Intent information processing method, electronic device and readable storage medium

Abstract

The present application belongs to the technical field of computers. Disclosed are an intent information processing method, an electronic device and a readable storage medium. The method comprises: by means of analyzing input intent information, determining a primary functional interface matching the intent information; acquiring primary parameters corresponding to the primary functional interface, the primary parameters being primary execution parameters of the primary functional interface; in response to a determination of the number of the primary parameters being greater than 0, acquiring a parameter value of each primary parameter by means of analyzing the intent information; on the basis of the parameter value of each primary parameter, determining an execution sequence corresponding to the intent information, the execution sequence comprising all functional interfaces used for processing the intent information; and successively calling the functional interfaces in the execution sequence, so as to obtain a processing result matching the intent information.

Description

Intent information processing methods, electronic devices and readable storage media

[0001] Cross-referencing

[0002] This application claims priority to Chinese Patent Application No. 202411489477.0, filed on October 24, 2024, entitled "Intent Information Processing Method, Electronic Device and Readable Storage Medium", the entire contents of which are incorporated herein by reference. Technical Field

[0003] This application relates to the field of computer technology, and in particular to an intent information processing method, an electronic device, and a readable storage medium. Background Technology

[0004] With the rapid development of information technology and the significant increase in network scale, the coexistence of multiple standards such as 3G, 4G, and 5G has led to considerable uncertainty in services and traffic, making network operation and maintenance increasingly complex. Currently, network operation and maintenance largely relies on the experience and skills of maintenance personnel. However, in the current network environment, technology is updating rapidly, and traditional network operation and maintenance methods can no longer meet the requirements for accuracy in the face of complex and ever-changing network architectures and diverse user needs. Summary of the Invention

[0005] This application provides an intent information processing method, an electronic device, and a readable storage medium.

[0006] In a first aspect, an intent information processing method is provided, the method comprising: determining a main functional interface matching the intent information by parsing input intent information; obtaining main parameters corresponding to the main functional interface, wherein the main parameters are main execution parameters of the main functional interface; in response to a determination that the number of main parameters is greater than 0, obtaining a parameter value of each main parameter by parsing the intent information; determining an execution sequence corresponding to the intent information based on the parameter value of each main parameter, wherein the execution sequence includes all functional interfaces used to process the intent information; and sequentially calling the functional interfaces in the execution sequence to obtain a processing result matching the intent information.

[0007] In a second aspect, an electronic device is provided, including a processor and a memory, the memory storing a program or instructions executable on the processor, the program or instructions, when executed by the processor, implementing the method as described in the first aspect.

[0008] Thirdly, a readable storage medium is provided, wherein at least one computer program is stored therein, which, when loaded and executed by a processor, implements the method described in the first aspect.

[0009] Fourthly, a computer program product is provided, the computer program product comprising at least one computer program that, when loaded and executed by a processor, implements the method described in the first aspect.

[0010] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0011] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0012] Figure 1 shows a schematic flowchart of an intent information processing method provided in an exemplary embodiment of this application;

[0013] Figure 2 shows a schematic diagram of a parameter spectrum provided in an exemplary embodiment of this application;

[0014] Figure 3 shows a schematic diagram of a parameter spectrum provided in an exemplary embodiment of this application;

[0015] Figure 4 shows a schematic diagram of a parameter spectrum provided in an exemplary embodiment of this application;

[0016] Figure 5 shows a schematic diagram of a task-oriented dialogue operation and maintenance system based on a large model provided in an exemplary embodiment of this application;

[0017] Figure 6 illustrates an interactive schematic diagram of a task-oriented dialogue operation and maintenance system based on a large model provided in an exemplary embodiment of this application;

[0018] Figure 7 illustrates an interactive schematic diagram of a task-oriented dialogue operation and maintenance system based on a large model provided in an exemplary embodiment of this application;

[0019] Figure 8 shows a flowchart of an intent information processing method provided in an exemplary embodiment of this application;

[0020] Figure 9 shows a schematic flowchart of an intent information processing method provided in an exemplary embodiment of this application;

[0021] Figure 10 shows a flowchart of an intent information processing method provided in an exemplary embodiment of this application;

[0022] Figure 11 shows a schematic flowchart of an intent information processing method provided in an exemplary embodiment of this application;

[0023] Figure 12 shows a schematic diagram of the structure of an electronic device provided in an exemplary embodiment of this application. Detailed Implementation

[0024] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0025] Figure 1 illustrates a flowchart of an intent information processing method provided in an exemplary embodiment of this application. This method can be executed by an electronic device, which may include a terminal device. In other words, the method can be executed by software or hardware installed on the electronic device, and the method may include the following steps:

[0026] S110: By parsing the input intent information, determine the main function interface that matches the intent information.

[0027] The main function interface is used to generate results that match the intent information. This main function interface may include other modules or interfaces such as application programming interfaces (APIs) and execution modules used to generate results that match the intent information. In other words, by calling this main function interface, results that match the intent information can be obtained.

[0028] Understandably, to address complex operational challenges, this application proposes a "conversational" operational solution. Users can input intent information using natural language through a dialog box. However, the user's input may be ambiguous or imprecise. By parsing the input intent information, the core of the user's intent can be identified. Based on this core, the main functional interface is determined from multiple functional interfaces. This allows for rapid focus on executing tasks at the main functional interface, thereby reducing unnecessary resource consumption. Optionally, the intent information can be audio, video, or text.

