based adaptation, and closed-loop self-optimization. It seamlessly integrates with the enterprise's existing outbound calling system without requiring any modification to the original system's architecture, functionality, or data storage mode. This solution acquires voice data in real-time through the plug-in intelligent agent. For example, voice filtering is specifically optimized for outbound calling scenarios, addressing customer verbal tics, call noise, and customer service interruptions. It designs a dedicated hybrid retrieval rule combining semantic matching and keyword search. Semantic analysis and core question extraction are performed on conversational and fragmented questions in outbound calling scenarios. Combined with comprehensive scoring and ranking, target answers are accurately selected to meet the diverse selection needs of customer service representatives. Furthermore, real-time voice recognition and question extraction enable second-level response to customer inquiries, significantly reducing customer service query time and improving communication efficiency. All functions in this solution are customized around the scenario of real-time question and answer support for outbound calls. It integrates functions such as real-time voice acquisition and filtering, speech transcription, semantic analysis and core question extraction, hybrid knowledge base retrieval, answer ranking and push, question and answer log recording, and model self-optimization into a customized plug-in intelligent agent. Through the agent's exclusive architecture and scenario-based adaptation, it can identify the questions raised by customers in real time, automatically search the enterprise knowledge base, quickly match and push the most likely relevant target answers to customer service personnel, without the need for manual knowledge base queries, thus improving the real-time performance, accuracy, and efficiency of question and answer support.

[0083] refer to Figure 2 The image shows an intelligent assisted question-answering processing method disclosed in an embodiment of this application, which is applied to the above embodiment. Figure 1The disclosed intelligent auxiliary question-answering processing system includes an external intelligent agent, which is an end-to-end dedicated intelligent agent composed of an outbound call scenario customization module, an external architecture, a scenario-based adaptation module, and a closed-loop self-optimization module. The method mainly includes the following steps:

[0084] S201: When the external intelligent agent is connected to the outbound calling system, it performs real-time transcription of the target voice data to obtain the text information corresponding to the target voice data.

[0085] The plug-in intelligent agent establishes a connection with the outbound calling system, and acquires outbound voice data of customer service and customers in real time through the audio acquisition interface. At the same time, it establishes data communication with the enterprise knowledge base to ensure that the knowledge base data can be retrieved in real time.

[0086] In S201, the plug-in intelligent agent collects the user's two-way voice stream data in real time, filters invalid voice data from the two-way voice stream data to obtain the filtered target voice data, and then transcribes the filtered voice data in real time to obtain the text information corresponding to the target voice data.

[0087] The plug-in intelligent agent performs real-time transcription of the collected two-way voice stream data, converting the customer's speech into text information and filtering out invalid speech (such as noise, filler words, and repetitive sentences). Specifically, the plug-in intelligent agent collects real-time two-way voice stream data between customer service representatives and customers, and can call a speech recognition engine to transcribe the two-way voice stream data in real time, converting it into text information (transcription accuracy is no less than 96%). At the same time, a built-in speech filtering algorithm removes invalid speech data such as noise, customer service interruptions, customer filler words (such as "um," "ah," "this"), and repetitive sentences, retaining the text content corresponding to the core speech. For example, if a customer says, "Um... I want to ask, how is the penalty interest calculated if a credit card payment is overdue?", the filtered text will be "How is the penalty interest calculated for overdue credit card payments?"

[0088] Real-time transcription and invalid speech filtering are built-in functions of the plug-in intelligent agent. The plug-in intelligent agent has made exclusive filtering rules for outbound call scenarios (such as accurately filtering outbound call slang such as "um," "ah," and "this," invalid speech such as customer service interruptions, and retaining the core content of fragmented customer questions). Transcription and filtering are completed autonomously and in real time by the intelligent agent without human / main system intervention. The transcription latency is controlled within 100ms and the filtering accuracy rate is over 98%, achieving the goal of real-time and accurate speech processing in outbound call scenarios.

[0089] S202: The core problem of extracting text information based on NLP algorithm models for plug-in intelligent agents.

[0090] In S202, the plug-in intelligent agent performs semantic analysis on text information based on the NLP algorithm model, obtains semantic analysis results, interprets the meaning and intent of the semantic analysis results, and extracts the core issues of the text information.

