User intention analysis method and system, terminal and storage medium
By constructing skill tasks and intention descriptions, using big models to generate intent sample data and train intent models, the problem of intent model training in the existing technology is solved, and efficient user intention analysis is achieved.
Patent Information
- Application Number
- CN202510082903.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-01-20
- Publication Date
- 2025-05-13
AI Technical Summary
In the prior art, the data annotation cost of training sample of intention models is high, resulting in low training efficiency of intention models and the analysis of user intention cannot be quickly realized.
By determining multiple skill tasks and intention descriptions, the first preset big model is used to generate intent sample data, and the second preset big model is trained based on the target intent sample data to obtain the target intent model, thereby achieving efficient analysis of user intentions.
It realizes the generation of intent sample data without labels, improves the training efficiency of the intent model, and improves the analysis efficiency of user intent.
Smart Images
Figure CN119990151A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of data processing technology, and in particular to a user intention analysis method, system, terminal and computer-readable storage medium. Background Art
[0002] In the process of analyzing user intentions in the prior art, it is generally necessary to set up a corresponding intention analysis model, so as to analyze the user's input speech through the intention analysis model to obtain the intention analysis result.
[0003] However, in the analysis process of the intent analysis model, a large amount of manual labeling is often required to obtain training sample data that meets the requirements. The labeling cost is high, which leads to low training efficiency of the intent model and inability to quickly analyze user intent.
[0004] Therefore, the prior art still needs to be improved and developed. Summary of the invention
[0005] The main purpose of the present invention is to provide a user intent analysis method, system, terminal and computer-readable storage medium, aiming to solve the problem in the prior art that the training sample data of the intent model is high in cost, resulting in low training efficiency of the intent model and inability to quickly realize the analysis of user intent.
[0006] To achieve the above object, the present invention provides a user intention analysis method, which comprises the following steps:
[0007] Determine a plurality of skill tasks, and create an intent description and example sample data corresponding to each of the skill tasks;
[0008] Determine a first preset large model, and input the intention description and the example sample data into the first preset large model to obtain a plurality of intention sample data corresponding to the plurality of skill tasks;
[0009] Determine a second preset large model and target intent sample data among the plurality of intent sample data, and train the second preset large model according to the target intent sample data to obtain a target intent model;
[0010] Acquire current user speech data, input the current user speech data into the target intent model, and output the user intent analysis result.
[0011] Optionally, the user intention analysis method, wherein the determining of multiple skill tasks and creating an intention description and example sample data corresponding to each of the skill tasks, specifically includes:
[0012] Determine multiple skill tasks, and create an intention name and slot information corresponding to each of the skill tasks;
[0013] Obtaining an intention description corresponding to the intention name, wherein the intention description includes a main task content or an intended execution operation corresponding to the intention name;
[0014] The user sample speech corresponding to each of the skill tasks is obtained, and example sample data is obtained according to the user sample speech, the intent name and the slot information.
[0015] Optionally, the user intention analysis method, wherein the determining a first preset macromodel, and inputting the intention description and the example sample data into the first preset macromodel to obtain a plurality of intention sample data corresponding to the plurality of skill tasks, specifically includes:
[0016] Determine a first preset macromodel, and input the intention description, the example sample data, and the number of samples corresponding to the example sample data into the first preset macromodel;
[0017] The sample data is generated through the first preset large model to obtain multiple intention sample data corresponding to the multiple skill tasks.
[0018] Optionally, the user intent analysis method, wherein the determining the second preset large model and the target intent sample data among the plurality of intent sample data, and training the second preset large model according to the target intent sample data to obtain the target intent model, specifically includes:
[0019] Determine preset target user needs, and determine the specified skill intentions based on the preset target user needs;
[0020] Searching in the plurality of the intent sample data according to the designated skill intent to obtain target intent sample data;
[0021] Determine a second preset large model that needs to be trained, and train the second preset large model according to the target intent sample data to obtain a target intent model.
[0022] Optionally, the user intent analysis method, wherein the determining of the second preset large model and the target intent sample data among the plurality of intent sample data, and training the second preset large model according to the target intent sample data to obtain the target intent model, further comprises:
[0023] The target intent model is deployed and a model address corresponding to the target intent model is obtained.
