Model prediction method and device, intelligent agent, equipment and storage medium

Abstract

The invention provides a model prediction method and device, an intelligent agent, equipment and a storage medium, and relates to the technical field of artificial intelligence, in particular to the fields of large models, intelligent agents and the like. According to the specific implementation scheme, model recommendation information is obtained according to demand information; in response to a selection command for the model recommendation information, obtaining a target model; and inputting data needing to be processed into the target model for processing to obtain an output result.

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence technology, and particularly to the fields of large models, intelligent agents, etc. Background Art

[0002] An intelligent agent is an intelligent application or entity that can perceive the environment, make decisions, and interact with the environment, and can affect the environment by performing actions. An intelligent agent can be physical or virtual. An intelligent agent usually has one or more goals or tasks, and these goals can be achieved through an intelligent decision-making process. Intelligent agents and artificial intelligence models such as large models can be used in combination. For example, an intelligent agent can be used as an interface for a large model to achieve natural language interaction, receive a user's natural language input, convert it into a format that the large model can understand, and then convert the output of the large model into natural language and feedback it to the user. The large model can provide knowledge and ability support for the intelligent agent. Summary of the Invention

[0003] This disclosure provides a model prediction method, apparatus, intelligent agent, device, and storage medium.

[0004] According to one aspect of this disclosure, there is provided a model prediction method, including:

[0005] Obtaining model recommendation information according to demand information;

[0006] Obtaining a target model in response to a selection command for the model recommendation information;

[0007] Inputting the data to be processed into the target model for processing to obtain an output result.

[0008] According to another aspect of this disclosure, there is provided a model prediction apparatus, including:

[0009] A first obtaining module, configured to obtain model recommendation information according to demand information;

[0010] A second obtaining module, configured to obtain a target model in response to a selection command for the model recommendation information;

[0011] An input module, configured to input the data to be processed into the target model for processing to obtain an output result.

[0012] According to another aspect of this disclosure, there is provided an intelligent agent, including:

[0013] An input module, configured to input demand information and the data to be processed;

[0014] A processing module, configured to call a large model to execute the model prediction method according to any embodiment of the present disclosure to process the received requirement information and the data to be processed by the input module, and obtain an output result;

[0015] An output module, configured to output the output result obtained by the processing module.

[0016] According to another aspect of the present disclosure, there is provided an electronic device, including:

[0017] At least one processor; and

[0018] A memory communicatively connected to the at least one processor; wherein,

[0019] The memory stores instructions executable by the at least one processor, and when the instructions are executed by the at least one processor, the at least one processor is enabled to execute any method in the embodiments of the present disclosure.

[0020] According to another aspect of the present disclosure, there is provided a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause the computer to execute any method in the embodiments of the present disclosure.

[0021] According to another aspect of the present disclosure, there is provided a computer program product, including a computer program, which when executed by a processor, implements any method in the embodiments of the present disclosure.

[0022] It should be understood that the content described in this part is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it used to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. Description of the Drawings

[0023] The drawings are used to better understand the solution and do not constitute a limitation to the present disclosure. Among them:

[0024] Figure 1 is a flowchart of a model prediction method according to an embodiment of the present disclosure;

[0025] Figure 2 is a flowchart of a model prediction method according to another embodiment of the present disclosure;

[0026] Figure 3 is a flowchart of a model prediction method according to another embodiment of the present disclosure;

[0027] Figure 4 is a flowchart of a model selection method according to another embodiment of the present disclosure;

[0028] Figure 5It is a schematic flowchart of performing function judgment according to an embodiment of the present disclosure;

[0029] Figure 6 It is a schematic flowchart of a model prediction method according to an embodiment of the present disclosure;

[0030] Figure 7 It is a schematic diagram of an agent interaction interface in the model recommendation process;

[0031] Figure 8 It is a schematic flowchart of an agent work process according to an embodiment of the present disclosure;

[0032] Figure 9 It is a schematic structural diagram of engineering layering according to an embodiment of the present disclosure;

[0033] Figure 10 It is a schematic flowchart of session management according to an embodiment of the present disclosure;

[0034] Figure 11 It is a schematic flowchart of performing function judgment according to an embodiment of the present disclosure;