[0091] The process of analyzing the meaning and intent of the semantic analysis results and extracting the core information from the text is as follows:

[0092] First, the sentence structure (such as subject, predicate, and object) is identified through part-of-speech tagging and syntactic analysis.

[0093] Then, an intent recognition algorithm is used to extract the customer's core questions and generate standardized question descriptions (such as extracting "How is the penalty interest for overdue credit card payments calculated?" as the core question "How is the penalty interest for overdue credit card payments calculated?").

[0094] Finally, add keyword tags (such as "credit card", "overdue payment", "penalty interest", "calculation method") to the core questions for subsequent knowledge base retrieval.

[0095] NLP semantic analysis is the core scenario-based function of the plug-in intelligent agent. The plug-in intelligent agent, tailored to the conversational, fragmented, and non-standardized questioning characteristics of customers in outbound calling scenarios, has developed a core question extraction algorithm (such as using syntactic analysis to identify the "core appeal words + business keywords" of customer questions, ignoring irrelevant conversational modifiers). For example, it accurately extracts the customer's question, "I want to ask, how is the penalty interest calculated if a credit card is overdue?" into "How is the penalty interest calculated for overdue credit card payments?" with an accuracy rate exceeding 95%. This achievement does not solely rely on NLP technology but rather on the deep adaptation of the plug-in intelligent agent to the customer service outbound calling scenario, a direct result of the agent's customized design.

[0096] S203: The plug-in intelligent agent retrieves multiple relevant answers to the core question through exclusive hybrid retrieval rules.

[0097] The plug-in intelligent agent, based on the extracted core question, uses a combination of semantic matching and keyword retrieval to search for relevant answers in the knowledge base, sorts them in order of matching degree from high to low or low to high, and selects the most likely relevant preset number (e.g., 10) of answers.

[0098] In S203, the plug-in intelligent agent uses a proprietary hybrid retrieval rule consisting of semantic matching and keyword retrieval to obtain the semantic similarity of the user's question and the weight of the business keywords corresponding to the core question. Based on the semantic similarity and the weight of the business keywords, it retrieves multiple related answers corresponding to the core question from the knowledge base.

[0099] The system receives the core questions and keyword tags output by the question extraction submodule and employs a proprietary hybrid retrieval strategy combining semantic matching and keyword search. The specific process is as follows:

[0100] First, keyword retrieval is performed using the open-source search and data analysis engine (Elasticsearch) to match knowledge data in the knowledge base that contain the same keyword tags;

[0101] The second step is to calculate the semantic similarity between the core question and the "problem description" field in the retrieved knowledge data, based on a pre-trained semantic similarity model (such as the BERT model).

[0102] The third step is to combine keyword matching and semantic similarity to give a comprehensive score to the search results (out of 100 points), sort them from high to low scores, and select the standard answers corresponding to the knowledge data with the highest scores in a preset number (such as the top 10).

[0103] It should be noted that the preset quantity is set according to the actual situation, and this application does not impose a specific limit.

[0104] S204: The plug-in intelligent agent sorts multiple related answers according to a preset sorting method, selects a preset number of target answers from the sorting results and recommends them.

[0105] The specific execution process of S204 is shown in A1-A3.

[0106] A1: The plug-in intelligent agent searches for multiple related answers in the knowledge base and obtains the matching degree of multiple related answers.

[0107] A2: The plug-in intelligent agent sorts the matching degree of multiple related answers according to the matching degree ranking method, and obtains the ranking result.

[0108] The matching degree ranking method is either based on the matching degree of multiple related answers in descending order, or based on the matching degree of multiple related answers in descending order.

[0109] A3: The plug-in intelligent agent combines answer suggestions to filter out a preset number of target answers from the sorted results and recommends them through an independent display interface.

[0110] In A3, 10 answers pushed by the plug-in AI agent are displayed in a separate interface, ordered by rating from high to low or low to high. Each answer is marked with a matching score (such as "98 points", "95 points", etc.), making it easier for customer service to quickly identify the best answer. At the same time, it supports copying and editing the answer content. Customer service can directly copy the answer and send it to the customer, or modify it according to the actual communication scenario before responding.

[0111] Semantic matching and keyword retrieval constitute a proprietary hybrid retrieval rule, along with the ranking and push of multiple related answers. This ensures the accuracy of the answers while providing customer service representatives with diverse options to suit different communication styles and personalized customer needs (e.g., some customer service representatives prefer concise answers, while others prefer detailed explanations). This design is a dedicated implementation of the "human-machine collaborative assistance" positioning of the plug-in intelligent agent.