[0024] Optionally, the user intent analysis method, wherein the determining of the second preset large model and the target intent sample data among the plurality of intent sample data, and training the second preset large model according to the target intent sample data to obtain the target intent model, further comprises:
[0025] When a model test instruction is received, obtaining the model address according to the model test instruction, and determining the target intent model according to the model address;
[0026] Acquire a user speech test sample, and perform model verification processing on the target intent model according to the user speech test sample to obtain a model verification result;
[0027] If the model verification result does not meet the preset requirements, the example sample data is adjusted to obtain updated example sample data;
[0028] The target intent model is retrained according to the updated example sample data until the model verification result corresponding to the target intent model meets the preset requirements.
[0029] Optionally, the user intention analysis method, wherein the step of obtaining current user speech data, inputting the current user speech data into the target intention model, and outputting the user intention analysis result, specifically includes:
[0030] Acquire current user speech data, and call the model address according to the current user speech data;
[0031] Determine the target intent model according to the model address, and input the current user speech data into the target intent model;
[0032] Perform intent analysis on the current user speech data according to the target intent model to obtain an intent analysis result.
[0033] In addition, to achieve the above-mentioned purpose, the present invention further provides a user intention analysis system, wherein the user intention analysis system comprises:
[0034] A skill task creation module, used to determine multiple skill tasks and create an intention description and example sample data corresponding to each of the skill tasks;
[0035] An intention sample data generation module, used to determine a first preset macromodel, and input the intention description and the example sample data into the first preset macromodel to obtain a plurality of intention sample data corresponding to a plurality of the skill tasks;
[0036] A target intent model training module, used to determine a second preset large model and target intent sample data from the plurality of intent sample data, and train the second preset large model according to the target intent sample data to obtain a target intent model;
[0037] The user intention analysis result generation module is used to obtain the current user speech data, input the current user speech data into the target intention model, and output the user intention analysis result.
[0038] In addition, to achieve the above-mentioned purpose, the present invention also provides a terminal, wherein the terminal includes: a memory, a processor, and a user intention analysis program stored on the memory and executable on the processor, and when the user intention analysis program is executed by the processor, the steps of the user intention analysis method as described above are implemented.
[0039] In addition, to achieve the above-mentioned purpose, the present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a user intention analysis program, and when the user intention analysis program is executed by a processor, the steps of the user intention analysis method as described above are implemented.
[0040] In the present invention, multiple skill tasks are determined, and the intent description and example sample data corresponding to each of the skill tasks are created; a first preset large model is determined, and the intent description and the example sample data are input into the first preset large model to obtain multiple intent sample data corresponding to the multiple skill tasks; a second preset large model and target intent sample data in the multiple intent sample data are determined, and the second preset large model is trained according to the target intent sample data to obtain a target intent model; the current user speech data is obtained, and the current user speech data is input into the target intent model, and the user intent analysis result is output. The present invention can realize the unlabeled generation of intent sample data by constructing intent descriptions and example sample data corresponding to multiple skill tasks, and inputting the intent descriptions and example sample data into the first preset large model. Further, the second preset large model is trained according to the intent sample data to obtain the target intent model, and the target intent model is deployed, which can efficiently realize the call of the target intent model, which not only effectively improves the training efficiency of the target intent model, but also effectively improves the analysis efficiency of user intent. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] Figure 1 is a flow chart of a preferred embodiment of the user intention analysis method of the present invention;
[0042] Figure 2It is an overall flow chart of the preferred embodiment of the user intention analysis method of the present invention, which is based on a large model and fully automated from sample generation to intention model training and deployment;
[0043] Figure 3 is a structural diagram of a preferred embodiment of the user intention analysis system of the present invention;
[0044] Figure 4 It is a structural diagram of a preferred embodiment of the terminal of the present invention. DETAILED DESCRIPTION
[0045] In order to make the purpose, technical solution and advantages of the present invention clearer and more specific, the present invention is further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.