[0035] Figure 12 It is a schematic flowchart of model prediction according to an embodiment of the present disclosure;

[0036] Figure 13 It is a schematic flowchart of progress query according to an embodiment of the present disclosure;

[0037] Figure 14 It is a schematic flowchart of template selection according to an embodiment of the present disclosure;

[0038] Figure 15 It is a schematic flowchart of prediction task progress query according to an embodiment of the present disclosure;

[0039] Figure 16 It is a schematic flowchart of prediction task startup according to an embodiment of the present disclosure;

[0040] Figure 17 It is a schematic diagram of a status field according to an embodiment of the present disclosure;

[0041] Figure 18 It is a schematic structural diagram of a model prediction device according to an embodiment of the present disclosure;

[0042] Figure 19 It is a schematic structural diagram of a model prediction device according to another embodiment of the present disclosure;

[0043] Figure 20 It is a schematic structural diagram of an agent according to an embodiment of the present disclosure;

[0044] Figure 21It is a block diagram of an electronic device for implementing the embodiments of the present disclosure. Detailed implementation manners

[0045] The following describes exemplary embodiments of the present disclosure with reference to the accompanying drawings. Various details of the embodiments of the present disclosure are included to facilitate understanding, and they should be considered merely exemplary. Therefore, those of ordinary skill in the art should recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope of the present disclosure. Similarly, for the sake of clarity and conciseness, the description of well-known functions and structures is omitted below.

[0046] Figure 1 It is a flowchart of a model prediction method 100 according to an embodiment of the present disclosure. In one implementation, the method may include:

[0047] S101. Obtain model recommendation information according to demand information;

[0048] S102. Obtain a target model in response to a selection command for the model recommendation information;

[0049] S103. Input data to be processed into the target model for processing to obtain an output result.

[0050] In the embodiments of the present disclosure, an agent can perceive the environment and take actions to achieve specific goals. The agent can have autonomy, adaptability, and interaction capabilities. The agent perceives changes in the environment, makes judgments and decisions based on the knowledge and algorithms learned by itself, and then executes actions to affect the environment or achieve a predetermined goal.

[0051] In the embodiments of the present disclosure, in an agent system, multiple types of agents can be included. For example, a label design agent, a data analysis agent, etc. Different agents can use different interaction interfaces. For example, the data analysis agent can analyze the data uploaded by the client through the data analysis agent interface. The label design agent can add appropriate labels to the data through the label design agent interface. Data between some agents can be transferred or shared with each other. The agent interaction interface can display various information on the interaction between the user and the agent through the client. For example, the agent interaction interface can prompt the user to provide some content through the client, such as data to be processed, requirements to be met, etc. The requirement information of the client can include content such as text, images, files, etc. directly input by the user through the client on the agent interaction interface, and can also include key information obtained after the large model understands the content provided by the user through the client. For example, the content provided by the user through the client, such as the requirement description, etc., is input into the large model for operations such as requirement understanding and / or intention recognition to obtain the requirement information of the client. The content provided by the user through the client can include, but is not limited to: model scope, basis for model selection, number of models, model application, etc. Then, model recommendation information can be obtained through agent planning based on the requirement information of the client. For example, the requirement information of the client can be sent to the planner of the agent. The planner of the agent has functions such as a chain of thought and sub-goal decomposition. By decomposing and thinking about the requirement information through the planner of the agent, recommended model information that meets the requirement information can be obtained. The recommended model information can include specific information of one or more recommended models. For example, one or more of the capabilities, names, descriptions, etc. of the recommended models, and the recommended model information can also include some optional functions, such as a model removal function, a model addition function, a model confirmation function, etc.

[0052] The model recommendation information can be displayed on the agent interaction interface. The model recommendation information can include specific information of one or more recommended models. For example, the names of multiple recommended models are displayed. For another example, the names, capabilities, and descriptions of multiple recommended models are displayed. The user can select the target model they need through the client, and select one or more target models in the recommended model information through a selection command. It is also possible to finally obtain the target model by means of removal, addition, confirmation, etc.