[0112] S205: The plug-in intelligent agent determines the accuracy of answer matching based on the user's question and a preset number of target answers.

[0113] In S205, the plug-in intelligent agent generates a question-and-answer log based on the user's question and a preset number of target answers, and obtains the answer matching accuracy corresponding to the question-and-answer log.

[0114] After the outbound call ends, the plug-in intelligent agent records the user's question, the 10 target answers pushed, and the answer finally selected by the customer service representative, forming a question and answer log. At the same time, it calculates the answer matching accuracy, providing data support for knowledge base optimization and intelligent agent algorithm iteration.

[0115] Regularly analyze the accuracy of answer matching in the Q&A log, which is the percentage of selected answers in the preset number (e.g., Top 10) of push results.

[0116] S206: The plug-in intelligent agent updates the NLP algorithm model and knowledge base index based on the answer matching accuracy to complete closed-loop self-optimization.

[0117] Feedback tags (such as "accurate", "inaccurate", "irrelevant" etc.) are assigned to the target number of pushed responses. The feedback information is synchronized to the plug-in intelligent agent in real time for model optimization and knowledge base updates.

[0118] For answers with matching accuracy rates lower than the preset accuracy rate, the plug-in agent pushes the cases corresponding to the answers with matching accuracy rates lower than the preset accuracy rate to the algorithm optimization module. The algorithm optimization module then performs incremental training on the semantic analysis model and the retrieval matching model, marks the answers with matching accuracy rates lower than the preset accuracy rate as invalid answers, and updates the knowledge base with invalid answers.

[0119] The preset accuracy rate should be set according to the actual situation; this application does not impose specific limitations.

[0120] The plug-in intelligent agent autonomously records the entire data chain from "customer question - pushed answer - customer service selected answer," and independently calculates the matching accuracy. For cases with incorrect matching (such as incorrect extraction of the core question or irrelevant answer push), the agent automatically pushes the case to the built-in algorithm optimization module for incremental training of semantic analysis and retrieval matching models. Simultaneously, it feeds back "invalid answers" marked by customer service to the knowledge base, enabling autonomous updates to the knowledge base. This closed-loop self-optimization function is a built-in, exclusive feature of the plug-in intelligent agent, completed autonomously without human intervention, achieving self-iterative upgrades for the plug-in intelligent agent.

[0121] This application designs and implements a dedicated plug-in intelligent agent for customer service outbound call human assistance scenarios. This agent features an external architecture, scenario-based adaptation, integrated functionality, and closed-loop self-optimization. Applied to customer service outbound call Q&A support scenarios, it achieves technical effects that are "real-time, accurate, low-cost to modify, and highly efficient in human-machine collaboration," which are unattainable with existing technologies. The core innovations are as follows:

[0122] 1. Pioneering "plug-in intelligent agent" architecture for customer service outbound calling scenarios: The customized intelligent agent is designed as a pure plug-in, non-intrusively connecting to the enterprise's existing outbound calling system. This solves the pain points of existing technologies where AI modules require modification of the main system, resulting in high modification costs and long implementation cycles. It achieves the technical effect of "zero modification, rapid deployment, and no impact on existing outbound calling business". This architecture design is unique in existing technologies.

[0123] 2. Customized development of dedicated AI agent functions for customer service outbound calling scenarios: Focusing on the scenario of "real-time Q&A assistance from human customer service", we have customized outbound calling-specific voice filtering rules, core question extraction algorithms, hybrid retrieval rules, and Top 10 answer push mode for the AI ​​agent. This solves the pain points of existing general AI modules being unable to adapt to the conversational and fragmented questions in outbound calling scenarios, and existing intelligent robots failing to meet the human-machine collaboration needs in "replacing human agents". It achieves "accurate and efficient Q&A assistance in outbound calling scenarios". This scenario-based customization function is not available in existing technologies.

[0124] 3. The design realizes the "end-to-end integration + closed-loop self-optimization" of the plug-in intelligent agent: It integrates functions such as speech processing, semantic analysis, knowledge base retrieval, answer push, log recording, and model optimization into the intelligent agent, realizing the end-to-end integrated operation of the intelligent agent in "autonomous speech reception - autonomous processing and analysis - autonomous answer push - autonomous optimization and iteration", and forming a closed-loop self-optimization system. It solves the pain points of existing technologies, such as the dispersion of functional modules, the need for manual / main system coordination, and the inability to iterate autonomously. It realizes the self-upgrading of the plug-in intelligent agent and the continuous improvement of the auxiliary effect.

