Communication service processing method, electronic equipment, storage medium and computer product

Abstract

The invention discloses a communication service processing method, electronic equipment, a storage medium and a computer product, and the method comprises the steps: obtaining first information related to a communication service, the first information comprising multi-dimensional information obtained by carrying out intention recognition on a conversation related to the communication service; obtaining second information related to the communication service, wherein the second information comprises an execution action of each executable task unit obtained by dividing the first information and an access condition of each executable task unit; third information is obtained, and the third information comprises analysis data generated by analyzing a result obtained by completing the execution action of each executable task unit under the access condition; and inputting the analysis data into the large language model to obtain a processing strategy related to the communication service output by the large language model.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a communication service processing method, electronic equipment, storage medium and computer product. Background Art

[0002] The existing support system has established instant messaging and work order scheduling modules. When field personnel encounter specific problems in the production process, they can submit work orders through the mobile application (application, app) of the support system and ask questions to the back-end personnel to seek technical support. After receiving the order on the personal computer (Personal Computer, PC) of the support system, the back-end personnel will communicate with the field personnel, check the existing network data according to the specific problem situation, and provide technical solutions to the field personnel based on their personal experience.

[0003] It can be seen that the problems encountered in the communication business are currently handled by relying on the background side. Summary of the invention

[0004] The embodiments of the present application provide a communication service processing method, an electronic device, a storage medium, and a computer product to enhance the ability of the electronic device to process problems encountered in communication services on site.

[0005] The technical solution of the embodiment of the present application is implemented as follows:

[0006] The present invention provides a communication service processing method, the method comprising:

[0007] Acquiring first information related to a communication service, where the first information includes multi-dimensional information obtained by performing intent recognition on a conversation related to the communication service;

[0008] Acquire second information related to the communication service, the second information including an execution action of each executable task unit obtained by dividing the first information, and an access condition of each executable task unit;

[0009] Acquire third information, the third information including analysis data generated by analyzing a result obtained by completing an execution action of each executable task unit under the access condition;

[0010] The analysis data is input into a large language model to obtain a processing strategy related to the communication service output by the large language model.

[0011] In the above solution, the acquiring of first information related to the communication service includes:

[0012] Obtaining a target text obtained by combining the text of the n+1th round of dialogue related to the communication service with the text of the first n rounds of dialogue, where n is a positive integer;

[0013] The target text is input into the large language model to obtain the first information output by the large language model.

[0014] In the above solution, the obtaining of the second information related to the communication service includes:

[0015] Inputting the first information into the large language model, obtaining a return result and related analysis information of the execution action of each executable task unit output by the large language model;

[0016] Obtaining interactive prompt information generated by the large language model based on the returned result and the relevant analysis information, wherein the interactive prompt information is used to guide the target object to perform a target operation and return a result of the target operation, wherein the target operation includes on-site troubleshooting and / or operating a target device;

[0017] Business data extracted by the large language model based on the result of the target operation is obtained as the access condition of each executable task unit.

[0018] In the above solution, the step of inputting the analysis data into a large language model to obtain a processing strategy related to the communication service output by the large language model includes:

[0019] The analysis data is input into a large language model to obtain a first strategy corresponding to each executable task unit output by the large language model, wherein the first strategy includes first text information that conforms to communication business logic; and the processing strategy includes the first strategy.

[0020] In the above solution, after inputting the analysis data into the large language model to obtain the first strategy corresponding to each executable task unit output by the large language model, the method further includes:

[0021] A second strategy is generated based on the first strategy, wherein the second strategy includes one or more of the following: second text information, picture information, and document information that conform to communication service logic, wherein the second text information is obtained by text expansion of the first text information.

[0022] In the above solution, before inputting the analysis data into the large language model to obtain the first strategy corresponding to each executable task unit output by the large language model, the method further includes:

[0023] The autoregressive language model is trained, and the autoregressive language model is fine-tuned based on the content target composed of each sample executable task unit and the sample strategy to obtain the large language model.

[0024] The present application provides a communication service processing device, including:

[0025] an acquisition unit, configured to acquire first information related to a communication service, wherein the first information includes multi-dimensional information obtained by performing intent recognition on a conversation related to the communication service;

[0026] an acquiring unit, configured to acquire second information related to the communication service, wherein the second information includes an execution action of each executable task unit obtained by dividing the first information, and an access condition of each executable task unit;

[0027] an acquisition unit, configured to acquire third information, wherein the third information includes analysis data generated by analyzing a result of completing an execution action of each executable task unit under the access condition;

[0028] The processing unit is used to input the analysis data into a large language model to obtain a processing strategy related to the communication service output by the large language model.

[0029] The embodiment of the present application also provides an electronic device, comprising: a processor and a memory for storing a computer program that can be run on the processor; wherein:

[0030] The processor is used to execute the steps of any one of the above methods when running the computer program.

[0031] An embodiment of the present application further provides a storage medium on which a computer program is stored, wherein the computer program implements the steps of any of the above methods when executed by a processor.

[0032] An embodiment of the present application further provides a computer product, including a computer program, which implements the steps of any of the above methods when executed by a processor.