[0029] In one implementation, S110 may include the following steps:

[0030] S112: Obtain a functional characteristic classification description file that matches the intent information. The functional characteristic classification description file includes at least one functional interface, and each functional interface corresponds to a different function.

[0031] In this step, the functional feature classification description file includes at least one functional interface. These functional interfaces are used to solve the same type of business problem. That is, each functional feature classification description file corresponds to a different type of business problem, and each functional interface under each functional feature classification description file corresponds to a different function. The functional feature classification description file may include the following fields: name, description, and semantic_apis. Here, name represents the name of the functional feature classification description file, description represents the functional description of the functional feature classification description file, and semantic_apis represents the functional interfaces included in the functional feature classification description file, as exemplified in Table 1.

[0032] Each functional interface also corresponds to a functional description file, which may include the following fields: name, version, description, input_parameters, and entity. Here, name represents the name of the functional interface; version indicates the version information of the functional interface, such as V1, V2, V3, etc.; description represents the functional description of the functional interface; input_parameters represents the semantic description of the input parameters corresponding to the functional interface, and may include the following attributes: parameter name, parameter type, parameter description information, whether it is a required parameter, etc.; entity represents the entity definition of the functional interface, and may include the following attributes: functional interface protocol type, operation type, address, parameter entity definition, etc., as exemplified in Table 2.

[0033] Table 1

[0034] Table 2

[0035] In another implementation, obtaining the functional characteristic classification description file that matches the intent information includes: generating a first prompt word based on the intent information, and providing the first prompt word to the target large model to obtain the functional characteristic classification description file that matches the intent information.

[0036] In other words, after obtaining the intent information, a first prompt word is generated based on this intent information, where the first prompt word is the prompt instruction of the target large language model. Then, the first prompt word is input into the target large language model to obtain a functional feature classification description file that matches the intent information.

[0037] Optionally, the first prompt word generated based on this intent information can be generated based on a pre-set first input template. For example, the first input template could be: "The current task is to select an appropriate skill based on the user's input intent '…'. Please select from the following skills: 'A' or 'B'." Then the generated first prompt word could be: "The current task is to select an appropriate skill based on the user's input intent 'Base station xxxx cannot access the internet, please diagnose.' Please select from the following skills: 'diagnose, this skill is used to diagnose and analyze faults existing in the network' or 'monitoring, this skill is used to monitor network traffic, device status, and performance parameters'."

[0038] S114: Determine the main functional interface that matches the intent information from at least one of the functional interfaces.

[0039] In this step, after determining the functional characteristic classification description file that matches the intent information, at least one functional interface under the functional characteristic classification description file can be obtained. Then, the functional interface that matches the intent information is selected from the at least one functional interface as the main functional interface.

[0040] In one implementation, S114 may include: generating a second prompt word based on the intent information and the at least one functional interface, and providing the second prompt word to the target large model to obtain the main functional interface that matches the intent information.

[0041] In other words, after obtaining at least one functional interface under the functional feature classification description file that matches the intent information, a second prompt word is generated based on the intent information and the at least one functional interface, wherein the second prompt word is the prompt instruction of the target large language model. Then, the second prompt word is input into the target large language model to obtain the main functional interface that matches the intent information.

[0042] Optionally, the generation of the second prompt word based on the intent information and the at least one functional interface can be based on a pre-set second input template. For example, the second input template could be: "The current task is to select the main functional interface based on the user's input intent '…'. Please select from the following functional interfaces: 'A' or 'B'." Then the generated second prompt word could be: "The current task is to select the main functional interface based on the user's input intent 'Base station xxxx cannot access the internet, please diagnose'. Please select from the following functional interfaces: 'siteDiagnose, this functional interface is responsible for base station fault diagnosis, diagnosing the base station based on a clear base station object and phenomenon', or 'lineServiceDiagnose, this functional interface is responsible for leased line fault diagnosis'.

[0043] In S112 and S114 above, a functional characteristic classification description file matching the intent information is first obtained from at least one functional characteristic classification description file. Then, the main functional interface is determined from at least one functional interface included in the matched functional characteristic classification description file. This avoids directly matching the main functional interface from multiple functional interfaces. Furthermore, in each processing step of the target large model, if the number of functional interfaces exceeds the target large model's token limit, the target large model may truncate some functional interfaces, resulting in the inability to read and process all functional interfaces, thus reducing the accuracy of matching. Therefore, classifying multiple functional interfaces, first obtaining the functional characteristic classification description file matching the intent information, and then searching for the matching main functional interface under that functional characteristic classification description file, improves the efficiency and accuracy of determining the main functional interface through two-level matching. In addition, the target large model is introduced in the process of determining the main functional interface. Utilizing the semantic understanding capability and generalization of the target large model further improves the accuracy of intent understanding and main functional interface determination, thereby improving the precision of network operation and maintenance.

[0044] S120: Obtain the main parameters corresponding to the main function interface.

[0045] The main parameter is the main execution parameter of the main function interface.