[0053] On the intelligent agent interaction interface, the user can be prompted to input the data to be processed through the client. This step can be executed after selecting the target model or before displaying the recommended models, and can be flexibly selected according to the actual application scenario. For example, based on the data to be processed, the demand information of the client can be identified, and then the model recommendation and subsequent steps can be executed based on the demand information of the client. Another example is that based on the data to be processed, a part of the demand information of the client can be identified, and then the model recommendation and subsequent steps can be executed in combination with another part of the demand information provided by the client. Through the intelligent agent according to the demand information of the client, the recommended model information that meets the client's needs can be planned, which is convenient for the client to select the required target model and accurately process the client data.

[0054] Figure 2 FIG. 200 is a schematic flowchart of a model prediction method 200 according to another embodiment of the present disclosure. The method 200 can be used to implement step S101 in the model prediction method 100. In one implementation, the method 200 includes: obtaining model recommendation information according to the demand information, further including:

[0055] S201. Invoke the planner of the intelligent agent to plan the demand information to obtain the model selection tool to be invoked;

[0056] S202. Invoke the model selection tool to obtain the recommended model information that meets the demand information.

[0057] In the embodiments of the present disclosure, the planner of the intelligent agent (Planner) can also be referred to as a planning module, and can adopt the Reasoning and Acting (ReAct) mode to process the demand information of the client using a large model, simulate the reasoning and acting process of human intelligence, and alternately generate Verbal Reasoning Traces and Actions. And, it allows the large language model to interact with external tools to obtain additional information feedback. During the reasoning and acting process, one or more tools that need to be used may be involved. For example, if a model selection tool needs to be used, the planner can call the model recommendation tool according to the planned reasoning trace to execute the model recommendation task to obtain the model recommendation information. Through the planner, the action plan of the intelligent agent can be dynamically maintained and updated, the model selection tool can be reasonably called, and the model recommendation information can be provided quickly and accurately.

[0058] In one implementation, obtaining the recommended model information that meets the demand information includes:

[0059] Sending a request to obtain a model list to the model data set;

[0060] Receive the first model list returned by the model dataset;

[0061] According to the first model list and the requirement information, assemble the prompt words for the large model, which are used to ask the large model for model recommendation results;

[0062] Input the prompt words into the large model to obtain the recommended second model list returned by the large model.

[0063] In the embodiments of the present disclosure, the model selection tool can send a request for obtaining a model list to the model dataset, and this request can be used to obtain the key information of all models supported by the client. The model selection tool can receive the first model list returned by the model dataset, and the first model list can include one or more of the key information such as the names, identifiers, important parameters, storage addresses, etc. of all models supported by the client. The model selection tool can assemble prompt words according to the first model list and the requirement information of the client, and then call the large model to output the recommended second model list according to the prompt words. After the model selection tool obtains the second model list from the large model, it can submit the second model list to the planner. The planner can directly use the second model list as the recommendation result and send it to the front end for display on the agent interaction interface; it can also perform other processing on the second model list and then send it to the front end for display on the agent interaction interface. The planner can perform the next round of planning based on the execution result of this tool, and sequentially complete the actions required for the inference trajectory. Through the model dataset and the large model, a more demand-information-compliant recommended model list can be obtained, facilitating the client to select the target model suitable for processing the client's data.

[0064] In the embodiments of the present disclosure, the large models used in different stages can be the same or different, and can be flexibly set according to the actual application scenario requirements.

[0065] In one implementation, obtaining the recommended model information that meets the requirement information further includes:

[0066] Send a model filtering request to the model filtering module, and this model filtering request is used to request filtering of the second model list;

[0067] Receive the third model list recommended after the model filtering module filters the second model list.

[0068] In the embodiments of the present disclosure, after the model selection tool obtains the second model list from the large model, it can filter the second model list. For example, the model selection tool sends a model filtering request to the model filtering module to request the model filtering module to filter the second model list according to key information such as the model name. The model names used for filtering may include the names of all models supported by the client, and these names can be set in advance or updated regularly. The model names used for screening can also be determined based on the first model list returned by the model dataset. In one case, the names of all models in the first model list are used as the model names for filtering. For example, if the preset model names for filtering include models M1, M2, M3, M4, and M5, and the recommended second model list includes M1, M2, M3, M6, and M7. In this case, M6 and M7 can be filtered out, and only M1, M2, and M3 are retained. Another example, if the first model list returned by the model dataset includes models M1, M2, M3, M4, M5, M6, and M7, and the recommended second model list includes M1, M2, M3, M6, M7, and M8. In this case, M8 can be filtered out, and only M1, M2, M3, M6, and M7 are retained.