[0033] An embodiment of the present application provides a communication service processing method, electronic device, storage medium and computer product, the method comprising: obtaining first information related to the communication service, the first information comprising multi-dimensional information obtained by performing intent recognition on a conversation related to the communication service; obtaining second information related to the communication service, the second information comprising an execution action of each executable task unit obtained by dividing the first information, and an access condition for each executable task unit; obtaining third information, the third information comprising analyzing a result obtained by completing the execution action of each executable task unit under the access condition to generate analysis data; inputting the analysis data into a large language model to obtain a processing strategy related to the communication service output by the large language model. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1A flow chart of a communication service processing method according to an embodiment of the present application;

[0035] Figure 2 This is a schematic diagram of an embodiment of the present application using a sliding window mechanism to process a conversation;

[0036] Figure 3 A schematic diagram of the communication service processing flow of an embodiment of the present application;

[0037] Figure 4 A schematic diagram for identifying the intended embodiments of the present application;

[0038] Figure 5 A schematic diagram of the process from the abbreviated scheme to the detailed scheme of the embodiment of the present application;

[0039] Figure 6 This is a schematic diagram of the interface of the business processing scenario of the embodiment of the present application;

[0040] Figure 7 A schematic diagram of a distributed deployment of a business support system according to an embodiment of the present application;

[0041] Figure 8 This is a schematic diagram of the structure of a communication service processing device according to an embodiment of the present application;

[0042] Fig. 9 It is a schematic diagram of the structure of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0043] The present application is further described in detail below in conjunction with the accompanying drawings and embodiments.

[0044] The existing support system has established instant messaging and work order scheduling modules. When field personnel encounter specific problems in the production process, they can submit work orders through the mobile app of the support system and ask questions to the back-end personnel to seek technical support. After receiving the order on the PC side of the support system, the back-end personnel will have a dialogue with the field personnel, check the existing network data according to the specific problem situation, and provide technical solutions to the field personnel based on their personal experience. It can be seen that the current problem encountered in the communication business is handled by relying on the back-end side.

[0045] This application introduces a large language model (LLM) construction support system on the electronic device side, provides an intelligent assistant, builds multi-round dialogue, task scheduling, automatic execution, and solution generation capabilities, and improves the ability of electronic equipment to handle problems encountered in communication services on-site, thereby increasing the proportion of on-site personnel handling problems by themselves and reducing the workload of back-end personnel.

[0046] The electronic device of the present application can be implemented as a laptop computer, a tablet computer, a desktop computer, a mobile device (for example, a mobile phone, a portable music player, a personal digital assistant, a dedicated messaging device, a portable gaming device), an intelligent robot, or any other terminal with a screen display function, and can also be implemented as a server.

[0047] The present application embodiment provides a communication service processing method, which is applied to electronic equipment, such as Figure 1 As shown, the method includes:

[0048] Step 101: Acquire first information related to a communication service, where the first information includes multi-dimensional information obtained by performing intent recognition on a conversation related to the communication service.

[0049] In actual application, the electronic device obtains multi-dimensional information by performing intent recognition on conversations related to communication services (which can be multiple rounds of conversations). The multi-dimensional information includes multi-dimensional attributes of the communication device providing the communication service, such as the device's serial number (SN) and broadband account number.

[0050] Step 102: Acquire second information related to the communication service, where the second information includes the execution action of each executable task unit obtained by dividing the first information, and the access condition of each executable task unit.

[0051] In actual application, when the first information is obtained, the first information can be planned into smaller executable task units T1, T2, ..., TP, and each executable task unit is as small as possible. Each executable task unit has a corresponding access condition. Only when a certain access condition is met will the execution of the executable task unit corresponding to the access condition be triggered, where P is the total number of executable task units.

[0052] In actual application, the execution of a task is divided into multiple steps. The key business data extracted by the data analysis in the mth step is used as the access condition to determine the executable task unit in the m+1th step, where m is a positive integer.

[0053] Step 103: Acquire third information, where the third information includes analysis data generated by analyzing results obtained by completing execution actions of each executable task unit under the access conditions.

[0054] In practical application, the third information includes integrated information obtained by executing all executable task units after multiple rounds of dialogue and performing data analysis (also known as live network verification data).

[0055] Step 104: Input the analysis data into the large language model to obtain a processing strategy related to the communication service output by the large language model.

[0056] In practical applications, LLM refers to a deep learning model trained with a large amount of text data, which can generate natural language text or understand the meaning of language text. Large language models can handle a variety of natural language tasks, such as text classification, question answering, dialogue, etc., and are an important path to artificial intelligence.

[0057] In actual application, the analysis data is input into the large language model, and the large language model can call the fault case library for comparison and analysis to obtain the communication service-related processing strategy output by the large language model.

[0058] An embodiment of the present application provides a communication service processing method, the method comprising: obtaining first information related to the communication service, the first information comprising multi-dimensional information obtained by performing intent recognition on a conversation related to the communication service; obtaining second information related to the communication service, the second information comprising an execution action of each executable task unit obtained by dividing the first information, and an access condition for each executable task unit; obtaining third information, the third information comprising analyzing a result obtained by completing the execution action of each executable task unit under the access condition to generate analysis data; inputting the analysis data into a large language model to obtain a processing strategy related to the communication service output by the large language model.