[0046] It is understandable that the main execution parameters refer to the parameters required to call the main function interface. To call the main function interface to generate a result matching the intent information, the main parameters corresponding to the main function interface must first be determined. Only then can the main function interface generate a result matching the intent information based on the values ​​of the main parameters. For example, in the case of intent information "Base station xxxx cannot access the internet, please diagnose," the main function interface can be determined to be siteDiagnose. As shown in Table 2 above, the main parameters corresponding to siteDiagnose include siteIP (base station IP) and faultType (fault type).

[0047] In this step, by obtaining the main parameters corresponding to the main function interface, the main function interface can avoid processing unnecessary data, thereby speeding up the processing efficiency.

[0048] S130: In response to the determination that the number of main parameters is greater than 0, the parameter value of each main parameter is obtained by parsing the intent information.

[0049] In another implementation, S130 may include: generating a third prompt word based on the intent information and each of the main parameters, and providing the third prompt word to the target large model to obtain the parameter value of each of the main parameters.

[0050] In other words, after determining the main parameters, it is necessary to obtain the parameter values ​​corresponding to the main parameters from the intent information. Therefore, a third prompt word can be generated based on the intent information and each of the main parameters. The third prompt word can be the prompt instruction of the target large language model. Then, the third prompt word is input into the target large language model to obtain the parameter value of each of the main parameters.

[0051] Optionally, the third prompt word generated based on the intent information and the main parameters can be generated based on a pre-set third input template. For example, the third input template can be: the current task is to parse the following parameters "E" and "F" from the user's input intent "...". Then the generated first prompt word can be: the current task is to parse the following parameters from the user's input intent "Base station xxxx cannot access the internet, please diagnose": "siteIP: represents the base station IP address, in IP address format", "faultType: describes the base station fault type, with possible values ​​1 (interruption, unable to access the internet), 2 (deterioration, lag, packet loss)". Then, the target model outputs the parameter values ​​of the main parameters siteIP and faultType based on this third prompt word. If the parameter value corresponding to a certain main parameter cannot be obtained from the intent information, then the parameter value of that main parameter is NULL. For example, the parameter value of siteIP is xxxx, and the parameter value of faultType is 1.

[0052] S140: Based on the parameter value of each of the main parameters, determine the execution sequence corresponding to the intent information.

[0053] The execution sequence includes all functional interfaces used to process the intent information.

[0054] It is understood that by parsing the intent information, the parameter value of each main parameter is obtained. If the value corresponding to a certain main parameter cannot be obtained from the intent information, the parameter value corresponding to that main parameter is empty, which can be represented by NULL. If the parameter value corresponding to that main parameter is empty, the main function interface will not be called correctly due to the lack of parameters, thus resulting in the inability to obtain a processing result matching the intent information. Therefore, before calling the main function interface, it is necessary to obtain the function interface used to determine the parameter value of the main parameter. Therefore, in this embodiment of the application, the execution sequence can be dynamically adjusted according to the determination of the parameter value of each main parameter, that is, the function interface used to process the intent information can be dynamically determined, thereby improving the accuracy of the returned processing result in subsequent execution.

[0055] S150: The function interfaces in the execution sequence are called sequentially to obtain the processing result that matches the intent information.

[0056] Understandably, after determining the execution sequence, each functional interface is called sequentially to execute the corresponding task, ultimately obtaining a processing result that matches the intent information. For example, this processing result could be: Fault location: Port x; Fault cause: Optical module damaged; Fault repair suggestion: Replace the optical module.

[0057] It should be noted that the order in which the functional interfaces are called in the execution sequence is determined based on parameter dependencies. If the execution parameters of a certain functional interface are not yet determined, the functional interface used to obtain those parameters is called first. This ensures that each functional interface executes its task only when the required parameters are met. If the parameters required by a functional interface are incomplete or incorrect, the execution results may be inaccurate, thereby reducing the accuracy of network operation and maintenance. For example, assuming the intent information is "Base station xxxx cannot access the internet, please diagnose", and the main parameters siteIP and faultType of the corresponding main function interface siteDiagnose are both determined (i.e., siteIP = xxxx, faultType = 1), then the execution sequence of this intent information only includes the main function interface, which can be executed directly: siteDiagnose(xxxx, 1). Assuming the intent information is "Base station in area xx is interrupted, please diagnose", and one of the main parameters faultType of the corresponding main function interface siteDiagnose is determined (faultType = 1), while the other main parameter siteIP needs to be determined based on the secondary parameter siteArea through the secondary function interface querySiteService, then the execution sequence of this intent information includes the secondary function interface and the main function interface, and they are executed sequentially:

[0058] 1. querySiteService(xx area) outputs siteIP.

[0059] 2. siteDiagnose(the result of 1 above, siteIP, 1).

[0060] In this context, calling each functional interface to execute the corresponding task means first obtaining the address corresponding to that functional interface, and then accessing the corresponding functional interface based on that address to complete the call. For example, the above intent information "Base station xxxx cannot access the internet, please diagnose" is accessed as follows:

[0061] / api / bn-nm-diagnose / v1 / siteDiagnose.

