Intelligent customer service method, device and system

Abstract

The invention discloses an intelligent customer service method, device and system. The method comprises the following steps: S1, a service center accesses a first message; s2, the service center cleans the acquired first message; s3, performing semantic analysis; s4, when the first message is an after-sales message, the service center extracts an equipment identifier and an operation intention; and S5, executing feedback. According to the method, the multi-modal first message can be uniformly accessed, automatic analysis of a multi-language mixed instruction and equipment operation intention recognition are realized by utilizing a cloud AI large model, the multi-language mixed instruction is sent to a global equipment management platform to execute remote operation in combination with an automatic API workflow, and an intelligent reply is generated based on execution feedback; the problems that a traditional system is insufficient in cross-language analysis capacity, low in equipment operation efficiency and low in semantic recognition precision are solved, and the method has the advantages that the multi-language mixed instruction analysis capacity is improved, efficient integration of a customer service platform and an equipment management platform is achieved, and semantic analysis precision and an automatic closed-loop feedback mechanism are optimized.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent customer service robots, and in particular to a method, device and system for intelligent customer service. Background Art

[0002] In the existing intelligent customer service systems, there are several significant technical deficiencies.

[0003] First of all, the existing intelligent customer service robots lack the ability to parse multi-language mixed instructions and usually rely on a single language or a private database for semantic processing. This limitation causes cross-language users to often rely on manual translation when requesting services, which not only increases the communication cost, but also easily leads to semantic deviations due to translation errors, affecting the service quality and user experience. Secondly, there is a lack of effective integration between the existing customer service systems and device management platforms. The existing customer service systems only support text or voice interactions and cannot directly convert user instructions into operations on remote devices, such as fault diagnosis or firmware upgrade. This fragmented design requires user instructions to be relayed manually, which is not only inefficient but also prone to errors due to human operation mistakes. In addition, the accuracy of the existing semantic parsing models is insufficient. They usually perform parsing based on a static knowledge base or single-modal input (such as pure text) and fail to dynamically optimize by combining device historical data and multi-modal information (such as text and images), resulting in a low recognition accuracy for complex instructions. Finally, the existing systems lack an automated closed-loop feedback mechanism. The execution results usually rely on manual confirmation, lacking real-time feedback and execution log records, which not only reduces the user experience but also makes it difficult to continuously improve the intelligence level of the system.

[0004] All in all, when users initiate multi-language service requests through fragmented communication platforms, the traditional mode also faces two major pain points. First, the dispersion of cross-platform customer service systems causes manual customer service to need to switch responses between multiple software, which not only increases the response time but also easily leads to incomplete information transmission due to communication barriers. Second, the parsing of multi-language requirements relies on manual translation conversion, which not only increases the communication cost but also easily leads to semantic deviations due to translation errors, further affecting the service quality and user experience.

[0005] In view of the above problems, the existing technologies urgently need to be improved. Summary of the Invention

[0006] To solve at least one of the problems existing in the above-mentioned prior art, the first aspect of the present invention provides a method for intelligent customer service based on AI semantic understanding, which includes the following steps: Step S1: The service center accesses the first message; wherein, the first message is sent by the user through the communication software, and the first message includes at least one of the following: a consultation message or an after-sales message, and the communication software sends the message to the service center; Step S2: The service center performs data cleaning on the obtained first message; Step S3: Perform semantic parsing to identify mixed language instructions and extract the device identifier and operation intention; Step S4: When the first message is an after-sales message, the service center extracts the device identifier and operation intention according to the cloud AI large model and sends them to the global device management platform, generates and sends a device operation instruction to the global device management platform, so as to directly perform after-sales / maintenance operations on the device to be after-saled; Step S5: Execute feedback; wherein, the global device management platform executes the device operation instruction and returns the execution feedback information of the device to the service center.

[0007] In the method of the intelligent customer service as described above, optionally, in the step S1, the following steps are further included: the communication software determines whether the received message is the user's first message, if so, the message is sent to the service center; otherwise, the execution ends.

[0008] In the method of the intelligent customer service as described above, optionally, the cloud AI large model is a multi-modal semantic parsing model, and the first message includes at least one modal data of text, voice or image; In the step S2, the data from different channels are cleaned separately, and the corpus of the user's message for each channel is passed to the cloud AI large model for training, to understand the communication habits, speaking tones and styles of users in different channels, and to more accurately analyze the device identifier and operation intention, and the corpus is provided to the large model for training according to the different speaking habits of users.