[0059] In some embodiments, step 101 of acquiring first information related to a communication service includes:

[0060] Obtaining a target text obtained by combining the text of the n+1th round of conversation related to the communication service with the text of the previous n rounds of conversation, where n is a positive integer;

[0061] The target text is input into the large language model to obtain the first information output by the large language model.

[0062] In actual applications, a sliding window mechanism is used to process multiple rounds of dialogue. For the n+1th round of dialogue, when the large language model answers the questions in this round, it combines the text of the current question with the text of the previous n rounds of historical dialogue to obtain a new question text, namely the target text, which is then input into the large language model.

[0063] See also Figure 2 As shown, taking the use of a sliding window mechanism to process two adjacent rounds of conversations as an example, during the process of human-computer interaction, the user conducts multiple rounds of conversations with the electronic device, and new conversations are constantly generated. The sliding window 201 slides to the side indicated by the arrow 202 (for example, the right side) to obtain a new question text as the target text. The present application uses a sliding window mechanism to predict the current answer by splicing the previous conversation and the current conversation (ie, the current question) when processing multiple rounds of conversations. When the content of the conversation exceeds the window length, the earliest sentence will be discarded to ensure the continuity and relevance of the conversation and improve the accuracy of the first information obtained.

[0064] See also Figure 3 As shown, in an achievable communication service processing flow, the communication service processing method provided by the embodiment of the present application is described by taking three links as an example: the three links include a "clarifying the problem" link 301, a "planning and execution" link 302, and an "output result" link 303;

[0065] Among them, the "clarify the problem" link 301: LLM completes information collection through multiple rounds of dialogue with the user, identifies the user's true intention, and clarifies the problem that the user really wants to solve.

[0066] In the "clarify the problem" stage 301, the user asks questions on the spot, the electronic device recognizes the intention, and determines whether the information dimension is complete. Otherwise, the question is completed through multiple rounds of interaction between the user and the electronic device; if so, it enters the "planning and execution" stage 302.

[0067] Dialogue example 1 is shown in Table 1 below. In this example, LLM guides the user to input device information based on the scene recognition result.

[0068]

[0069] Table 1

[0070] Dialogue sample 2 is shown in Table 2 below. In this sample, LLM guides the user to input the problem scenario based on the recognition situation.

[0071]

[0072] Table 2

[0073] Through multiple rounds of dialogue with people, LLM completes information collection, identifies the user's true intention, and clarifies the problem that the user really wants to solve. In other words, the user's intention is obtained through multiple rounds of dialogue interaction to complete the problem and then analyze it. Through multiple rounds of dialogue, LLM stores the attributes of multiple dimensions of the current device in short-term memory, such as the device's SN, broadband account, etc.

[0074] See also Figure 4As shown in the figure, the user initiates a conversation (i.e., asks questions on the spot), and the content of the questions supports text, voice, pictures, etc. LLM cooperates with the image character recognition model capability and voice-to-text capability to perceive user input. LLM recognizes and integrates the information of text input, image input, and voice input, and obtains the following content: questions (including phenomena, categories, dislocation codes, etc.), equipment (including types, logos, etc.), and other information. Here, the expert experience in long-term memory (also known as domain knowledge, such as user questioning habits, case libraries, interface documents, equipment manuals, etc.) can also be combined to determine whether the information dimension provided by the current user is complete. If the information is incomplete, LLM generates guide words to guide the user to supplement the missing information in a new round of dialogue. Until LLM infers and recognizes complete question information from the user conversation context.

[0075] "Planning and Execution" link 302: includes task scheduling, task execution and data analysis; the actual production scenario of the planning and execution link is relatively complex and there may be multiple situations. For this situation, this link can also include user on-site operations and multiple rounds of interaction.

[0076] Among them, task orchestration refers to domain experts using multiple groups of natural languages ​​to describe the problem handling process under different user intentions, the application programming interface (API) and system involved in each executable task unit, and the judgment rules of the data returned by the system. Then the domain knowledge is imported into the LLM application through pre-training, fine-tuning or prompt word engineering.

[0077] Among them, task execution refers to the interaction between LLM and dozens of surrounding support systems to access the existing network data of multiple data sources, including data query, modification and deletion operations. And in task execution, permission control is introduced, first, permission control is performed on functional modules, and second, domain control is performed on data.

[0078] Data analysis refers to analyzing the result data returned by LLM calling dozens of surrounding support systems. Parse according to the interface definition specifications and extract key business data. The key business data here refers to the access conditions for executable task units in the process. After each subtask is completed, it is necessary to analyze the data generated by the subtask execution. Combined with the expert experience learned by LLM, it is necessary to guide the installation and maintenance personnel to conduct on-site inspections at certain nodes in the entire process. Through the guide words output by LLM, multiple rounds of dialogue with the installation and maintenance personnel are conducted to collect more information to assist in problem troubleshooting or processing. LLM puts the integrated information into the context and makes inferences about the next executable task. Until the task is completed.