[0062] The above intent message "Base station in area xx is down, please diagnose" should be accessed sequentially:

[0063] / api / bn-nm-sitemng / v1 / querySiteService; / api / bn-nm-diagnose / v1 / siteDiagnose.

[0064] In this embodiment, by parsing the input intent information, the main functional interface matching the intent information is determined, thereby improving the accuracy of main functional interface matching. After obtaining the main functional interface, the main parameters corresponding to the main functional interface are obtained. Then, in response to the judgment that the number of main parameters is greater than 0, the parameter value of each main parameter is obtained by parsing the intent information. Based on the parameter value of each main parameter, the execution sequence corresponding to the intent information is determined. The execution sequence includes all functional interfaces used to process the intent information, realizing the dynamic determination of the functional interface used to process the intent information according to the parameter value of each main parameter. Finally, the functional interfaces in the execution sequence are called sequentially to obtain the processing result matching the intent information, thereby improving the accuracy and flexibility of network operation and maintenance.

[0065] In one implementation, the method further includes: responding to a determination that the number of main parameters is equal to 0, processing the intent information by calling the main function interface to obtain a processing result matching the intent information. It is understood that if the main parameters corresponding to the main function interface are empty, i.e., the number of main parameters is 0, it means that the main function interface does not need to input main execution parameters. The intent information can be passed to the main function interface for processing, thereby generating a processing result matching the intent information, thus increasing the flexibility of intent processing.

[0066] In this embodiment of the application, dynamically determining the execution sequence corresponding to the intent information may include one of the following:

[0067] (1) In one implementation, determining the execution sequence corresponding to the intent information based on the parameter value of each of the main parameters includes: in response to the judgment that the parameter value of each of the main parameters is not empty, adding the main function interface to the execution sequence, wherein the execution result corresponding to the main function interface is the processing result.

[0068] Understandably, if the parameter value of the main parameter corresponding to the main function interface can be successfully extracted from the intent information, then the execution sequence only includes that main function interface. By calling this main function interface, a processing result matching the intent information can be directly obtained.

[0069] (2) In one implementation, determining the execution sequence corresponding to the intent information based on the parameter value of each of the main parameters includes: in response to a determination that at least one first main parameter has an empty parameter value, obtaining a secondary function interface for each first main parameter, wherein the secondary function interface is a function interface used to determine the parameter value of the first main parameter; and sequentially adding the secondary function interface and the main function interface to the execution sequence.

[0070] Understandably, if the parameter value of the main parameter corresponding to the main function interface cannot be extracted from the intent information, then the secondary function interface used to determine the parameter value of the first main parameter can be added to the execution sequence first, and then the main function interface can be added to the execution sequence, thus achieving flexible assembly of multiple function interfaces. Subsequently, each function interface in this execution sequence can be called sequentially, avoiding chaotic execution and the problem of execution failure due to missing parameter values.

[0071] The following are two methods for obtaining the functional interface as provided in an exemplary embodiment of this application:

[0072] (I) In one implementation, obtaining the secondary function interface includes: obtaining a first secondary parameter corresponding to the first primary parameter and the parameter value of the first secondary parameter according to the intent information and the parameter map corresponding to the primary function interface; obtaining the secondary function interface corresponding to the first secondary parameter, wherein the primary execution parameter of the secondary function interface is the first secondary parameter, and the execution result corresponding to the secondary function interface is the parameter value of the first primary parameter.

[0073] In this embodiment, the secondary function interface corresponding to the first secondary parameter is used to determine the parameter value of the first primary parameter. The first secondary parameter is determined based on the parameter graph corresponding to the first primary parameter and the primary function interface, and the value of the first secondary parameter is obtained from the intent information. That is, if the parameter value of the first primary parameter cannot be obtained from the intent information, then the first secondary parameter of the first primary parameter is obtained based on the parameter graph. The parameter value of the first secondary parameter can be obtained from the intent information. Then, the function interface for determining the parameter value of the first primary parameter based on the parameter value of the first secondary parameter is obtained. By calling this function interface, the parameter value of the first primary parameter is indirectly obtained.

[0074] In another implementation, obtaining the first secondary parameter corresponding to the first primary parameter and its parameter value based on the intent information and the parameter graph corresponding to the primary function interface may include: searching in the parameter graph for the first secondary parameter that can be indirectly reached from the first primary parameter until a stopping condition is met. The stopping condition includes obtaining the parameter value of the first secondary parameter by parsing the intent information, and the number of hops to indirectly reach it is N, where N is an integer greater than or equal to 1.

[0075] The parameter graph describes the dependencies between parameters; that is, providing parameter A allows us to indirectly obtain parameter B. The parameter graph includes at least two parameters, and the connecting edges between adjacent parameters represent the functional interface for obtaining the output parameter based on the input parameter. For example, as shown in Figure 2, the parameter graph can be defined by `parameter` and `leadTo`. `parameter` represents a vertex of the parameter graph, i.e., a parameter, and its attributes include the parameter name, which is the unique identifier of that vertex. `leadTo` represents an edge of the parameter graph, used to connect two parameter vertices, indicating that the output parameter can be obtained based on the input parameter. The edge's attributes can include: functional interface, i.e., the output parameter can be obtained based on the input parameter through this functional interface. For example, Figure 3 shows the instantiated parameter graph.