[0125] Through the exclusive design of the plug-in intelligent agent, a highly efficient human-machine collaboration is achieved, in which the plug-in intelligent agent provides real-time answer references and human customer service can independently select / modify answers. This leverages the advantages of the plug-in intelligent agent in terms of "speed, accuracy, and fatigue-free operation" while retaining the advantages of human customer service in terms of "personalized and emotional" communication. It solves the dual pain points of "low efficiency of pure human and poor experience of pure intelligent" in existing technologies. The core of the implementation of this human-machine collaboration mode relies on the plug-in intelligent agent of this solution.

[0126] In this embodiment, a dedicated plug-in intelligent agent, customized for outbound calling scenarios, possesses core characteristics such as an external architecture, scenario-based adaptation, and closed-loop self-optimization. This agent seamlessly integrates with the enterprise's existing outbound calling system without requiring any modification to the original system's architecture, functionality, or data storage mode. This solution acquires voice data in real-time through the plug-in intelligent agent. For example, voice filtering is specifically optimized for outbound calling scenarios, addressing customer verbal tics, call noise, and customer service interruptions. Dedicated hybrid retrieval rules combining semantic matching and keyword search are designed. Semantic analysis and core question extraction are performed on conversational and fragmented questions in outbound calling scenarios. Combined with comprehensive scoring and ranking, target answers are accurately selected to meet the diverse selection needs of customer service representatives. Furthermore, real-time voice recognition and question extraction enable sub-second responses to customer inquiries, significantly reducing customer service query time and improving communication efficiency. All functions in this solution are customized around the scenario of real-time question and answer support for outbound calls. It integrates functions such as real-time voice acquisition and filtering, speech transcription, semantic analysis and core question extraction, hybrid knowledge base retrieval, answer ranking and push, question and answer log recording, and model self-optimization into a customized plug-in intelligent agent. Through the agent's exclusive architecture and scenario-based adaptation, it can identify the questions raised by customers in real time, automatically search the enterprise knowledge base, quickly match and push the most likely relevant target answers to customer service personnel, without the need for manual knowledge base queries, thus improving the real-time performance, accuracy, and efficiency of question and answer support.

[0127] Based on the above embodiments Figure 2 The present application discloses an intelligent assisted question-answering processing method and, in its embodiments, also discloses an intelligent assisted question-answering processing device, such as... Figure 3 As shown, this intelligent auxiliary question-answering processing device is applied to an external intelligent agent. The external intelligent agent is an end-to-end dedicated intelligent agent composed of an outbound call scenario customization module, an external architecture, a scenario-based adaptation module, and a closed-loop self-optimization module. The intelligent auxiliary question-answering processing device includes:

[0128] The real-time transcription unit 301 is used to transcribe the target voice data in real time while connected to the outbound calling system, so as to obtain the text information corresponding to the target voice data.

[0129] Extraction unit 302 is used to address the core issue of extracting text information based on natural language processing algorithm models;

[0130] Retrieval unit 303 is used to retrieve multiple relevant answers to the core question using a dedicated hybrid retrieval rule;

[0131] The sorting and filtering unit 304 is used to sort multiple related answers according to a preset sorting method, filter out a preset number of target answers from the sorting results and recommend them.

[0132] Unit 305 is used to determine the answer matching accuracy based on the user question and a preset number of target answers;

[0133] The first update unit 306 is used as the first update unit to update the natural language processing algorithm model and knowledge base index according to the answer matching accuracy, so as to complete the closed-loop self-optimization.

[0134] Furthermore, the real-time transcription unit 301 includes:

[0135] The real-time acquisition module is used to acquire the user's two-way voice stream data in real time, and to filter out invalid voice data from the two-way voice stream data to obtain the filtered target voice data.

[0136] The real-time transcription module is used to transcribe the filtered speech data in real time to obtain the text information corresponding to the target speech data.

[0137] Furthermore, the extraction unit 302 includes:

[0138] The semantic analysis module is used to perform semantic analysis on text information based on natural language processing algorithm models to obtain semantic analysis results.