[0069] After the model filtering module filters the second model list to obtain the filtered third model list, it can return the third model list to the model selection tool. The model selection tool can submit the third model list to the planner. The planner can send the third model list as a recommended result to the front end for display on the agent interaction interface. By filtering the initial model recommendation list, the final model recommendation list that better meets the client's demand information can be retained, facilitating the client to select a target model more suitable for processing the client's data.

[0070] In one implementation, as Figure 2 shown, the method 200 can be used to implement step S103 in the agent-based model prediction method 100. In one implementation, the method 200 includes: inputting the data to be processed into the target model for processing, further including:

[0071] S203. Starting a prediction task in response to a confirmation command for the prediction task start information;

[0072] S204. Inputting the data to be processed into the target model for prediction in response to the prediction task.

[0073] In the embodiments of the present disclosure, after selecting a target model, the data to be processed uploaded by the client can be processed by the target model. In the agent interaction interface, the prediction task start information can be displayed first to prompt the user to start the prediction task. If the user confirms to start the prediction task, the client (front end) can send a command to the agent to confirm model prediction, that is, a confirmation command for the prediction task start information. If there is a target model that needs to take effect, the agent can send the target model that needs to take effect to the prediction service. The agent can also instruct the prediction service to start the prediction task. After starting the prediction task, the target model can predict the data that the client needs to process to obtain a prediction result. For example, the type of the target model can include one or more of a composite label model, a label classification model, etc. The prediction result output by the composite label model can include the label combination of the data in different dimensions. The prediction result output by the label classification model can include the label classification to which the data belongs.

[0074] Through the interaction in the agent interaction interface, the prediction task can be flexibly started when the user needs it, and the target model can be used to process the data uploaded by the client, which can meet more diverse application scenarios. In the embodiments of the present disclosure, the data uploaded by the client can participate in the model recommendation process. In this case, the data can be first input into the planner to plan the recommended model information in combination with demand information, etc., and then input into the target model for prediction; the data can also not participate in the model recommendation process. In this case, the data is only input into the target model for prediction.

[0075] Figure 3 It is a schematic flowchart of a model prediction method 300 based on an agent according to another embodiment of the present disclosure. In one implementation, the method may further include:

[0076] S301: Call the planner of the agent to plan the progress query intention to obtain the progress query tool to be called;

[0077] S302: Call the progress query tool to query the prediction task progress information.

[0078] In the embodiments of the present disclosure, the planner can call the progress query tool to query the progress information of the prediction task. For example, if the user wants to know the current execution status of the prediction task, when the user clicks the progress query icon on the agent interaction interface, a progress query intention (or called task progress intention, query progress intention, query progress request, etc.) can be generated. After the user sends the progress query intention to the planner through the client, the planner can plan the progress query intention to obtain the inference trajectory and actions, and determine that the tool to be called is the progress query tool. The planner can call the progress query tool to execute the progress query of the prediction task to obtain the prediction task progress information. In addition, the execution result of the prediction task and / or the prediction task progress information can also be polled regularly. For example, the prediction task progress information can include one or more of the task name, task execution time, estimated end time, whether abnormal, etc. For another example, the execution result of the prediction task can include one or more of the data prediction result, analysis report, task execution completed, task execution successful, task execution failed, etc. By planning the progress query intention through the planner, the progress query tool can be called to query the prediction task progress information, which is convenient for the user to timely know the progress of the prediction task.

[0079] In one implementation, querying the prediction task progress information includes:

[0080] Sending a prediction task progress query request to the prediction service;

[0081] Receiving the prediction task progress information returned by the prediction service.