[0076] Understandably, in the parameter graph, the search begins with the first primary parameter as the starting vertex. Each time a first secondary parameter that can be directly or indirectly reached from the first primary parameter is found, it is determined whether the parameter value of the first secondary parameter can be extracted from the intent information. If the parameter value of the first secondary parameter can be extracted, the search stops, and the secondary function interfaces of the edges traversed from the first secondary parameter to the first primary parameter are obtained in sequence. For example, continuing to refer to Figure 3 above, assuming the functional interface is an API interface and the first primary parameter is parameter1, in the first search, parameter3 and parameter4 adjacent to parameter1 are searched. If it is determined that the parameter values ​​corresponding to parameter3 and parameter4 cannot be obtained from the intent information, the search continues. Since parameter4 is the last vertex, in the second search process, only the vertex adjacent to parameter3, i.e., parameter7, is searched. If it is determined that the parameter value corresponding to parameter7 can be obtained from the intent information, the search stops, and then the secondary functional interfaces on each edge from parameter7 to parameter1 are obtained. That is, the obtained secondary functional interfaces include: functional interface_7 and functional interface_3 in sequence.

[0077] In another implementation, obtaining the parameter value of the first secondary parameter by parsing the intent information includes: generating a fourth prompt word based on the intent information and the first secondary parameter, and providing the fourth prompt word to the target large model to obtain the parameter value of the first secondary parameter.

[0078] In other words, after obtaining a first secondary parameter, it is necessary to determine whether the parameter value corresponding to the first secondary parameter can be obtained from the intent information. Therefore, a fourth prompt word can be generated based on the intent information and the first secondary parameter. The fourth prompt word can be the prompt instruction of the target large language model. Then, the fourth prompt word is input into the target large language model to obtain the parameter value of the first secondary parameter.

[0079] (II) In one implementation, obtaining the slave function interface includes: obtaining a second slave parameter of the main function interface, wherein the second slave parameter is a slave execution parameter of the main function interface; in response to determining the parameter value corresponding to the second slave parameter based on the intent information, obtaining the slave function interface between the second slave parameter and the main parameter based on the parameter map corresponding to the main function interface.

[0080] It is understandable that some functional interfaces can also correspond to secondary parameters, that is, secondary parameters are parameters that are not required when calling the functional interface. If the parameter value corresponding to the primary parameter cannot be determined, the secondary parameter corresponding to the functional interface is obtained, and it is determined whether the parameter value of the secondary parameter can be extracted from the intent information. If the parameter value of the secondary parameter can be extracted, the secondary functional interface from the secondary parameter to the primary parameter is obtained based on the parameter map corresponding to the functional interface. For example, as shown in Figure 4, assuming the intent information is: "Base station in area xx is interrupted, please diagnose", the primary functional interface corresponding to this intent information is siteDiagnose. As shown in Table 2 above, the primary parameters corresponding to this primary functional interface are siteIP (base station IP) and faultType (fault type). The parameter value of faultType (fault type) can be determined according to the intent information and is 1. The parameter value of siteIP cannot be extracted from the intent information "Base station in area xx is interrupted, please diagnose". Although the intent information does not describe the base station IP, it describes the area where the base station is located as "area xx". Therefore, the parameter value of the second secondary parameter siteArea can be determined to be "area xx". In other words, the intent information describes a primary parameter faultType and a secondary parameter siteArea in the primary function interface. Then, based on the secondary parameter, the primary parameter, and the parameter graph, the secondary function interface from the secondary parameter to the primary parameter is obtained.

[0081] It should be noted that, in the embodiments of this application, each functional interface may correspond to 0 or 1 or more main parameters and 0 or 1 or more slave parameters.

[0082] In (I) or (II) above, based on the parametric graph, parameter identification and automatic assembly of functional interfaces are achieved, effectively expanding the scope of task-oriented intent support and enhancing the executability of intent tasks. Simultaneously, during the input phase, users can freely input known parameter values ​​into the intent information without needing to target specific parameters. This is because, based on the parametric graph, the parameter value of a specific parameter can be indirectly determined based on the user-inputted value of a particular parameter, thus ensuring the arbitrariness of intent expression.

[0083] (3) In one implementation, determining the execution sequence corresponding to the intent information based on the parameter value of each of the main parameters includes: in response to a determination that at least one of the main parameters has an empty parameter value, receiving the parameter value of each of the second main parameters; and adding the main function interface to the execution sequence.

[0084] Understandably, if the value of a certain secondary primary parameter cannot be obtained based on the intent information, the user can be prompted to input the value to obtain the corresponding value, thus ensuring the accuracy and reliability of the data. Introducing user input in this implementation enhances the fault tolerance of the intent information processing method.

[0085] The following is a schematic diagram of a task-oriented dialogue operation and maintenance system based on a large model, as shown in Figure 5, based on the above embodiments. It may include: a natural language front-end interaction module 510, an API selector 520, an API parameter parser 530, an API executor 540, and a result parser 550. Wherein:

[0086] Natural Language Front-End Interaction Module 510: Supports input of natural language intents, extracts contextual dialogue information and passes it to the back-end, and supports presentation of intent execution results.