[0009] In the method of the intelligent customer service as described above, optionally, the method further includes step S6: customer service response; wherein, the service center intelligently generates a reply message based on the execution feedback information, and sends the reply message to the user through the communication software interface.

[0010] In the method of the intelligent customer service as described above, optionally, the API provided by the service center to the outside is used to receive messages from each channel, and the API of the global device management platform is called by the service center to operate the devices in the hands of channel users.

[0011] In the method of the intelligent customer service as described above, optionally, in the step S2, the semantic parsing includes the following steps: judging the language, performing semantic conversion of a preset language, and then analyzing the device identifier and the operation intention that the user currently needs to perform on the device according to the semantics of the preset language.

[0012] In the method of the intelligent customer service as described above, optionally, the method further includes the service center judging whether it is necessary to call the global device management platform for operation: when the service center receives the device identifier and the operation intention extracted by the cloud AI large model, it determines that the first message is an after-sales message and needs to call the global device management platform for operation; on the contrary, when the cloud AI large model fails to extract the device identifier and the operation intention, it determines that the first message is a consultation message and there is no need to call the global device management platform to operate the device.

[0013] In the method of the intelligent customer service as described above, optionally, in the step S6, the service center automatically generates a corresponding reply message to the communication software according to the execution feedback information of the global device management platform; wherein, the reply message is determined by the operation intention.

[0014] To achieve the above object, the second aspect of the present invention provides an AI model training method, wherein, implementing the intelligent customer service method as described in any one of the first aspect includes: Step T1: Construct a multi-language training corpus, which at least includes one of the following: user conversations from various channels, mixed language instruction texts, device operation logs, and corresponding intention labels; Step T2: Initialize the AI parsing model, and jointly train the language translation task and the device operation intention recognition task through multi-task learning, wherein the language translation task is used to align the multi-language semantic space, and the intention recognition task is used to extract the device identifier and the operation instruction; Step T3: Introduce a device operation knowledge base, embed the device historical operation and maintenance data as prior knowledge into the model training, and dynamically weight the device-related features through the attention mechanism; wherein the historical operation and maintenance data at least includes one of the following: failure type, after-sales operation record, firmware version information; Step T4: Regularly optimize the model based on reinforcement learning to make the analysis results increasingly accurate; Step T5: Deploy an incremental learning mechanism, and regularly input the newly added multi-language instruction data and device operation logs into the model for fine-tuning to adapt to the semantic changes in the globalized scenario.

[0015] To achieve the above object, the third aspect of the present invention provides an intelligent customer service device, wherein, using the intelligent customer service method as described in any one of the foregoing first aspect includes: A communication interface module for receiving a first message from communication software, where the first message includes at least one of: a consultation message or an after-sales message; A data preprocessing module, connected to the communication interface module, for removing dirty data from the received first message; An AI semantic parsing module, deploying a cloud AI large model and a multilingual processing unit, connected to the data preprocessing module through a high-speed network interface, for semantic parsing of the cleaned first message, automatically identifying mixed-language instructions and extracting device identifiers and operation intentions; An instruction conversion control module, connected to the AI semantic parsing module, including a protocol adapter and an API workflow controller, sending device operation instructions to the global device management platform according to the device identifiers and operation intentions extracted by the cloud AI large model, and sending device operation instructions to the global device management platform; A device operation execution module, communicatively connected to the instruction conversion control module, including a distributed control node of the global device management platform and an execution feedback sensor, where the distributed control node remotely operates the device according to the device operation instructions, and the execution feedback sensor collects device status data in real time; among them, the global device management platform executes the device operation instructions and returns execution feedback information to the response generation module; A response generation module, respectively connected to the device operation execution module and the communication interface module, including an intelligent reply generation chip and a communication interface drive unit, where the intelligent reply generation chip intelligently generates a multilingual reply message based on the execution feedback information and returns it to the communication software of the user terminal through the communication interface module.

[0016] To achieve the above object, a fourth aspect of the present invention provides a system for an intelligent customer service, where an intelligent customer service method as described in any one of the foregoing first aspects is used, including: a user terminal equipped with communication software, a service center, a cloud AI large model service, and a global device management platform; among them, the service center is respectively bidirectionally communicatively connected to the user terminal, the cloud AI large model service, and the global device management platform.

[0017] To achieve the above object, a fifth aspect of the present invention provides a computer-readable storage medium, where the computer-readable storage medium stores computer-executable instructions or a computer program, and when the computer-executable instructions or the computer program are processed and executed, the intelligent customer service method as described in any one of the foregoing first aspects is implemented.