[0079] In actual application, in the entire diagnosis and processing process, if the intermediate link requires manual on-site verification or operation, and the self-closed loop cannot be scheduled through the API, LLM accurately guides manual on-site inspection or operation of a certain device, and guides people to return the results of the verification or operation through dialogue. People can return this part of information through natural language text or on-site photos. LLM interacts with people through multiple rounds of dialogue and obtains the "entry conditions" of a certain "executable action A". LLM extracts this part of information from the dialogue, puts it in the context as an environmental parameter, and continues to think about the next operation according to the chain of thought.

[0080] It can be seen that the multiple rounds of interaction in this application are reflected in the "clarify the problem" link 301 and the "plan and execute" link 302. In the entire process of problem identification, diagnosis and processing, LLM introduces multiple rounds of dialogue. In the two key links of "clarify the problem" and "plan and execute", text output is used to guide users to give feedback, and the user's feedback is used as input for subsequent process processing.

[0081] This application improves the problem through multiple rounds of dialogue, understands the problem through intention recognition, automatically solves the problem through planning and execution, gives recommended solutions, and finally realizes a series of processes for automatic problem handling. This application can also write preset solutions into the prompt model (prompt) to improve the accuracy and usability of the solution recommendation function in the self-service problem handling device.

[0082] In some embodiments, step 102 of acquiring second information related to the communication service includes:

[0083] Inputting the first information into the large language model, obtaining a return result and related analysis information of the execution action of each executable task unit output by the large language model;

[0084] Obtaining interactive prompt information generated by the large language model based on the returned results and related analysis information, where the interactive prompt information is used to guide the target object to perform a target operation and return the result of the target operation, where the target operation includes on-site troubleshooting and / or operating a target device;

[0085] The business data extracted by the large language model based on the result of the target operation is obtained as the entry condition for each executable task unit.

[0086] In actual application, when the first information is obtained (for example, the problem Q that the user wants to solve), LLM plans Q into smaller executable task units T1, T2, ..., TP. Each executable task unit is as small as possible, and the execution action A of each executable task unit and the entry conditions of each executable task unit under problem Q are inferred from the expert experience in long-term memory.

[0087] Exemplarily, Table 3 provides executable task units, access conditions, and execution actions planned for user question Q (ie, user intention Q):

[0088]

[0089] Table 3

[0090] In some scenarios, in the "Planning and Execution" link 302, the actual production scenario is relatively complex and there may be multiple situations. Possibility one: There is a situation where the status of the equipment is not connected to the network, and it is impossible to query from the information management system through automated means. To understand the status of these devices, it is only possible through manual on-site inspections. Including: 1) Failures of some non-intelligent facilities, for example, the existing network line is broken, the interface is not plugged in tightly, power failure, etc. 2) The equipment fails and is disconnected from the upstream equipment. The equipment status cannot be obtained, and deeper causes cannot be diagnosed. For example, the optical modem crashes or fails, and is disconnected from the upstream OLT. Possibility two: Some operations need to be performed by the user on-site and cannot be handled remotely. After guiding the user to complete the operation, the diagnosis or processing process still needs to continue. For example, the user needs to change the password on-site, and then the user is kicked offline. Here, taking Table 4 as an example, the dialogue in the "Planning and Execution" link is illustrated:

[0091]

[0092]

[0093] Table 4

[0094] It should be noted that when calling the optical modem's self-discovery tool, it is found to be an array. For example, the corresponding values ​​in R / F / S / P are different.

[0095] From the above content, it can be seen that this application introduces multiple rounds of dialogue into the self-service problem handling process, and supports people to supplement the on-site situation through text conversations, photos, screenshots, etc. In the process of troubleshooting existing network problems, the status of devices that are not connected to the network and operations that require manual intervention can be taken into consideration.

[0096] In some embodiments, step 104 inputs the analysis data into the large language model to obtain a communication service-related processing strategy output by the large language model, including:

[0097] The analysis data is input into the large language model to obtain a first strategy corresponding to each executable task unit output by the large language model, wherein the first strategy includes first text information that conforms to the communication business logic; and the processing strategy includes the first strategy.

[0098] In actual application, the first strategy is called the abbreviation scheme. For example, for a short text, keyword analysis and identification are performed from the existing case library combined with expert experience to screen out unique corpus that conforms to the business logic. LLM is trained based on the corpus, and model parameters are optimized with reference to business cases, and the number of model layers and parameter settings are modified. Under the premise that the previous process is accurate, LLM combines the existing network verification data to infer the abbreviation scheme with an accuracy rate of about 98%.

[0099] This application provides specific recommendations based on the data verified on the current network and the failure cases in the large model trained / fine-tuned by experts in the previous field, according to the different states of the equipment. In order to avoid the large model providing a correct reference with little reference value after probability calculation, and to improve the accuracy of the recommended solution, the large model can provide more constructive and practical suggestions for specific production scenarios. The solution is divided into two parts: abbreviated solution and solution details.

[0100] In some embodiments, before inputting the analysis data into the large language model to obtain the first strategy corresponding to each executable task unit output by the large language model, the communication service processing method further includes:

[0101] The autoregressive language model is trained and fine-tuned based on the content target consisting of each sample executable task unit and sample strategy to obtain a large language model.