[0087] API Selector 520: As shown in Figure 6, it includes API description files and skill description files that group APIs according to business characteristics. After receiving natural language intent, API Selector 520 first matches the appropriate skill based on the semantics of the skill's functional description according to the large model, and then semantically matches the appropriate API from the API set carried by the skill.

[0088] API parameter parser 530: As shown in Figure 7, it includes a parameter graph describing the dependencies between parameters. The API parameter parser 530 records information about which API (e.g., API_slave) is used to obtain another parameter (e.g., parameter_master) based on a certain parameter (e.g., parameter_slave), that is, it records the triple information of [parameter_slave, leadTo(attribute: API_slave), parameter_master]. After obtaining the main API, the API selector can retrieve the parameter `parameter_master` required to execute the main API. Based on the parameter `parameter_master` of the main API, it obtains indirectly accessible secondary parameters `parameter_slave` from the parameter graph, as well as secondary API interfaces for obtaining `parameter_master` from `parameter_slave`. This gives the system access to all possible parameters `parameter_slave` and `parameter_master` that the user can freely express. Based on the large model, it parses all or one of `parameter_slave` and `parameter_master` from the user's raw natural language intent, thus obtaining the API execution combination `[main API]` or `[secondary API, main API]`. This allows the system to automatically assemble interfaces to complete tasks based on the user's freely input intent. It should be noted that the main API can have multiple input parameters, and each parameter can find its own secondary parameters through the parameter graph.

[0089] API Executor 540: Based on the obtained API execution sequence and parsed parameters, it generates the correct parameters and request body for the API according to the API description document, and executes them sequentially, such as calling a RESTful API.

[0090] Result parser 550: Converts the API execution result into an interpretable result and returns it to the front end for presentation.

[0091] Figure 8 illustrates a flowchart of an intent information processing method provided in an exemplary embodiment of this application. This method can be executed by the aforementioned task-oriented dialogue operation and maintenance system based on a large model. As shown in Figure 8, it may include the following steps:

[0092] S810: Enter the original intent.

[0093] Users input their intent using natural language on the front end.

[0094] S820: Select the main API based on the large model.

[0095] In one implementation, as shown in Figure 9, S820 may include the following steps:

[0096] S821: Reads skill files.

[0097] S822: Automatically generate prompts for skill selection.

[0098] S823: Skill selection based on large model.

[0099] S824: Reads the API description file attached to the skill.

[0100] S825: Automatically generate prompts for selecting APIs.

[0101] S826: Select the API based on the large model to obtain the main API.

[0102] Regarding API description files, it's important to understand that an API refers to the functional interface definition that the system can execute. To ensure that the larger model can understand it, the API's supported functions and the meaning of its required input parameters are described using natural language. The fields included in an API description file may include:

[0103] name: Represents the name of the functional interface; version: Expresses the version information of the functional interface, such as V1, V2, V3, etc.; description: Represents the functional description of the functional interface; input_parameters: Represents the semantic description of the input parameters corresponding to this functional interface, which may include the following attributes: parameter name, parameter type, parameter description information, whether it is a required parameter, etc.; entity: Represents the entity definition of this functional interface, which may include the following attributes: functional interface protocol type, operation type, address, parameter entity definition, etc., as exemplified in Table 2 above.

[0104] Regarding skill description files, it can be understood that a skill is a capability description that corresponds to solving a type of business problem. It is associated with a set of APIs. Skills are used to group APIs, reducing the amount of API information presented to a large model at once, and thus improving the accuracy of intent recognition. The fields in its description file may include:

[0105] name: indicates the name of the functional feature classification description file; description: indicates the functional description of the functional feature classification description file; semantic_apis: indicates the functional interfaces included in the functional feature classification description file, for example, as shown in Table 1 above.

[0106] Understandably, after receiving the natural language description, the system first reads the skill description file. Based on the pre-designed prompt template, the system automatically generates a large model prompt. The large model, based on the prompt, first retrieves a suitable skill description file from the skill set according to the user's natural language input. This step is to avoid directly matching APIs from a large API set, as the number of APIs would be too large, exceeding the large model's token limit and reducing matching accuracy. Therefore, APIs are grouped according to their business functions; the group is found first, and then APIs are searched within that group. After finding the skill, the system reads the API set associated with that skill and, combined with the original natural language intent, automatically generates a large model prompt based on the pre-designed prompt template. This drives the large model to match the API name to be executed, called API_master, from the API set associated with the skill according to the API function description.

[0107] S830: Based on large model analytical parameters.

[0108] In one implementation, as shown in Figure 10, S830 may include the following steps:

[0109] S831: Obtain the input parameter description of the main API by reading the description file of the main API.

[0110] S832: Obtain the secondary parameters from the parameter map that can indirectly reach the primary parameters.

[0111] S833: Generate a prompt word based on the main parameter and the secondary parameter.

[0112] S834: Based on prompt words, a large model extracts the parameter values ​​of the main and secondary parameters from natural language intent.

[0113] The parametric graphs can be found in the descriptions in the foregoing embodiments, and will not be repeated here.