[0082] In the embodiments of the present disclosure, the agent can start the prediction task through the prediction service, and then use various large models or other tools to execute the prediction task. The prediction service can record the situation of each prediction task, such as whether it is started, the execution situation of the prediction task, the execution tool of the prediction task, etc. The planner of the agent can call the progress query tool to send a prediction task progress query request to the prediction service, and the request can include one or more of the basic information of the prediction task to be queried, such as name, number, description, etc. If the prediction service includes the record of the prediction task, the prediction service can return the progress query result to the planner, which can include the prediction task progress information and / or the prediction task execution result. The prediction task progress information can be quickly queried in the prediction task, improving the query efficiency.

[0083] Figure 4 FIG. 400 is a schematic flowchart of a model selection method 400 based on an agent according to another embodiment of the present disclosure. In one implementation, the method may further include:

[0084] S401, obtaining a first prompt information; the first prompt information is used to prompt to provide the data to be processed;

[0085] S402. In response to the data providing command, obtain the data to be processed provided according to the first prompt message.

[0086] In the embodiments of the present disclosure, various prompt messages can be used in the agent interaction interface to prompt the user to perform various operations through the client. The agent interaction interface can include conversation forms such as questionnaires and dialogues. For example, the first prompt message displayed in the agent interaction interface can prompt the client to provide the data to be processed. For example, the first prompt message can include "Hello! Please provide the data you want to analyze, and I will select a suitable model for data analysis according to your needs." When the user uploads data through the client in the agent interaction interface, a data providing command can be generated. In response to the data providing command, the uploaded data, the information of the agent, etc. will exist in the session of the agent. The data uploaded by the client can participate in the model recommendation process or not. Prompting the client to provide the data to be processed through the interface interaction method can support the user to flexibly upload data through the client and meet the requirements of various application scenarios.

[0087] In one implementation, as Figure 4 shown, the method may further include:

[0088] S403. Obtain a second prompt message; the second prompt message is used to prompt the provision of the requirement information;

[0089] S404. In response to the requirement providing command, obtain the requirement information provided according to the second prompt message.

[0090] In the embodiments of the present disclosure, a second prompt message can be displayed in the agent interaction interface to prompt the user to provide requirement information through the client. The user can provide some requirement descriptions, and then obtain the requirement information of the client by analyzing the requirement descriptions through a large model, etc. The user can also directly provide the requirement information of the client. For example, the second prompt message displayed in the agent interaction interface can include "I have successfully received your data. Please describe in detail which aspects you specifically want to analyze from the data." Through one or more rounds of conversations, guide the user to input or select requirements through the client. After the user inputs or selects requirement descriptions such as "which aspects need to be analyzed" through the client in the agent interaction interface, the client can send the requirement description to the agent, and the agent can understand the requirement description through a large model to obtain the requirement information. Then enter the model recommendation process and recommend a list of models according to the requirement information. Prompting the client to provide the data to be processed through the interface interaction method can support the user to flexibly upload data through the client and meet the requirements of various application scenarios.

[0091] In one implementation, the method may further include one of the following:

[0092] If there is a target agent record in the session of the agent, and when the task type in the sub - session of the target agent is model training, model training is executed;

[0093] If there is a target agent record in the session of the agent, and when the task type in the sub - session of the target agent is model prediction, model prediction is executed;

[0094] If there is no target agent record in the session of the agent, and there is a target agent record in the agent analysis report, model training is executed;

[0095] If there is no target agent record in the session of the agent, and there is no target agent record in the agent analysis report, model prediction is executed.

[0096] In the embodiments of the present disclosure, the session of the agent may include the interaction content between the user and the agent through the client. According to the session of the agent, the task type, the analysis report of the agent, etc., it can be determined whether to execute model prediction (or model selection) or model training.

[0097] See Figure 5 , an exemplary judgment process includes:

[0098] S501. Determine whether there is a target agent record in the session of the agent. If so, execute S502; otherwise, execute S503.

[0099] S502. Obtain the task type in the sub - session of the target agent. If the task type is model training, execute S504; if the task type is model prediction, execute S505.

[0100] S503. Determine whether there is a target agent record in the agent analysis report. If so, execute S504; otherwise, execute S505.

[0101] S504. Execute model training.

[0102] S505. Execute model prediction.