[0139] The parsing module is used to analyze the meaning and intent of the semantic analysis results and extract the core issues of the textual information.

[0140] Furthermore, the retrieval unit 303 includes:

[0141] The first acquisition module is used to acquire the semantic similarity of user questions and the weight of business keywords corresponding to core questions by using exclusive hybrid retrieval rules consisting of semantic matching and keyword retrieval.

[0142] The first retrieval module is used to retrieve multiple relevant answers to the core question from the knowledge base based on semantic similarity and business keyword weights.

[0143] Furthermore, the sorting and filtering unit 304 includes:

[0144] The second retrieval module is used to search for multiple related answers in the knowledge base and obtain the matching degree of multiple related answers;

[0145] The sorting module is used to sort the matching degree of multiple related answers according to the matching degree sorting method, and obtain the sorting result;

[0146] The filtering module is used to filter out a preset number of target answers from the sorted results based on the answer suggestions, and then recommend them through a separate display interface.

[0147] Furthermore, the defining unit 305 includes:

[0148] The generation module is used to generate question-and-answer logs based on user questions and a preset number of target answers;

[0149] The second acquisition module is used to obtain the answer matching accuracy rate corresponding to the question and answer log.

[0150] Furthermore, the intelligent assisted question-answering processing device also includes:

[0151] The training labeling unit is used to push cases corresponding to answer matching accuracy rates below the preset accuracy rate to the algorithm optimization module. The algorithm optimization module performs incremental training on the semantic analysis model and the retrieval matching model, and marks the answers corresponding to answer matching accuracy rates below the preset accuracy rate as invalid answers.

[0152] The second update unit is used to update the knowledge base with invalid answers.

[0153] In this embodiment, a dedicated plug-in intelligent agent, customized for outbound calling scenarios, possesses core characteristics such as an external architecture, scenario-based adaptation, and closed-loop self-optimization. This agent seamlessly integrates with the enterprise's existing outbound calling system without requiring any modification to the original system's architecture, functionality, or data storage mode. This solution acquires voice data in real-time through the plug-in intelligent agent. For example, voice filtering is specifically optimized for outbound calling scenarios, addressing customer verbal tics, call noise, and customer service interruptions. Dedicated hybrid retrieval rules combining semantic matching and keyword search are designed. Semantic analysis and core question extraction are performed on conversational and fragmented questions in outbound calling scenarios. Combined with comprehensive scoring and ranking, target answers are accurately selected to meet the diverse selection needs of customer service representatives. Furthermore, real-time voice recognition and question extraction enable sub-second responses to customer inquiries, significantly reducing customer service query time and improving communication efficiency. All functions in this solution are customized around the scenario of real-time question and answer support for outbound calls. It integrates functions such as real-time voice acquisition and filtering, speech transcription, semantic analysis and core question extraction, hybrid knowledge base retrieval, answer ranking and push, question and answer log recording, and model self-optimization into a customized plug-in intelligent agent. Through the agent's exclusive architecture and scenario-based adaptation, it can identify the questions raised by customers in real time, automatically search the enterprise knowledge base, quickly match and push the most likely relevant target answers to customer service personnel, without the need for manual knowledge base queries, thus improving the real-time performance, accuracy, and efficiency of question and answer support.

[0154] This application embodiment also provides a storage medium, the storage medium including stored instructions, wherein, when the instructions are executed, the device where the storage medium is located is controlled to perform the intelligent auxiliary question-answering processing method described above.

[0155] This application also provides an electronic device, the structural schematic diagram of which is shown below. Figure 4 As shown, it specifically includes a memory 401 and one or more instructions 402, wherein one or more instructions 402 are stored in the memory 401 and configured to be executed by one or more processors 403 to perform the above-mentioned intelligent auxiliary question-answering processing method.

[0156] For the foregoing method embodiments, in order to simplify the description, they are all described as a series of actions. However, those skilled in the art should understand that this application is not limited to the described order of actions, because according to this application, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily essential to this application.

[0157] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For apparatus embodiments, since they are basically similar to method embodiments, the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0158] The steps in the methods of the various embodiments of this application can be adjusted, combined, or deleted according to actual needs.

[0159] Finally, it should be noted that in this paper, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations.