[0018] The method, device and system of an intelligent customer service provided by the present invention can uniformly access multimodal first messages, use a cloud AI large model to realize automatic parsing of multilingual mixed instructions and recognition of device operation intentions, establish a dual association mechanism between operation intentions and execution feedback, and send them to the global device management platform through an automated API workflow to directly perform remote operations on devices, and generate intelligent responses based on the execution feedback of the operation completion status of the devices, solving the problems of insufficient cross-language parsing ability, low device operation efficiency and low semantic recognition accuracy in traditional systems, and having the advantages of improving the parsing ability of multilingual mixed instructions, realizing efficient integration of customer service and device management platform, and optimizing semantic parsing accuracy and automated closed-loop feedback mechanism. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings in the following description are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0020] Figure 1 is a flowchart of an embodiment of a method of an intelligent customer service of the present invention; Figure 2 is Figure 1 a detailed flowchart of the method of the intelligent customer service in Figure 3 is a message flow diagram of an embodiment of an intelligent customer service and device operation integration system of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0021] To make the objectives, technical solutions and advantages of the embodiments of the present invention clearer, the following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the drawings in the embodiments of the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts fall within the scope of protection of the present invention.

[0022] Terms such as "including" and "comprising" indicate that in addition to the components directly and clearly described in the specification and claims, the technical solutions of the present invention do not exclude the situation of having other components not directly or clearly described.

[0023] As Figures 1 to 2 shown, the method of the intelligent customer service based on AI semantic understanding of the present invention may include the following steps: Step S1: The service center receives the first message. Here, the service center is a global or multi-device after-sales service intelligent middle platform, which refers to a central hub platform that centrally processes user service requests and device operation instructions. Specifically, it can be implemented using a microservices architecture and containerization deployment technology to achieve the unified access and management functions of cross-regional device services.

[0024] In step S1, the first message sent by the user through the communication software refers to multi-channel messages, which can be specifically divided into consultation messages or after-sales messages. The communication software sends the messages to the service center through a standardized interface. The standardized interface refers to a standardized data interaction channel that follows an open communication protocol. Specifically, it can be implemented using RESTful API or WebSocket protocol to ensure data transmission compatibility between the communication software and the service center. For example, the service center receives these messages through the RESTful API interface and stores them uniformly in the message queue. Among them, the API provided by the service center to the outside is used to receive messages from each channel. At the same time, the global device management platform provides API interfaces for the service center to call to perform remote operations on user devices. These APIs may include functions such as device status query, remote control, and firmware update. After receiving the operation intention parsed by AI, the service center will call the corresponding device management API. For example, if the operation intention is "restart the device", the service center will call the "reboot" API of the device management platform and pass in the corresponding device identifier and authentication information.

[0025] However, in this process, there may be messages from non-user sources misconnected to the service center, resulting in resource waste, redundant data processing, and potential system interference problems.

[0026] In response, step S1 can also include the following steps: The communication software determines whether the received message is the user's first message. If it is, the message is sent to the service center; otherwise, the execution ends.

[0027] In this embodiment, the communication software can be the customer service system in a website or app. The information received through these website or app customer service systems is uniformly classified as the user's channel information. Another type of communication software can be a third-party communication software. This application sets up an independent account for this type of software as the unified input and output for communicating with the user. The information coming from this account is the user's channel information because this account only communicates with the user.

[0028] As a result, redundant calculations for data cleaning and semantic parsing are reduced, improving the overall processing efficiency of the system. Further, this solution prevents malicious or irrelevant messages from interfering with the device operation process, ensuring that the service center only processes legitimate user requests, enhancing the security and reliability of the system. Specifically, by screening at the initial stage of message reception, the waste of system resources is significantly reduced, and potential system interference problems are minimized. For example, system error messages, test messages, or other non-user requests are effectively intercepted and do not enter the processing flow of the service center. This not only optimizes the system resource allocation but also improves the response speed of the service center to real user requests. Thus, the technical solution of this application enhances the user service quality while improving the system efficiency.

[0029] Step S2: The service center performs data cleaning on the obtained first message. Data cleaning refers to the process of filtering out noise and converting the format of the original message, which can be specifically implemented using regular expression matching and outlier detection algorithms to eliminate the interference of unstructured data on semantic parsing.