[0102] In actual application, the content is the result of executing each executable task unit, the target pair is the abbreviated solution, and the content and the target pair constitute a content-target pair.

[0103] In a feasible scenario, the steps for handling business processes that cannot access the Internet are as follows (some steps are omitted for ease of reading, and key steps are retained to reflect the process of obtaining the recommended solution):

[0104] Step 1: Check the online status of the account and call the "Query Account Online Time" tool;

[0105] If the "online time" returned in the result is empty, the account is not online, and execute step 2;

[0106] If the "online time" returned in the result is not empty, the account is online, and step 3 is executed;

[0107] Step 2: Check the account authentication status, call the "Account Online and Offline Records" tool, and view the most recent record;

[0108] If the "Package Type" returned in the result is not "Authentication Failed", go to step 12;

[0109] If the content of the "Package Type" returned in the result is "Authentication Failed" and the authentication result is "Binding Information Error", execute step 4;

[0110] If the content of the "Package Type" returned in the result is "Authentication Failed" and the authentication result is "User Online Authentication Rejected", execute step 5;

[0111] If the content of the "Package Type" returned in the result is "Authentication Failed" and the authentication result is "Wrong Password", execute step 6;

[0112] If the content of the "Package Type" returned in the result is "Authentication Failed" and the authentication result is "User Locked", execute step 7;

[0113] If the content of the "Package Type" returned in the result is "Authentication Failed" and the authentication result is "User does not exist", execute step 8;

[0114] If the content of the "Package Type" returned in the result is "Authentication Failed", and the authentication result is not any of ["Binding Information Error", "User Online Authentication Rejected", "Password Error", "User Locked", "User Does Not Exist"], execute step 35;

[0115] Step 3: Analyze the account offline records, call the "Account Online and Offline Records" tool, and view the most recent record;

[0116] If the "package type" returned in the result does not belong to "authentication failure" or "billing end", execute step 24;

[0117] If the "Package Type" returned in the result is "Authentication Failed" and the authentication result is "User Online Authentication Rejected", execute step 5;

[0118] If the "Package Type" returned by the result is "Accounting Ended" and the authentication result is "User Offline (Session Timed Out)", execute step 23;

[0119] Step 24: Determine whether the online disconnection is frequent and call the "account online and offline record" tool;

[0120] If the "frequent disconnection" field in the first dictionary of the returned data list is True, execute step 28;

[0121] In the first dictionary of the data list returned as a result, the "frequent disconnection" field is False, and step 30 is executed;

[0122] Step 30: Determine whether the optical power is abnormal and call the "Optical Modem and OLT Light Receiving" tool;

[0123] If the tool result returns that "ONU received optical power" is "normal" and the tool result returns that "OLT received ONU optical power" is "normal", execute step 31;

[0124] If the tool result returns that "ONU received optical power" is "normal" and the tool result returns that "OLT received ONU optical power" is "strong light", execute step 36;

[0125] If the tool result returns that "ONU received optical power" is "normal" and the tool result returns that "OLT received ONU optical power" is "weak light", execute step 37;

[0126] step36:

[0127] Diagnostic results: OLT receiving ONU optical power: strong light; ONU receiving optical power: normal;

[0128] Recommended solution: Please make light power and strong light rectification

[0129] The process ends.

[0130] The above is the entire business logic.

[0131] Sample output:

[0132] Question is an input question that must be answered.

[0133] Please extract the original complete fragment of the corresponding step from the business logic based on the step number given in the question or the step number to be executed mentioned in the question. Please ignore the other content in the question and do not modify any content of the extracted step fragment.

[0134] Example 1:

[0135] Question:

[0136] step1.

[0137] Answer:

[0138] Step 1: Check the online status of the account, determine whether the online time is empty, and call the "Query Account Online Time" tool;

[0139] If the "online time" returned in the result is empty, the account is not online, and execute step 2;

[0140] If the "online time" returned in the result is not empty, the account is online, and step 3 is executed;

[0141] Example 2:

[0142] Question:

[0143] Thinking process: In the results returned by the account online time query tool, the "Online Time" field is not empty.

[0144] Select the number of the next step: step3.

[0145] Answer:

[0146] Step 3: Analyze the account offline records, call the "Account Online and Offline Records" tool, and view the most recent record;

[0147] If the "Package Type" returned in the result is "Authentication Failed" and the authentication result is "User Online Authentication Rejected", execute step 5;

[0148] If the "Package Type" returned in the result is "Accounting Ended" and the authentication result is "User Offline (Session Timed Out)", execute step 23;

[0149] If the "Package Type" returned in the result is not "Authentication Failed" or "Accounting Ended", execute step 24;

[0150] It should be noted that: based on the step number given in the question or the step number to be executed mentioned in the question, please extract the original complete fragment of the corresponding step from the business logic. Please ignore other content in the question and do not modify any content of the extracted step fragment.