[0114] Understandably, the system parses the API_master description file obtained from S826 to obtain the input parameter description of the API, called the master parameter_master. There can be multiple master parameters. Then, from the parameter graph (in the form of triples [paramer_slave, leadTo(API_slave), parameter_master], indicating that the parameter paramer_slave is known, and the parameter parameter_master can be obtained by executing API_slave), it finds the slave parameters that can indirectly reach the master parameter. The number of hops for indirect access is N (N>=1). At this point, all parameter names that need to be parsed by the large model have been listed. Based on the designed prompt template, prompt words are automatically generated to drive the large model to extract these master and slave parameters from natural language. The extraction results indicate which parameters are described in the original natural language intent and which parameters are not described in the original natural language intent.

[0115] In another implementation, the value of the master or slave parameter of API_master can be input by the user.

[0116] S840: Generate API execution sequence.

[0117] In one implementation, as shown in Figure 11, S840 may include the following steps:

[0118] S841: Determine whether all parameters corresponding to the main API have been successfully extracted from the natural language intent.

[0119] If extraction is successful, proceed to S842; otherwise, proceed to S845.

[0120] S842: Retrieve the main parameters that were not successfully extracted and their corresponding successfully extracted secondary parameters.

[0121] S843: Query the parameter graph to obtain the API from the secondary parameter to the primary parameter.

[0122] S844: Add the API to the API execution sequence.

[0123] S845: Add the main API to the API execution sequence.

[0124] Understandably, based on the parsed parameter list obtained from S830 above, and combined with the parameter graph, the API execution sequence is obtained. The parameter graph is in triplet form: [parameter_slave, leadTo(API_slave), parameter_master], indicating that the parameter parameter_slave is known, and by executing API_slave, the parameter parameter_master can be obtained.

[0125] If all the parameters required by the main API are successfully extracted from the user's natural language intent, then the API execution sequence will only have this one main API, namely API_master. If the parameters required by the main API are not extracted from the natural language intent, but indirect secondary parameter information is extracted, then the system can obtain the secondary API that reaches the main parameters, namely API_slave, based on this secondary parameter secondary parameter graph, and add this API_slave to the execution sequence. That is, the secondary API, namely API_slave, is executed first to obtain the main parameters, and then the main API, namely API_master, is executed.

[0126] S850: Execute the APIs in the execution sequence sequentially.

[0127] Understandably, the actual API interfaces, such as REST interfaces, are called sequentially according to the API execution sequence and the description of the API entities in the API description file to execute the task.

[0128] S860: API result parsing and front-end presentation.

[0129] Understandably, after execution, the results are parsed and returned to the front end for display.

[0130] This application provides a task-oriented dialogue operation and maintenance system based on a large model. By leveraging the powerful semantic understanding and generalization capabilities of the large model, along with a proposed supporting application architecture and specification definition, the system ensures that the dialogue system meets requirements in terms of generalization, flexibility, and network operation and maintenance accuracy, thereby simplifying operation and maintenance. Specifically, the large model, combined with skill-API two-level matching, enables intent understanding and API selection. On one hand, the large model enhances the naturalness and versatility of intent language expression, improving user experience; on the other hand, it improves the accuracy of intent recognition and API method selection, enhancing network operation and maintenance precision. Simultaneously, the large model, in conjunction with parameter graphs, enables API parameter recognition and automatic API assembly. This significantly expands the scope of task-oriented intent support and enhances the executability of intent tasks.

[0131] As shown in Figure 12, this application embodiment also provides an electronic device 1200, including a processor 1210 and a memory 1220. The memory 1220 stores a program or instructions that can run on the processor 1210. When the program or instructions are executed by the processor 1210, they implement the various processes of the embodiments shown in Figures 1 to 11 above and can achieve the same technical effect. To avoid repetition, they will not be described again here.

[0132] This application also provides a readable storage medium storing a program or instructions. When the program or instructions are executed by a processor, they implement the various processes of the embodiments shown in Figures 1 to 11 above and achieve the same technical effect. To avoid repetition, they will not be described again here.

[0133] The processor mentioned above is the processor in the terminal described in the above embodiments. The readable storage medium includes computer-readable storage media, such as computer read-only memory (ROM), random access memory (RAM), magnetic disk, or optical disk. In some examples, the readable storage medium may be a non-transient readable storage medium.

[0134] This application embodiment also provides a chip, which includes a processor and a communication interface. The communication interface is coupled to the processor. The processor is used to run programs or instructions to implement the various processes of the embodiments shown in Figures 1 to 11 above, and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0135] It should be understood that the chip mentioned in the embodiments of this application may also be referred to as a system-on-a-chip, system chip, chip system, or system-on-a-chip, etc.

[0136] This application also provides a computer program / program product, which is stored in a storage medium and executed by at least one processor to implement the various processes of the embodiments shown in Figures 1 to 11 above, and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0137] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element. Furthermore, it should be noted that the scope of the methods and apparatuses in the embodiments of this application is not limited to performing functions in the order shown or discussed, but may also include performing functions substantially simultaneously or in the reverse order, depending on the functions involved. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Additionally, features described with reference to certain examples may be combined in other examples.

[0138] From the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of computer software products plus necessary general-purpose hardware platforms, and of course, they can also be implemented by hardware. The computer software product is stored in a storage medium (such as ROM, RAM, magnetic disk, optical disk, etc.) and includes several instructions to cause the terminal or network-side device to execute the methods described in the various embodiments of this application.