[0030] Step S3: Perform semantic parsing according to the cloud AI large model to automatically identify the mixed-language instructions and extract the device identifier to be serviced and the operation intention.

[0031] After the service center removes the dirty data from the received first message, semantic parsing is performed by calling the cloud AI large model through the API to automatically identify the mixed-language instructions and extract the device identifier and the operation intention. Among them, calling the cloud AI large model through the API means remotely accessing artificial intelligence computing resources through an application programming interface, which can be specifically implemented using HTTP requests and the OAuth2.0 authentication mechanism to call the multi-modal semantic parsing model deployed on the cloud server. The cloud AI large model in this embodiment is a multi-modal semantic parsing model, and the first message can include various modal data such as text, voice, or images.

[0032] Since user messages from different channels may include at least one of the modal data of text, voice, or images, and there are differences in communication habits, speaking tones, and styles among different channels, if no differential processing is performed for multi-modal data and channel characteristics, the semantic parsing model will not be able to accurately analyze the device identifier and the operation intention, thereby affecting the accuracy of subsequent instruction conversion.

[0033] Further, in the step S3, data from different channels is cleaned separately, and the corpus of different user messages for each channel is transmitted to the cloud AI large model for training, to understand the communication habits, speaking tones, and styles of different channels, and to more accurately analyze the device identifier and the operation intention. For example, for the same problem of poor network, the user's feedback may be "device number + poor network" or "device number + very slow to use" or directly "device number + switch the operator". Through this embodiment, the corpus can be provided to the large model for training according to the different speaking habits of users, making the analysis of each channel more and more accurate. The device number is the unique identifier for the global device management platform to operate on the device, and the operation intention that the user needs to perform on the device may include, for example, restarting the device, switching the network, querying the WIFI name and password, etc.

[0034] Specifically, the multi-modal semantic parsing model realizes cross-modal feature extraction by fusing a convolutional neural network and a sequence modeling module. For example, voice data can be input into the speech recognition sub-model after Mel spectrogram conversion, and image data is used to extract device interface elements through an object detection algorithm. Data cleaning for different channels uses independent data pipelines. For example, messages from the social software channel are filtered for emojis through regular expressions, and messages from the email channel have signature information removed through keyword matching. For the corpus training of each channel, a dynamic learning rate adjustment strategy is adopted. For example, a higher learning rate is set during the corpus training of the social software channel to adapt to the rapid iteration of colloquial expressions, and a lower learning rate is set during the corpus training of the email channel to maintain the stability of written descriptions. The differential training of user speaking habits is achieved by constructing channel-specific word vectors. For example, for user groups who are accustomed to using abbreviations, a mapping rule of "device ID → device identifier" is established in the corpus of the social software channel.

[0035] During the analysis process, the large model trains on the received corpus, determines the language, performs semantic conversion of the preset language, and then analyzes the device number and the operation intention that the user currently needs to perform on the device according to the semantics of the preset language. For example, there will be different customized trainings for channel messages in different national languages, which fit the channel communication language and communication habits. For example, if the channel uses English as the communication language, the AI semantic parsing model will be set as an English expert, and with customized corpus training, the channel problems are parsed to extract the device number and the operation that the channel wants to perform on the device. This step is an automatic recognition, and different branches are taken for channels with different languages.

[0036] Semantic parsing may include the following steps: determining the language, performing semantic conversion of the preset language, and then analyzing the device identifier and the operation intention that the user currently needs to perform on the device according to the semantics of the preset language.

[0037] Among them, the language judgment step uses a natural language processing model to identify the grammatical structure and vocabulary features of the original message to determine whether it is a preset language or a mixed language containing non-preset language characters. Assuming that the preset language in the country is Chinese, the Chinese semantic conversion step can use a cross-language conversion model based on a neural network to map non-Chinese vocabulary or sentence patterns to the corresponding Chinese semantic framework, such as converting the English device model "Model-X" to the Chinese standard model "Model X", or converting the Japanese grammatical structure "操作を実行" to the Chinese subject-object structure "执行操作". The device identification and operation intention analysis step extracts the device identifier based on the converted Chinese semantics through the named entity recognition model, and parses the operation verb and target parameter in combination with the context dependency, such as extracting the device ID "Device123" and the operation parameter "温度设置25℃" from "Please adjust the temperature of Device123 to 25℃".