[0046] In the process of analyzing user intentions in the prior art, it is generally necessary to set up a corresponding intention analysis model, so as to analyze the user's input speech through the intention analysis model to obtain the intention analysis result.
[0047] However, in the analysis process of the intent analysis model, a large amount of manual labeling is often required to obtain training sample data that meets the requirements. The labeling cost is high, which leads to low training efficiency of the intent model and inability to quickly analyze user intent.
[0048] To solve the above problems, the present invention provides a method and platform for fully automated deployment from sample generation to intent model training based on a large model. First, create a skill (i.e., multiple skill tasks in the present invention) on the platform. After the skill is created, create the corresponding intent name and slot information under the skill. According to the intent description (the main task content or intent execution operation corresponding to the intent name) and the example sample (i.e., the example sample data in the present invention), the sample data (i.e., multiple intent sample data in the present invention) can be automatically generated based on the large model. The training of the corresponding intent model can be completed by selecting the sample data of the specified skill intent (i.e., the target intent sample data in the present invention). After the model training is completed, click one-click deployment. The intent information can be successfully identified through the deployed model address. The present invention can be applied to products such as televisions.
[0049] The user intention analysis method described in the preferred embodiment of the present invention is as follows: Figure 1 As shown, the user intention analysis method includes the following steps:
[0050] Step S10: determine multiple skill tasks, and create an intent description and example sample data corresponding to each of the skill tasks.
[0051] like Figure 2As shown, before generating the intent sample data, the present invention first needs to create multiple skill tasks on the platform, including the intent name and slot information in the skill task, and then generate the intent description and example sample data according to the intent name and slot information.
[0052] There can be multiple intent sample data. You only need to select multiple different skill tasks when creating a skill. You can create them separately on the platform and generate multiple intent sample data at the same time.
[0053] Specifically, multiple skill tasks are determined, and an intent name and slot information corresponding to each skill task is created; an intent description corresponding to the intent name is obtained, wherein the intent description includes the main task content or intended execution operation corresponding to the intent name; user sample scripts corresponding to each skill task are obtained, and example sample data is obtained based on the user sample scripts, the intent name and the slot information.
[0054] When creating a skill (i.e., the skill task in the present invention), you need to fill in the skill name (skill key), where the skill name is the colloquial name of the skill, usually in Chinese, which can describe a certain function of the skill. The skill key is generally used as the unique identifier of the skill to reflect the function and purpose of the skill. The successfully created skill information will be stored in the database of the server.
[0055] It is understandable that skill is a large concept and entity, corresponding to entities in real life, and can complete a certain function that people can perceive, so skills can cover all aspects of life, such as film and television skills on TV, volume control skills, music skills, train ticket skills, takeaway skills, etc.
[0056] like Figure 2 As shown, the intent name information (i.e., the intent name in the present invention) is generally an English identifier, and then the intent description information (i.e., the intent description in the present invention) is filled in, wherein the intent description can better describe the main task of the intent or the operation that the system should perform. After that, fill in the slot information associated with the intent, including the name of the slot, which is generally an English identifier. After successful creation, the intent (including the intent name and intent description information) and slot information will be stored in the database on the server side.
[0057] It can be understood that the intent description is a detailed description of the intent name, and the example sample (i.e., the example sample data in the present invention) is sample data that includes the intent name, slot information and user speech (i.e., the user sample speech in the present invention), which provides a prerequisite for the automatic generation of subsequent samples.
[0058] like Figure 2As shown, the present invention takes the takeaway skill as an example. For example: first, the user creates a skill on the platform, and creates an intent to search for takeaways under the takeaway skill. The intent description is: waimai_search, and the intent is an intent to search for takeaways. The name of the intent is: waimai_search, and the name of the slot is: extre_info, where the value of the slot is the original input words of the user.
[0059] The generated sample data may be: open the takeaway and see what's delicious. Sample example: {'waimai_search':{'extra_info':['Open the takeaway and see what's delicious']}}.
[0060] Step S20, determine a first preset large model, and input the intention description and the example sample data into the first preset large model to obtain multiple intention sample data corresponding to the multiple skill tasks.