[0139] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other implementations under the guidance of this application without departing from the spirit and scope of the claims. All of these implementations are within the protection scope of this application.

Claims

1. An intent information processing method, wherein, include: By parsing the input intent information, the main functional interface that matches the intent information is determined; Obtain the main parameters corresponding to the main function interface, wherein the main parameters are the main execution parameters of the main function interface; In response to the determination that the number of main parameters is greater than 0, the parameter value of each main parameter is obtained by parsing the intent information; Based on the parameter value of each of the main parameters, an execution sequence corresponding to the intent information is determined, wherein the execution sequence includes all functional interfaces used to process the intent information; The functional interfaces in the execution sequence are called sequentially to obtain processing results that match the intent information.

2. The method according to claim 1, wherein, The step of determining the execution sequence corresponding to the intent information based on the parameter value of each of the main parameters includes: In response to the determination that the parameter value of each of the main parameters is not empty, the main function interface is added to the execution sequence, wherein the execution result corresponding to the main function interface is the processing result.

3. The method according to claim 1, wherein, The step of determining the execution sequence corresponding to the intent information based on the parameter value of each of the main parameters includes: In response to the determination that at least one of the primary parameters has an empty parameter value, for each primary parameter, a secondary function interface is obtained, wherein the secondary function interface is a function interface used to determine the parameter value of the primary parameter. The slave function interface and the master function interface corresponding to each of the first master parameters are sequentially added to the execution sequence, wherein each slave function interface is located before the master function interface in the execution sequence.

4. The method according to claim 1, wherein, The step of determining the execution sequence corresponding to the intent information based on the parameter value of each of the main parameters includes: In response to the determination that at least one of the main parameters has an empty parameter value, the parameter value of each of the second main parameters is received. Add the main function interface to the execution sequence.

5. The method according to claim 1, wherein, The step of obtaining the parameter value of each main parameter by parsing the intent information includes: Based on the intent information and each of the main parameters, a third prompt word is generated, and the third prompt word is provided to the target large model to obtain the parameter value of each of the main parameters.

6. The method according to claim 3, wherein, The acquisition function interface includes: Based on the intent information and the parameter map corresponding to the main function interface, obtain the first secondary parameter corresponding to the first main parameter and the parameter value of the first secondary parameter; Obtain the slave function interface corresponding to the first slave parameter, wherein the main execution parameter of the slave function interface is the first slave parameter, and the execution result corresponding to the slave function interface is the parameter value of the first main parameter.

7. The method according to claim 3, wherein, The acquisition function interface includes: Obtain the second slave parameter of the main function interface, wherein the second slave parameter is the slave execution parameter of the main function interface; In response to determining the parameter value corresponding to the second slave parameter based on the intent information, the slave function interface between the second slave parameter and the main parameter is obtained based on the parameter map corresponding to the main function interface.

8. The method according to claim 6, wherein, The step of obtaining the first secondary parameter and its parameter value corresponding to the first primary parameter based on the intent information and the parameter map corresponding to the primary function interface includes: In the parameter map, the first slave parameter that can be indirectly reached from the first master parameter is searched until a stopping condition is met. The stopping condition includes obtaining the parameter value of the first slave parameter by parsing the intent information, and the number of hops to indirectly reach it is N, where N is an integer greater than or equal to 1.

9. The method according to claim 8, wherein, The step of obtaining the parameter value of the first parameter by parsing the intent information includes: Based on the intent information and the first sub-parameter, a fourth prompt word is generated, and the fourth prompt word is provided to the target large model to obtain the parameter value of the first sub-parameter.

10. The method according to claim 1, wherein, The step of determining the main function interface that matches the input intent information by parsing the input intent information includes: Obtain a functional feature classification description file that matches the intent information, wherein the functional feature classification description file includes at least one functional interface, and each functional interface corresponds to a different function; The main functional interface that matches the intent information is determined from at least one of the functional interfaces.

11. The method according to claim 10, wherein, The acquisition of the functional characteristic classification description file matching the intent information includes: Based on the intent information, a first prompt word is generated, and the first prompt word is provided to the target large model to obtain a functional characteristic classification description file that matches the intent information.

12. The method according to claim 10, wherein, Determining the main functional interface that matches the intent information from at least one of the functional interfaces includes: A second prompt word is generated based on the intent information and the at least one functional interface, and the second prompt word is provided to the target large model to obtain the main functional interface that matches the intent information.

13. The method according to claim 1, wherein, The method further includes: In response to the determination that the number of main parameters is equal to 0, the intent information is processed by calling the main function interface to obtain a processing result that matches the intent information.

14. An electronic device, wherein, It includes a processor, a memory, and a program or instructions stored in the memory and executable on the processor, wherein when the program or instructions are executed by the processor, they implement the steps of the intent information processing method as described in any one of claims 1-13.

15. A readable storage medium, wherein, The readable storage medium stores a program or instructions that, when executed by a processor, implement the steps of the intent information processing method as described in any one of claims 1-13.

16. A computer program product, wherein, The computer program product includes program instructions that, when executed by a computer, cause the computer to perform the steps of the intent information processing method as described in any one of claims 1-13.

Images