[0160] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

[0161] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. An intelligent assisted question-answering processing method, characterized in that, The method is applied to an add-on intelligent agent, which is an end-to-end dedicated intelligent agent composed of an outbound call scenario customization module, an add-on architecture, a scenario-based adaptation module, and a closed-loop self-optimization module. The method includes: While connected to the outbound calling system, the target voice data is transcribed in real time to obtain the text information corresponding to the target voice data. The core issue in extracting text information based on natural language processing algorithm models; Multiple relevant answers to the core question were retrieved using a proprietary hybrid retrieval rule. The multiple related answers are sorted according to a preset sorting method, and a preset number of target answers are selected and recommended from the sorting results. Determine the answer matching accuracy based on user questions and a preset number of target answers; The natural language processing algorithm model and knowledge base index are updated based on the accuracy of the answer matching to complete closed-loop self-optimization.

2. The method according to claim 1, characterized in that, While connected to the outbound calling system, the process of real-time transcription of the target voice data to obtain the corresponding text information includes: Real-time acquisition of the user's two-way voice stream data, and invalid voice filtering of the two-way voice stream data to obtain the filtered target voice data; The filtered speech data is transcribed in real time to obtain the text information corresponding to the target speech data.

3. The method according to claim 1, characterized in that, The core issues in extracting text information based on natural language processing algorithm models include: The text information is semantically analyzed based on a natural language processing algorithm model to obtain semantic analysis results. The meaning and intent of the semantic analysis results are analyzed to extract the core issues of the textual information.

4. The method according to claim 1, characterized in that, The process of retrieving multiple relevant answers to the core question using a specific hybrid retrieval rule includes: A proprietary hybrid retrieval rule consisting of semantic matching and keyword retrieval is used to obtain the semantic similarity of user questions and the weight of business keywords corresponding to core questions; Based on the semantic similarity and the weight of the business keywords, multiple relevant answers corresponding to the core question are retrieved from the knowledge base.

5. The method according to claim 1, characterized in that, The step of sorting the multiple related answers according to a preset sorting method, selecting a preset number of target answers from the sorting results, and recommending them includes: The multiple related answers are retrieved from the knowledge base to obtain the matching degree of the multiple related answers; The matching degree of multiple related answers is sorted according to the matching degree sorting method to obtain the sorting result; Based on the suggested answers, a preset number of target answers are selected from the sorted results and recommended through a separate display interface.

6. The method according to claim 1, characterized in that, The determination of answer matching accuracy based on user questions and a preset number of target answers includes: Generate a question-and-answer log based on user questions and a preset number of target answers; Obtain the answer matching accuracy corresponding to the question and answer log.

7. The method according to claim 1, characterized in that, Also includes: For answers with matching accuracy rates lower than a preset accuracy rate, the cases corresponding to these accuracy rates are pushed to the algorithm optimization module. The algorithm optimization module then performs incremental training on the semantic analysis model and the retrieval matching model, and marks the answers with matching accuracy rates lower than the preset accuracy rate as invalid answers. Update the knowledge base with the invalid answer.

8. An intelligent auxiliary question-and-answer processing device, characterized in that, The device is applied to an external intelligent agent, which is an end-to-end dedicated intelligent agent composed of an outbound call scenario customization module, an external architecture, a scenario-based adaptation module, and a closed-loop self-optimization module. The device includes: The real-time transcription unit is used to transcribe the target voice data in real time while connected to the outbound calling system, so as to obtain the text information corresponding to the target voice data. The extraction unit addresses the core issue of extracting text information based on natural language processing algorithm models. The retrieval unit is used to retrieve multiple relevant answers to the core question using a dedicated hybrid retrieval rule. The sorting and filtering unit is used to sort the multiple related answers according to a preset sorting method, filter out a preset number of target answers from the sorting results and recommend them; The determination unit is used to determine the answer matching accuracy based on the user's question and a preset number of target answers; The first update unit is used to update the natural language processing algorithm model and knowledge base index according to the answer matching accuracy, so as to complete the closed-loop self-optimization.

9. A storage medium, characterized in that, The storage medium includes stored instructions, wherein, when the instructions are executed, the device containing the storage medium is controlled to perform the intelligent assisted question-answering processing method as described in any one of claims 1 to 7.

10. An electronic device, characterized in that, It includes a memory, and one or more instructions, wherein one or more instructions are stored in the memory and configured to be executed by one or more processors as described in any one of claims 1 to 7.

Images