[0061] like Figure 2 As shown, after obtaining the intention description and example sample data, the present invention can directly input them into the first preset large model to generate sample data, thereby obtaining the intention sample data corresponding to each skill task.
[0062] Specifically, a first preset large model is determined, and the intention description, the example sample data, and the number of samples corresponding to the example sample data are input into the first preset large model; sample data is generated through the first preset large model to obtain multiple intention sample data corresponding to multiple skill tasks.
[0063] It can be understood that the generation of sample data is to take the example sample, intent description and sample quantity as the input of the big model (that is, the first preset big model in the present invention), and let the big model complete the sample generation according to the prompt words for generating samples. The generated sample data (that is, multiple intent sample data in the present invention) will be stored in the database of the server.
[0064] The present invention generates a large number of intent samples based on the large model according to the intent description and example samples, saving a lot of manual annotation for intent model training.
[0065] Step S30, determine a second preset large model and target intent sample data among the plurality of intent sample data, and train the second preset large model according to the target intent sample data to obtain a target intent model.
[0066] It can be understood that the training process of the model is to input the target intent sample data obtained above into the large model (i.e., the second preset large model in the present invention) for training. Taking the skill of taking takeout as an example, it is necessary to screen from a plurality of the intent sample data and select the intent sample data related to the skill of taking takeout. Afterwards, the second preset large model is trained according to the intent sample data related to the skill of taking takeout, and then an intent model related to taking takeout can be obtained.
[0067] Specifically, the preset target user needs are determined, and the specified skill intentions are determined based on the preset target user needs; the target intention sample data are searched in the multiple intention sample data based on the specified skill intentions to obtain the target intention sample data; the second preset large model that needs to be trained is determined, and the second preset large model is trained based on the target intention sample data to obtain the target intention model.
[0068] After completing the generation of intent samples (i.e., example sample data in the present invention), the present invention selects required skill intent sample data (i.e., target intent sample data in the present invention) according to requirements (i.e., preset target user requirements in the present invention) to complete the training of the intent model (i.e., the second preset large model in the present invention), which can greatly shorten the time for model training.
[0069] Furthermore, the target intent model is deployed and a model address corresponding to the target intent model is obtained.
[0070] like Figure 2 As shown, when the target intent model is trained, click the deploy button on the platform, and then the target intent model will be deployed on the cloud server, and then the model address will be generated (the generated model address is actually the API interface address). The present invention can realize one-click deployment of the target intent model.
[0071] Furthermore, when a model testing instruction is received, the model address is obtained according to the model testing instruction, and the target intent model is determined according to the model address; a user speech test sample is obtained, and model verification processing is performed on the target intent model according to the user speech test sample to obtain a model verification result; if the model verification result does not meet the preset requirements, the example sample data is adjusted to obtain updated example sample data; the target intent model is retrained according to the updated example sample data until the model verification result corresponding to the target intent model meets the preset requirements.
[0072] It can be understood that the present invention can use the model address url to complete the verification of intent, use similar sample speech (i.e., the user speech test sample in the present invention) for verification, and determine whether the corresponding intent information can be normally identified (equivalent to building a verification set, and then verifying the accuracy of the model through the verification set).
[0073] If the validation fails, adjust the sample data volume and richness and retrain.
[0074] Step S40: obtain the current user speech data, input the current user speech data into the target intention model, and output the user intention analysis result.
[0075] After the target intent model is trained and deployed, taking the TV as an example, when the TV receives what the user says, it will analyze what the user says to determine the corresponding model address, so as to select the corresponding target intent model to perform user intent analysis, thereby outputting the user intent analysis results.
[0076] Specifically, the current user speech data is obtained, and the model address is called according to the current user speech data; the target intent model is determined according to the model address, and the current user speech data is input into the target intent model; the current user speech data is subjected to intent analysis according to the target intent model to obtain an intent analysis result.
[0077] It can be understood that the model address generated after the target intent model is successfully deployed is input into the model address according to the user's speech (i.e., the current user speech data in the present invention), and the intent analysis is performed through the target intent model corresponding to the model address to obtain the corresponding training intent result (i.e., the intent analysis result in the present invention).