[0038] Specifically, when a mixed language instruction is received, the language components in the message are first identified through the language judgment module. If non-Chinese content is detected, the semantic conversion process is triggered. For example, the instruction containing mixed Chinese and English "Help me check the status of device A" is converted into a pure Chinese instruction "Help me check the status of device A". The converted Chinese text is input into the semantic analysis model, and the device identification "device A" and the operation intention "status query" are accurately extracted through Chinese grammatical rules and device database matching. In this process, Chinese is used as a unified analysis benchmark to avoid parsing errors caused by grammatical differences when multiple languages are mixed. For example, after the German verb postposition structure "Gerät B einschalten" is converted into Chinese "Turn on device B", the verb position is adjusted to make the intention recognition more accurate. Through the dual processing of language standardization and semantic mapping, it is ensured that instructions expressed in different languages are parsed under a unified logical framework, and the accuracy of device operation instruction conversion is improved.

[0039] In an optional embodiment, the present application further proposes that the service center determines whether it is necessary to call the global device management platform to perform an operation: in the semantic parsing stage, when the service center receives the device identification and operation intention extracted by the cloud-based AI big model, it determines that the first message is an after-sales message, and triggering the device operation requires calling the global device management platform to perform the operation; conversely, when the cloud-based AI big model does not extract the device number and operation intention, it determines that the first message is an advisory message, and there is no need to call the global device management platform to operate the device.

[0040] Among them, the core of the judgment logic lies in the dual-element extraction status of the device identifier and the operation intention. The device identifier can include structured data such as device serial numbers and product models, and the operation intention can be parsed into action types such as firmware upgrade instructions and remote restart commands. For example, when the message contains "Device SN12345 needs to be restarted", the AI model will extract the device identifier SN12345 and the operation intention "restart", triggering the device operation process; if the message only contains "How to set device parameters", since the device identifier and the operation intention are not extracted, only a customer service reply will be generated. The judgment process is directly related to the semantic parsing result. Specifically, the parsing result is transmitted to the service center through the API interface, and the classification decision is executed by the logic control unit built into the middle platform.

[0041] Specifically, in the message processing flow, after receiving the parsing result of the cloud AI model, the service center first verifies the integrity of the device identifier and the operation intention. If both exist, it sends an operation instruction containing the device identifier to the device management platform. For example, it generates an instruction packet in JSON format {"device_id":"SN12345", "action":"restart"}; if either element is missing, it skips the instruction conversion step and directly calls the customer service response module to generate a text reply. During the execution process, the API call frequency of the device management platform is dynamically matched with the message type. For example, the trigger rate of after-sales messages is 100%, while the trigger rate of consultation messages is 0%, thereby reducing the invalid interface calls by more than 30%. This mechanism collaborates with the multilingual parsing module: when multi-modal data is input into the AI model after cleaning, the device identifier can be extracted through QR code recognition in the image or serial number broadcast in the voice, and the operation intention may be comprehensively analyzed by combining text instructions and the device status in the image. For example, when the user sends a message containing a device failure photo and the text "Cannot power on", the model extracts the device identifier through image recognition and combines text parsing to parse the operation intention as "fault diagnosis", thereby triggering a call to the device management platform.

[0042] For the case of messages determined to be consultation messages, the service center does not trigger the device operation process. Instead, it forwards the message to a dedicated consultation processing module. This module can include functions such as knowledge base query and intelligent question answering, and is used to generate reply content for user consultations.

[0043] In addition, to ensure system security, the service center conducts permission verification before each call to the global device management platform. The verification content includes: the legality of the message source, the compliance of the operation instructions, and the user's operation permission for the target device. Only when all verification items are passed is the actual device operation allowed. The service center has also implemented an operation log recording function. Each device operation request, whether successful or not, will be detailedly recorded, including information such as timestamp, device identifier, operation type, execution result, etc. These logs are used for subsequent auditing, fault analysis, and system optimization.

[0044] Step S4: Instruction conversion.

[0045] In step S4, when the first message is an after-sales message, the service center sends the device operation instruction to the global device management platform according to the device identifier and operation intention extracted by the cloud AI large model through the automated API workflow. As mentioned above, when the cloud AI large model fails to extract the device number and operation intention, it is determined that the first message is a consultation message, and there is no need to call the global device management platform to operate the device. The user can be replied according to the preset reply template.

[0046] Among them, the automated API workflow refers to a preset logical chain for generating and transmitting device operation instructions, which can be specifically implemented by using workflow engines such as Apache Airflow or AWS Step Functions to ensure the real-time and accuracy of triggering instructions on the device management platform.