[0151] Exemplary, current issues:

[0152] Question:{question}

[0153] Answer:

[0154] """

[0155] During the above process, the large language model is trained using business-specific data with an efficient parameter fine-tuning solution (such as the lora method). During the fine-tuning process, a set of parameters, i.e., the total number of parameters Φ of the model, is found to maximize formula 1:

[0156]

[0157] And add noise ε to the loss function to obtain the following formula 2, which enhances the generalization of model reasoning:

[0158]

[0159] Adding a multilayer perceptron (MLP) layer to the last layer of the model yields Formula 3 for classification:

[0160]

[0161] Among them, P Φ (y|x,y) represents the pre-trained autoregressive language model, Z = {(x i ,y i )}i=1,…,N represents the content-target pair of each downstream task, the content is the result of the execution of each executable task unit, and the target pair is an abbreviated solution, such as {"Please make optical power and strong light rectification", "Please call 10086 to change the password", "Please close the port",..}.

[0162] In some embodiments, after the analysis data is input into the large language model to obtain the first strategy corresponding to each executable task unit output by the large language model, the communication service processing method further includes:

[0163] The second strategy is generated based on the first strategy, and the second strategy includes one or more of the following: second text information, picture information, and document information that conform to the communication business logic, and the second text information is obtained by text expansion of the first text information.

[0164] In actual application, the second strategy is called a detailed plan, which is based on the abbreviated plan and describes the specific operation steps in detail, or explains certain concepts in more detail, or provides more reference information. The detailed plan can be output in the format of text, pictures, tables, etc. A large number of case libraries and equipment documents are injected into the large language model through pre-training or external knowledge base. The system combines the existing network verification data with the abbreviated plan again, and infers the detailed plan through the large language model.

[0165] See also Figure 5 As shown, taking the abbreviated plan of "Please make optical power rectification" as an example, based on the abbreviated plan, the question is expanded, the knowledge base text is searched, and finally the image and text are associated to obtain the output detailed plan, that is, the detailed measures on how to rectify the strong light are output.

[0166] See also Figure 6 As shown in the figure, in a feasible business processing scenario, Figure 6Taking the interface shown as an example, the process from initiating diagnosis to presenting results is explained: the large language model processes the problem of user input, and the first interface 601 displays: We are diagnosing the problem that the user with broadband account A cannot access the Internet, please wait. Intelligent diagnosis in progress... The second interface 602 displays: Diagnosis result: Account binding information error; the information is as follows: 4. Account online status: offline; 5. Authentication information: authentication identification, authentication time, authentication information, reason for authentication failure (the specific content of each information is not shown); 6. Optical modem status: offline. The third interface 603 displays: Recommended solution: Port is closed, please handle it according to the local customs. The fourth interface 604 displays: More (display solution details), displayed in text, for example, describing the specific operation steps; displayed in pictures, for example, the processing requirements of different cities; displayed in documents, for example, operating specifications. In this embodiment, the large language model returns the result of intelligent processing of the question to the user. The question processing result is divided into several parts: guide words, data analysis results, data query results, abbreviated solutions, and solution details. It should be noted that the above first interface 601, second interface 602, third interface 603 and fourth interface 604 can be defined and displayed in combination according to actual conditions, and this application does not make specific limitations on this.

[0167] In actual applications, after the conversation ends, experts in the field can evaluate the intelligent processing of the large language model and continuously supplement expert experience. They can also use the domain knowledge feedback from experts to perform post-processing and correction on the large language model.

[0168] See also Figure 7 As shown, the business support system can be distributed and deployed on multiple servers, and the business server 701 of the business support system is responsible for the unified control and scheduling search engine. The component server 702 stores embedded information such as prior rules or knowledge, API information, and library tables, which is used to use the vector library to search for the most similar knowledge items as background knowledge supplements for the model input before the model is called. Among them, the component server 702 stores short-term memory in an in-memory database and stores long-term memory in a structured database. The business server 701 can communicate with different information management platforms through a unified API. The model server 703 can deploy one or more large language models after performing resource evaluation as needed. The entire system can be intelligently deployed based on cloud and container technologies, with high portability and easy to promote continuously.

[0169] The business support system improves problems through multiple rounds of dialogue, understands problems through intent recognition, automatically solves problems through planning and execution, and gives recommended solutions, ultimately realizing the process of automatic problem handling. In the troubleshooting of existing network problems, it can take into account the status of devices that are not connected to the network and situations that require manual intervention.

[0170] The present application also provides a communication service processing device, such as Figure 8 As shown, the communication service processing device includes:

[0171] An acquisition unit 801 is configured to acquire first information related to a communication service, where the first information includes multi-dimensional information obtained by performing intent recognition on a conversation related to the communication service;

[0172] The acquisition unit 801 is used to acquire second information related to the communication service, where the second information includes the execution action of each executable task unit obtained by dividing the first information, and the access condition of each executable task unit;

[0173] An acquisition unit 801 is used to acquire third information, where the third information includes analysis data generated by analyzing the results of executing actions of each executable task unit under the access conditions;

[0174] The processing unit 802 is used to input the analysis data into the large language model to obtain a processing strategy related to the communication service output by the large language model.

[0175] In some embodiments, the acquisition unit 801 is used to acquire a target text obtained by combining the text of the n+1th round of dialogue related to the communication service with the text of the previous n rounds of dialogue, where n is a positive integer;

[0176] The processing unit 802 is used to input the target text into the large language model to obtain the first information output by the large language model.