[0078] Take the takeaway search intent of the takeaway skill as an example. The takeaway search intent is: waimai_search, and the user's phrase is: "I want to find some delicious takeaways." Through simulation debugging of the model address, the intent of waimai_search can be output. If you want to enrich the skill, you can also add slot information such as takeaway locations in the sample.
[0079] Technical effects of the present invention:
[0080] 1. Based on the automation of large models, zero-label generation of intent sample data can be achieved.
[0081] 2. Ability to efficiently complete the training and deployment of intent models with just one click.
[0082] 3. The present invention can simultaneously support sample generation of multiple skill intents and training and deployment of skills, greatly reducing the cost of manual labeling and improving the efficiency of intent model training and deployment.
[0083] Furthermore, if Figure 3 As shown, based on the above user intention analysis method, the present invention also provides a user intention analysis system accordingly, wherein the user intention analysis system includes:
[0084] A skill task creation module 51 is used to determine a plurality of skill tasks and create an intention description and example sample data corresponding to each of the skill tasks;
[0085] An intention sample data generating module 52 is used to determine a first preset macromodel, and input the intention description and the example sample data into the first preset macromodel to obtain a plurality of intention sample data corresponding to the plurality of skill tasks;
[0086] A target intent model training module 53 is used to determine a second preset large model and target intent sample data from the plurality of intent sample data, and train the second preset large model according to the target intent sample data to obtain a target intent model;
[0087] The user intention analysis result generation module 54 is used to obtain the current user speech data, input the current user speech data into the target intention model, and output the user intention analysis result.
[0088] Furthermore, if Figure 4 As shown, based on the above-mentioned user intention analysis method and system, the present invention also provides a terminal accordingly, and the terminal includes a processor 10, a memory 20 and a display 30. Figure 4 Only some components of the terminal are shown, but it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.
[0089] The memory 20 may be an internal storage unit of the terminal in some embodiments, such as a hard disk or memory of the terminal. The memory 20 may also be an external storage device of the terminal in other embodiments, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (SecureDigital, SD) card, a flash card (Flash Card), etc. equipped on the terminal. Further, the memory 20 may also include both an internal storage unit of the terminal and an external storage device. The memory 20 is used to store application software and various types of data installed on the terminal, such as the program code of the installation terminal, etc. The memory 20 may also be used to temporarily store data that has been output or is to be output. In one embodiment, a user intention analysis program 40 is stored on the memory 20, and the user intention analysis program 40 can be executed by the processor 10, thereby realizing the user intention analysis method in the present application.
[0090] In some embodiments, the processor 10 may be a central processing unit (CPU), a microprocessor or other data processing chip, used to run the program code or process data stored in the memory 20, such as executing the user intent analysis method.
[0091] In some embodiments, the display 30 may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, an OLED (Organic Light-Emitting Diode) touch device, etc. The display 30 is used to display information on the terminal and to display a visual user interface.
[0092] In one embodiment, when the processor 10 executes the user intention analysis program 40 in the memory 20, the following steps are implemented:
[0093] Determine a plurality of skill tasks, and create an intent description and example sample data corresponding to each of the skill tasks;
[0094] Determine a first preset large model, and input the intention description and the example sample data into the first preset large model to obtain a plurality of intention sample data corresponding to the plurality of skill tasks;
[0095] Determine a second preset large model and target intent sample data among the plurality of intent sample data, and train the second preset large model according to the target intent sample data to obtain a target intent model;
[0096] Acquire current user speech data, input the current user speech data into the target intent model, and output the user intent analysis result.
[0097] The step of determining a plurality of skill tasks and creating an intention description and example sample data corresponding to each of the skill tasks specifically includes:
[0098] Determine multiple skill tasks, and create an intention name and slot information corresponding to each of the skill tasks;
[0099] Obtaining an intention description corresponding to the intention name, wherein the intention description includes a main task content or an intended execution operation corresponding to the intention name;
[0100] The user sample speech corresponding to each of the skill tasks is obtained, and example sample data is obtained according to the user sample speech, the intent name and the slot information.