[0047] In this embodiment, the service center can convert the structured data output by the AI model (including fields such as device ID and operation type) into JSON format. Through a pre-defined API mapping table, the JSON data is converted into an instruction format recognizable by the device management platform. For example, converting "Check the temperature sensor of device A" into {"device_id": "A", "action": "check_sensor", "sensor_type": "temperature"}.

[0048] Step S5: Execution feedback.

[0049] In step S5, the global device management platform executes the device operation instruction and returns the execution feedback information to the service center. Among them, the execution feedback information refers to a set of status data of the device operation result, which can be specifically implemented by encapsulating the device response code and operation log in JSON format to provide structured input for intelligent reply generation. It has been described in detail above and will not be elaborated here.

[0050] Step S6: Customer service response.

[0051] In step S6, the service center intelligently generates a reply message based on the execution feedback information and sends the reply message to the user through the communication software interface. Herein, intelligently generating a reply message means automatically constructing the user response content based on the execution result, which can be specifically implemented by using a template engine and natural language generation technology to generate multilingual feedback information in combination with the device operation status.

[0052] In an alternative embodiment, the service center automatically generates a corresponding reply message to the communication software according to the execution feedback information of the global device management platform; wherein, the reply message is determined by the operation intention.

[0053] The process of generating the reply message may include a dynamic template matching mechanism based on the operation intention.

[0054] Optionally, the specific implementation method may include establishing a mapping relationship library between the operation intention types and the standard reply templates. For example, when the operation intention is device restart, it is mapped to a template including the execution status confirmation and the estimated completion time. The parsing unit of the execution feedback information is configured to identify the status code or log data returned by the device management platform. For example, combining the "instruction received" status code with the operation intention triggers the template filling engine to generate a complete reply. The relevance verification module of the operation intention and the execution feedback is deployed to detect whether there is a logical conflict between the feedback information and the original intention. If a conflict is detected, an exception handling process is triggered.

[0055] After the device operation instruction is executed, the global device management platform returns the execution feedback information including the status code to the service center. The service center extracts the key fields in the status code through the parsing module, such as the instruction reception status and the execution progress code. Meanwhile, the operation intention database is retrieved to obtain the original operation intention type corresponding to the current session. The reply generation engine inputs the operation intention and the execution feedback information into a multi-dimensional matching algorithm. For example, when the operation intention is firmware upgrade and the feedback information contains "upgrade package verification failed", it automatically selects the reply template of "upgrade file exception detected, please re-upload the installation package". The filling of the dynamic parameters in the reply template is implemented according to the specific values returned by the device management platform. For example, in the device restart scenario, according to the average restart duration in the historical data of the device model, the reply content of "it is expected to be completed in 2 minutes and 30 seconds" is generated. Thus, the generated reply message not only accurately reflects the actual execution status of the device operation, but also forms a natural language expression that conforms to the user's cognitive habits through the semantic supplement of the original operation intention, and can complete the closed-loop response without manual intervention.

[0056] The solution of this embodiment establishes a dual association mechanism between operation intention and execution feedback. The response content can not only accurately reflect the operation result of the device, but also dynamically adapt to the context of the user's original request. This method effectively bridges the semantic gap between the device operation result and the user's understanding, and improves the reliability of the automated closed-loop. Further, since the response content is more in line with the actual needs of the user, the need for manual secondary verification is reduced, thereby improving the response efficiency and accuracy of the intelligent customer service system.

[0057] To achieve the above object, the present invention also provides an AI model training method to implement the intelligent customer service method as described in any one of the foregoing embodiments, including: Step T1: Construct a multilingual training corpus, which at least includes one of the following: user conversations from various channels, mixed-language instruction texts, device operation logs, and corresponding intention tags; Step T2: Initialize the AI parsing model, and jointly train the language translation task and the device operation intention recognition task through multi-task learning. Among them, the language translation task is used to align the multilingual semantic space, and the intention recognition task is used to extract the device identifier and operation instructions; Step T3: Introduce a device operation knowledge base, embed the device historical operation and maintenance data as prior knowledge into the model training, and dynamically weight the device-related features through the attention mechanism; among them, the historical operation and maintenance data at least includes one of the following: failure type, after-sales operation record, firmware version information; Step T4: Regularly optimize the model based on reinforcement learning to make the analysis result increasingly accurate; Step T5: Deploy an incremental learning mechanism, and regularly input the newly added multilingual instruction data and device operation logs into the model for fine-tuning to adapt to the semantic changes in the globalized scenario.