[0177] In some embodiments, the acquisition unit 801 is used to acquire a target text obtained by combining the text of the n+1th round of dialogue related to the communication service with the text of the previous n rounds of dialogue, where n is a positive integer;

[0178] The processing unit 802 is used to input the target text into the large language model to obtain the first information output by the large language model.

[0179] In some embodiments, the processing unit 802 is used to input the first information into the large language model to obtain the return result and related analysis information of the execution action of each executable task unit output by the large language model;

[0180] An acquisition unit 801 is used to acquire interactive prompt information generated by the large language model based on the returned result and related analysis information, where the interactive prompt information is used to guide the target object to perform a target operation and return a result of the target operation, where the target operation includes on-site troubleshooting and / or operating a target device;

[0181] The acquisition unit 801 is used to acquire the business data extracted by the large language model based on the result of the target operation as the access condition of each executable task unit.

[0182] In some embodiments, the processing unit 802 is used to input the analysis data into the large language model to obtain a first strategy corresponding to each executable task unit output by the large language model, wherein the first strategy includes first text information that conforms to the communication business logic; the processing strategy includes the first strategy.

[0183] In some embodiments, the processing unit 802 is used to generate a second strategy based on the first strategy, and the second strategy includes one or more of the following: second text information, picture information, and document information that conforms to the communication business logic, and the second text information is obtained by text expansion of the first text information.

[0184] In some embodiments, the processing unit 802 is used to train the autoregressive language model and fine-tune the autoregressive language model based on the content target composed of each sample executable task unit and the sample strategy to obtain a large language model.

[0185] In order to implement the communication service processing method of the embodiment of the present application, the embodiment of the present application also provides an electronic device, such as Fig. 9 As shown, the electronic device 900 includes: a communication interface 901 and a processor 902; wherein,

[0186] Communication interface 901;

[0187] Processor 902, connected to the communication interface 901, and configured to execute the above-mentioned method for information service processing when running a computer program;

[0188] The memory 903 stores computer programs that can be executed on the processor 902 .

[0189] The communication interface is used to: obtain first information related to the communication service, the first information including multi-dimensional information obtained by performing intent recognition on a conversation related to the communication service;

[0190] Acquire second information related to the communication service, the second information including an execution action of each executable task unit obtained by dividing the first information, and an access condition of each executable task unit;

[0191] Acquiring third information, the third information including analysis data generated by analyzing a result of completing an execution action of each executable task unit under the access condition;

[0192] Processor 902 is used to: input the analysis data into the large language model to obtain a processing strategy related to the communication service output by the large language model.

[0193] In some embodiments, the communication interface 901 is used to: obtain a target text obtained by combining the text of the n+1th round of dialogue related to the communication service with the text of the previous n rounds of dialogue, where n is a positive integer;

[0194] Processor 902 is used to: input the target text into the large language model to obtain first information output by the large language model.

[0195] In some embodiments, the processor 902 is configured to: input the first information into the large language model to obtain the return result and related analysis information of the execution action of each executable task unit output by the large language model;

[0196] Communication interface 901, used to: obtain interactive prompt information generated by the large language model based on the returned results and related analysis information, the interactive prompt information is used to guide the target object to perform the target operation and return the result of the target operation, the target operation includes on-site troubleshooting and / or operating the target device;

[0197] The communication interface 901 is used to obtain the business data extracted by the large language model based on the result of the target operation as the access condition of each executable task unit.

[0198] In some embodiments, the processor 902 is used to: input the analysis data into the large language model to obtain a first strategy corresponding to each executable task unit output by the large language model, the first strategy including first text information that conforms to the communication business logic; the processing strategy includes the first strategy.

[0199] In some embodiments, the processor 902 is used to: generate a second strategy based on the first strategy, the second strategy includes one or more of the following: second text information, picture information, document information that conforms to the communication business logic, and the second text information is obtained by text expansion of the first text information.

[0200] In some embodiments, the processor 902 is used to: train an autoregressive language model, and fine-tune the autoregressive language model based on a content target composed of each sample executable task unit and a sample strategy to obtain a large language model.

[0201] The method disclosed in the above embodiment of the present application can be applied to the processor 902, or implemented by the processor 902. The processor 902 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the hardware integrated logic circuit or software instructions in the processor 902. The above processor 902 may be a general-purpose processor, a digital signal processor (Digital Signal Processor, DSP), or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. The processor 902 can implement or execute the methods, steps and logic block diagrams disclosed in the embodiments of the present application. The general-purpose processor may be a microprocessor or any conventional processor, etc. In combination with the steps of the method disclosed in the embodiment of the present application, it can be directly embodied as a hardware decoding processor to execute, or it can be executed by a combination of hardware and software modules in the decoding processor. The software module can be located in a storage medium, which is located in the memory 903, and the processor 902 reads the information in the memory 903 and completes the steps of the above method in combination with its hardware.

[0202] In an exemplary embodiment, the electronic device 900 can be implemented by one or more application specific integrated circuits (ASIC), DSP, programmable logic device (PLD), complex programmable logic device (CPLD), field programmable gate array (FPGA), general processor, controller, microcontroller (MCU), microprocessor, or other electronic components to execute the aforementioned method.