[0101] The step of determining a first preset macromodel and inputting the intention description and the example sample data into the first preset macromodel to obtain a plurality of intention sample data corresponding to the plurality of skill tasks specifically includes:
[0102] Determine a first preset macromodel, and input the intention description, the example sample data, and the number of samples corresponding to the example sample data into the first preset macromodel;
[0103] The sample data is generated through the first preset large model to obtain multiple intention sample data corresponding to the multiple skill tasks.
[0104] The step of determining the second preset large model and target intent sample data from the plurality of intent sample data, and training the second preset large model according to the target intent sample data to obtain the target intent model specifically includes:
[0105] Determine preset target user needs, and determine the specified skill intentions based on the preset target user needs;
[0106] Searching in the plurality of the intent sample data according to the designated skill intent to obtain target intent sample data;
[0107] Determine a second preset large model that needs to be trained, and train the second preset large model according to the target intent sample data to obtain a target intent model.
[0108] The step of determining a second preset large model and target intent sample data from a plurality of the intent sample data, and training the second preset large model according to the target intent sample data to obtain a target intent model, further includes:
[0109] The target intent model is deployed and a model address corresponding to the target intent model is obtained.
[0110] The step of determining a second preset large model and target intent sample data from a plurality of the intent sample data, and training the second preset large model according to the target intent sample data to obtain a target intent model, further includes:
[0111] When a model test instruction is received, obtaining the model address according to the model test instruction, and determining the target intent model according to the model address;
[0112] Acquire a user speech test sample, and perform model verification processing on the target intent model according to the user speech test sample to obtain a model verification result;
[0113] If the model verification result does not meet the preset requirements, the example sample data is adjusted to obtain updated example sample data;
[0114] The target intent model is retrained according to the updated example sample data until the model verification result corresponding to the target intent model meets the preset requirements.
[0115] The step of obtaining the current user speech data, inputting the current user speech data into the target intention model, and outputting the user intention analysis result specifically includes:
[0116] Acquire current user speech data, and call the model address according to the current user speech data;
[0117] Determine the target intent model according to the model address, and input the current user speech data into the target intent model;
[0118] Perform intent analysis on the current user speech data according to the target intent model to obtain an intent analysis result.
[0119] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a user intention analysis program, and when the user intention analysis program is executed by a processor, the steps of the user intention analysis method as described above are implemented.
[0120] In summary, the present invention provides a user intention analysis method and related equipment, the method comprising: determining multiple skill tasks, and creating an intention description and example sample data corresponding to each skill task; determining a first preset large model, and inputting the intention description and the example sample data into the first preset large model, and obtaining multiple intention sample data corresponding to multiple skill tasks; determining a second preset large model and target intention sample data in the multiple intention sample data, and training the second preset large model according to the target intention sample data to obtain a target intention model; obtaining current user speech data, and inputting the current user speech data into the target intention model, and outputting a user intention analysis result. The present invention can realize the generation of intention sample data without annotation by constructing intention descriptions and example sample data corresponding to multiple skill tasks, and inputting the intention description and example sample data into the first preset large model. Further, the second preset large model is trained according to the intention sample data to obtain the target intention model, and the target intention model is deployed, which can efficiently realize the call of the target intention model, which not only effectively improves the training efficiency of the target intention model, but also effectively improves the analysis efficiency of user intention.
[0121] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or terminal including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or terminal. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or terminal including the element.
[0122] Of course, those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be implemented by instructing related hardware (such as a processor, a controller, etc.) through a computer program, and the program can be stored in a computer-readable storage medium that can be read by a computer, and the program can include the processes of the above-mentioned method embodiments when executed. The computer-readable storage medium can be a memory, a disk, an optical disk, etc.
[0123] It should be understood that the application of the present invention is not limited to the above examples. For ordinary technicians in this field, improvements or changes can be made based on the above description. All these improvements and changes should fall within the scope of protection of the claims attached to the present invention.
Claims
1. A user intention analysis method, characterized in that: The user intention analysis method comprises: Determine a plurality of skill tasks, and create an intent description and example sample data corresponding to each of the skill tasks; Determine a first preset large model, and input the intention description and the example sample data into the first preset large model to obtain a plurality of intention sample data corresponding to the plurality of skill tasks; Determine a second preset large model and target intent sample data among the plurality of intent sample data, and train the second preset large model according to the target intent sample data to obtain a target intent model; Acquire current user speech data, input the current user speech data into the target intent model, and output the user intent analysis result.