[0058] To achieve the above object, the present invention also provides a device for intelligent customer service. Among them, the method for intelligent customer service as described in any one of the foregoing embodiments is used, including: a communication interface module, a data preprocessing module, an AI semantic parsing module, an instruction conversion control module, a device operation execution module, and a response generation module.

[0059] Specifically, the communication interface module is used to receive a first message from communication software, and the first message may include a consultation message or an after-sales message; the data preprocessing module is connected to the communication interface module to remove dirty data from the received first message; the AI semantic parsing module deploys a cloud AI large model and a multilingual processing unit, and is connected to the data preprocessing module through a high-speed network interface to perform semantic parsing on the cleaned first message, automatically identify mixed-language instructions, and extract the device identifier and operation intention; the instruction conversion control module is connected to the AI semantic parsing module, including a protocol adapter and an API workflow controller, and according to the device identifier and operation intention extracted by the cloud AI large model, sends them to the global device management platform through an automated API workflow, and sends device operation instructions to the global device management platform; the device operation execution module is communicatively connected to the instruction conversion control module, including a distributed control node and an execution feedback sensor of the global device management platform, the distributed control node remotely operates the device according to the device operation instructions, and the execution feedback sensor collects device status data in real time; among them, the global device management platform executes the device operation instructions and returns the execution feedback information to the response generation module; the response generation module is respectively connected to the device operation execution module and the communication interface module, and may include an intelligent reply generation chip and a communication interface drive unit, and the intelligent reply generation chip intelligently generates a multilingual reply message based on the execution feedback information and returns it to the communication software of the user terminal through the communication interface module.

[0060] To achieve the above object, as Figure 3 shown, the present invention also provides a system for intelligent customer service, wherein the method for intelligent customer service described in any of the foregoing embodiments is used, including: a user terminal equipped with communication software, a service center, a cloud AI large model service, and a global device management platform; wherein, the service center is respectively bidirectionally communicatively connected to the user terminal, the cloud AI large model service, and the global device management platform. Its functions and specific ways of communication transmission have been described in detail above and will not be elaborated here.

[0061] To achieve the above object, the present invention also provides a computer-readable storage medium, and the computer-readable storage medium stores executable instructions or programs, and when the executable instructions or programs are processed and executed, the method for intelligent customer service described in any previous embodiment is implemented.

[0062] The readable storage medium is, for example, a memory. The memory can be a volatile memory or a non-volatile memory, or the memory can include both a volatile memory and a non-volatile memory at the same time. Among them, the non-volatile memory can be a read-only memory (ROM), a programmable ROM (PROM), an erasable programmable ROM (EPROM), an electrically erasable PROM (EEPROM), or a flash memory. The volatile memory can be a random access memory (RAM), which is used as an external cache. By way of example but not limitation, many forms of RAM are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchlink DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0063] If the integrated unit in the above embodiments is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in the above-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. The software product is stored in the storage medium and includes several instructions for causing one or more devices (which can be personal terminals, user terminals, or network devices, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention.

[0064] The preferred specific embodiments of the present invention have been described in detail above. Only several implementation manners of the present invention are expressed, but it should not be construed as a limitation to the scope of the patent. The technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in this specification. It should be understood that those of ordinary skill in the art can make many modifications and variations according to the concept of the present invention without creative labor. Therefore, without departing from the concept of the present invention, all technical solutions that can be obtained by those skilled in the art in the technical field according to the concept of the present invention through logical analysis, reasoning or limited experiments on the basis of the prior art should fall within the protection scope determined by the claims.

Claims

1. A method for intelligent customer service, characterized in that, It includes the following steps: Step S1: The service center obtains a first message; wherein, the first message is sent by the user through a communication software, and the first message includes at least one of the following: a consultation message or an after-sales message, and the communication software sends the message to the service center; Step S2: The service center performs data cleaning on the obtained first message; Step S3: Perform semantic parsing to identify mixed-language instructions and extract the device identifier and operation intention; Step S4: When the first message is an after-sales message, the service center extracts the device identifier and operation intention of the device to be after-saled according to the cloud AI large model and sends them to the global device management platform, generates and sends a device operation instruction to the global device management platform, so as to directly perform after-sales / maintenance operations on the device to be after-saled; Step S5: Execute feedback; wherein, the global device management platform executes the device operation instruction and returns the execution feedback information of the device to the service center.