[0203] It can be understood that the memory of the embodiment of the present application can be a volatile memory or a non-volatile memory, and can also include both volatile and non-volatile memories. Among them, the non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), a magnetic random access memory (FRAM), a flash memory (FlashMemory), a magnetic surface memory, an optical disc, or a compact disc read-only memory (CD-ROM); the magnetic surface memory can be a disk memory or a tape memory. The volatile memory can be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as Static Random Access Memory (SRAM), Synchronous Static Random Access Memory (SSRAM), Dynamic Random Access Memory (DRAM), Synchronous Dynamic Random Access Memory (SDRAM), Double Data Rate Synchronous Dynamic Random Access Memory (DDRSDRAM), Enhanced Synchronous Dynamic Random Access Memory (ESDRAM), Sync Link Dynamic Random Access Memory (SLDRAM), and Direct Rambus Random Access Memory (DRRAM).The memories described in the embodiments of the present application are intended to include, but are not limited to, these and any other suitable types of memories.

[0204] In an exemplary embodiment, the present application also provides a storage medium, namely a computer storage medium, specifically a computer-readable storage medium, for example, a memory 903 storing a computer program, and the computer program can be executed by a processor 902 of an electronic device 900 to complete the steps of the aforementioned method. The computer-readable storage medium can be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, Flash Memory, magnetic surface storage, optical disk, or CD-ROM.

[0205] In an exemplary embodiment, the embodiment of the present application further provides a computer product, including a computer program, which can be executed by the processor 902 of the electronic device 900 to complete the steps of the aforementioned method.

[0206] It should be noted that: "first", "second", etc. are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0207] In addition, the technical solutions described in the embodiments of the present application can be combined arbitrarily without conflict.

[0208] The above description is only a preferred embodiment of the present application and is not intended to limit the protection scope of the present application.

Claims

1. A communication service processing method, characterized in that: The method comprises: Acquiring first information related to a communication service, where the first information includes multi-dimensional information obtained by performing intent recognition on a conversation related to the communication service; Acquire second information related to the communication service, the second information including an execution action of each executable task unit obtained by dividing the first information, and an access condition of each executable task unit; Acquire third information, the third information including analysis data generated by analyzing a result obtained by completing an execution action of each executable task unit under the access condition; The analysis data is input into a large language model to obtain a processing strategy related to the communication service output by the large language model.

2. The method according to claim 1, characterized in that The acquiring of first information related to the communication service includes: Obtaining a target text obtained by combining the text of the n+1th round of dialogue related to the communication service with the text of the first n rounds of dialogue, where n is a positive integer; The target text is input into the large language model to obtain the first information output by the large language model.

3. The method according to claim 1, characterized in that The acquiring the second information related to the communication service includes: Inputting the first information into the large language model, obtaining a return result and related analysis information of the execution action of each executable task unit output by the large language model; Obtaining interactive prompt information generated by the large language model based on the returned result and the relevant analysis information, wherein the interactive prompt information is used to guide the target object to perform a target operation and return a result of the target operation, wherein the target operation includes on-site troubleshooting and / or operating a target device; Business data extracted by the large language model based on the result of the target operation is obtained as the access condition of each executable task unit.

4. The method according to claim 1, characterized in that: The step of inputting the analysis data into a large language model to obtain a processing strategy related to the communication service output by the large language model includes: The analysis data is input into a large language model to obtain a first strategy corresponding to each executable task unit output by the large language model, wherein the first strategy includes first text information that conforms to communication business logic; and the processing strategy includes the first strategy.

5. The method according to claim 4, characterized in that After inputting the analysis data into the large language model to obtain the first strategy corresponding to each executable task unit output by the large language model, the method further includes: A second strategy is generated based on the first strategy, wherein the second strategy includes one or more of the following: second text information, picture information, and document information that conform to the communication service logic, wherein the second text information is obtained by text expansion of the first text information.

6. The method according to claim 4, characterized in that Before inputting the analysis data into the large language model to obtain the first strategy corresponding to each executable task unit output by the large language model, the method further includes: The autoregressive language model is trained, and the autoregressive language model is fine-tuned based on the content target composed of each sample executable task unit and the sample strategy to obtain the large language model.

7. A communication service processing device, characterized in that: include: an acquisition unit, configured to acquire first information related to a communication service, wherein the first information includes multi-dimensional information obtained by performing intent recognition on a conversation related to the communication service; an acquiring unit, configured to acquire second information related to the communication service, wherein the second information includes an execution action of each executable task unit obtained by dividing the first information, and an access condition of each executable task unit; an acquisition unit, configured to acquire third information, wherein the third information includes analysis data generated by analyzing a result of completing an execution action of each executable task unit under the access condition; The processing unit is used to input the analysis data into a large language model to obtain a processing strategy related to the communication service output by the large language model.

8. An electronic device, characterized in that: include: A processor and a memory for storing a computer program capable of running on the processor; wherein, The processor is used to execute the steps of the method according to any one of claims 1 to 6 when running the computer program.

9. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

10. A computer product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images