2. The user intention analysis method according to claim 1, characterized in that: The determining of multiple skill tasks and creating an intent description and example sample data corresponding to each of the skill tasks specifically includes: Determine multiple skill tasks, and create an intention name and slot information corresponding to each of the skill tasks; Obtaining an intention description corresponding to the intention name, wherein the intention description includes a main task content or an intended execution operation corresponding to the intention name; The user sample speech corresponding to each of the skill tasks is obtained, and example sample data is obtained according to the user sample speech, the intent name and the slot information.
3. The user intention analysis method according to claim 1, characterized in that: The determining of the first preset macromodel and inputting the intention description and the example sample data into the first preset macromodel to obtain a plurality of intention sample data corresponding to the plurality of skill tasks specifically includes: Determine a first preset macromodel, and input the intention description, the example sample data, and the number of samples corresponding to the example sample data into the first preset macromodel; The sample data is generated through the first preset large model to obtain multiple intention sample data corresponding to the multiple skill tasks.
4. The user intention analysis method according to claim 1, characterized in that: The determining of the second preset large model and the target intent sample data among the plurality of intent sample data, and training the second preset large model according to the target intent sample data to obtain the target intent model specifically includes: Determine preset target user needs, and determine the specified skill intentions based on the preset target user needs; Searching in the plurality of the intent sample data according to the designated skill intent to obtain target intent sample data; Determine a second preset large model that needs to be trained, and train the second preset large model according to the target intent sample data to obtain a target intent model.
5. The user intention analysis method according to claim 1, characterized in that: The step of determining a second preset large model and target intent sample data from the plurality of intent sample data, and training the second preset large model according to the target intent sample data to obtain a target intent model, further includes: The target intent model is deployed and a model address corresponding to the target intent model is obtained.
6. The user intention analysis method according to claim 5, characterized in that: The step of determining a second preset large model and target intent sample data from the plurality of intent sample data, and training the second preset large model according to the target intent sample data to obtain a target intent model, further includes: When a model test instruction is received, obtaining the model address according to the model test instruction, and determining the target intent model according to the model address; Acquire a user speech test sample, and perform model verification processing on the target intent model according to the user speech test sample to obtain a model verification result; If the model verification result does not meet the preset requirements, the example sample data is adjusted to obtain updated example sample data; The target intent model is retrained according to the updated example sample data until the model verification result corresponding to the target intent model meets the preset requirements.
7. The user intention analysis method according to claim 6, characterized in that: The acquiring current user speech data, inputting the current user speech data into the target intention model, and outputting the user intention analysis result specifically includes: Acquire current user speech data, and call the model address according to the current user speech data; Determine the target intent model according to the model address, and input the current user speech data into the target intent model; Perform intent analysis on the current user speech data according to the target intent model to obtain an intent analysis result.
8. A user intention analysis system, characterized in that: The user intention analysis system comprises: A skill task creation module, used to determine multiple skill tasks and create an intention description and example sample data corresponding to each of the skill tasks; An intention sample data generation module, used to determine a first preset large model, and input the intention description and the example sample data into the first preset large model to obtain a plurality of intention sample data corresponding to the plurality of skill tasks; A target intent model training module, used to determine a second preset large model and target intent sample data from the plurality of intent sample data, and train the second preset large model according to the target intent sample data to obtain a target intent model; The user intention analysis result generation module is used to obtain the current user speech data, input the current user speech data into the target intention model, and output the user intention analysis result.
9. A terminal, characterized in that: The terminal includes: a memory, a processor, and a user intention analysis program stored in the memory and executable on the processor. When the user intention analysis program is executed by the processor, the steps of the user intention analysis method as described in any one of claims 1 to 7 are implemented.
10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a user intention analysis program, and when the user intention analysis program is executed by a processor, the steps of the user intention analysis method according to any one of claims 1 to 7 are implemented.