2. The method of the intelligent customer service according to claim 1, wherein In the step S1, the following steps are further included: the communication software determines whether the received message is the user's first message, if so, it sends the message to the service center; otherwise, it ends the execution.

3. The method of the intelligent customer service according to claim 1, characterized in that The cloud AI large model is a multi-modal semantic parsing model, and the first message includes at least one modal data of text, voice or image; In the step S2, the data from different channels are cleaned separately, and the corpus of the user message for each channel is passed to the cloud AI large model for training, to understand the communication habits, speaking tones and styles of users in different channels, more accurately analyze the device identifier and operation intention, and provide the corpus to the large model for training according to the different speaking habits of users.

4. The method of the intelligent customer service according to claim 1, wherein The method further includes step S6: customer service response; wherein, the service center intelligently generates a reply message based on the execution feedback information and sends the reply message to the user through the communication software interface.

5. The method of the intelligent customer service according to claim 1, wherein In the step S2, the semantic parsing includes the following steps: judge the language, perform semantic conversion of the preset language, and then analyze the device identifier and the operation intention that the user currently needs to perform on the device according to the semantic of the preset language.

6. The method of the intelligent customer service according to claim 1, characterized in that The method further includes that the service center judges whether it is necessary to call the global device management platform for operation: when the service center receives the device identifier and operation intention extracted by the cloud AI large model, it determines that the first message is an after-sales message and needs to call the global device management platform for operation; otherwise, when the cloud AI large model does not extract the device identifier and operation intention, it determines that the first message is a consultation message and there is no need to call the global device management platform to operate on the device.

7. The method of the intelligent customer service according to claim 4, wherein In the step S6, the service center automatically generates a corresponding reply message to the communication software according to the execution feedback information of the global device management platform; wherein, the reply message is determined by the operation intention.

8. An AI model training method, characterized in that, The method for implementing the intelligent customer service as described in any one of claims 1 to 7 includes: Step T1: Construct a multilingual training corpus, which at least includes one of the following: user conversations from various channels, mixed-language instruction texts, device operation logs, and corresponding intent tags; Step T2: Initialize the AI model, and jointly train the language translation task and the device operation intent recognition task through multi-task learning. Among them, the language translation task is used to align the multilingual semantic space, and the intent recognition task is used to extract device identifiers and operation instructions; Step T3: Introduce a device operation knowledge base, embed device historical operation and maintenance data as prior knowledge into model training, and dynamically weight device-related features through an attention mechanism; among them, the historical operation and maintenance data at least includes one of the following: failure types, after-sales operation records, firmware version information; Step T4: Regularly optimize the model based on reinforcement learning to make the analysis results increasingly accurate; Step T5: Deploy an incremental learning mechanism, and regularly input newly added multilingual instruction data and device operation logs into the model for fine-tuning to adapt to semantic changes in the globalized scenario.

9. An intelligent customer service device, characterized in that, The method of using the intelligent customer service as described in any one of claims 1 to 7 includes: A communication interface module for receiving a first message from a communication software, where the first message at least includes one of the following: a consultation message or an after-sales message; A data preprocessing module connected to the communication interface module to remove dirty data from the received first message; An AI semantic parsing module that deploys a cloud AI large model and a multilingual processing unit, and is connected to the data preprocessing module through a high-speed network interface to perform semantic parsing on the cleaned first message, automatically identify mixed-language instructions, and extract device identifiers and operation intents; An instruction conversion control module connected to the AI semantic parsing module, including a protocol adapter and an API workflow controller, and sends device operation instructions to the global device management platform through an automated API workflow according to the device identifiers and operation intents extracted by the cloud AI large model; A device operation execution module communicatively connected to the instruction conversion control module, including a distributed control node of the global device management platform and an execution feedback sensor. The distributed control node remotely operates the device according to the device operation instruction, and the execution feedback sensor collects device status data in real time; among them, the global device management platform executes the device operation instruction and returns the execution feedback information to the response generation module; A response generation module connected to the device operation execution module and the communication interface module respectively, including an intelligent reply generation chip and a communication interface drive unit. The intelligent reply generation chip intelligently generates multilingual reply messages based on the execution feedback information and returns them to the communication software of the user terminal through the communication interface module.

10. A system for intelligent customer service, characterized in that, The method of using the intelligent customer service as described in any one of claims 1 to 7 includes: a user terminal equipped with communication software, a service center, a cloud AI large model service, and a global device management platform; among them, the service center is communicatively connected to the user terminal, the cloud AI large model service, and the global device management platform bidirectionally.

Images