[0103] In the embodiments of the present disclosure, taking the target model as the data analysis agent as an example, if the session of the agent, such as agent_session, includes a data analysis agent record, the task type can be further obtained in the sub - session of the target agent, such as agent_analysis_sub_session. If the task type is model training, the data analysis agent can start the model training process. If the task type is model prediction, the data analysis agent can start the model prediction process.

[0104] In the embodiments of the present disclosure, if the session of the agent, such as agent_session, does not include the data analysis agent record, it is possible to further check whether the analysis report of the agent, such as agent_analysis_report, includes the data analysis agent record. The analysis report of the agent may include reports on the prediction or analysis of input data by the agent using various models. If there is no data analysis agent record in both the session and the analysis report of the agent, it indicates that the agent has not been used for analysis or prediction yet, and no model training task has been created based on the agent. In this case, the model prediction process can be entered. If there is no data analysis agent record in the session, but there is a data analysis agent record in the analysis report of the agent, it indicates that the agent has been used, and the agent can be used to train the model, and the model training process can be entered.

[0105] In the embodiments of the present disclosure, the model prediction process can refer to the relevant descriptions in the above embodiments. An example of a model training process may include: displaying label interaction information corresponding to the first label on the agent interaction interface; obtaining the label annotation result of the client according to the label interaction information; generating a training set according to the label annotation result; training the initial model according to the training set to obtain the target model. Generating a validation set according to the label annotation result; validating the target model according to the test set.

[0106] Figure 6 is a schematic flowchart of a model prediction method according to an embodiment of the present disclosure. This method can reasonably plan the selected model according to the user's needs and automatically create a data prediction task to help the user automatically process the data to be analyzed and obtain the data prediction result. As Figure 6 shown, the method includes:

[0107] S601. The user provides data analysis requirements and corresponding data (upload data).

[0108] S602. The large model understands the requirements.

[0109] S603. Plan the model capabilities, such as which models to use for prediction & what to predict.

[0110] S604. The user confirms which models to select for prediction.

[0111] S605. The agent automatically creates a prediction task for direct prediction.

[0112] S606. Obtain the data prediction result.

[0113] I. Function description of the user directly using the data analysis agent.

[0114] (1) Default display mode:

[0115] The default page of the "Data Analysis Agent" can display the Agent welcome card by default. For example, the welcome copy can include: Hello! I'm your data analysis consultant, and I'm glad to serve you. You can give me some data you want to analyze first, and later I will select a suitable model according to your needs to perform data analysis for you.

[0116] (2) Data upload.

[0117] Users first submit data to the Agent through the "Send Data" button at the bottom of the card. For example, the data selection prompt can include: "You can select existing data in the system or send me a data file." An example of the data upload logic is as follows.

[0118] 1. Users can choose "Data in the system" or "Table import".

[0119] Data in the system: Support users to filter data in the data management of the analysis end in the system.

[0120] Table import: Support users to upload data through an Excel file.

[0121] (1) Support file upload (refer to table import data in the management end insight Agent).

[0122] (2) The data submitted by users through file upload needs to be synchronized to the "Analysis end" - data management so that users can view the prediction results in data management after the prediction is completed.

[0123] 2. If the user sends a message to the Agent without selecting data, the Agent can reply to the user: "You haven't given me any data yet. Please give me some data first so that I can help you with data analysis."

[0124] 3. After the user provides the data, the upload data button can change color, for example, be grayed out, to prevent the user from providing data repeatedly.

[0125] (1) During the data import process, the Agent replies to the user: "Receiving your data, please wait a moment."

[0126] (2) If the data is not successfully imported, the Agent replies to the user: "Sorry, I couldn't understand the data you provided. Please provide me with another copy of the data."

[0127] (3) If the data is successfully imported, the Agent replies to the user: "Your data has been successfully received. Please describe in detail what aspects you specifically want to analyze from the data" to guide the user to send requirements.

[0128] (3) Requirement Analysis.

[0129] The user sends the analysis requirements to the Agent, and the large model understands the requirements described by the user and plans to call which models to help the user with predictive analysis.

[0130] Examples of analysis requirements are as follows:

[0131] Model scope: Select from the models in the template market.

[0132] Basis for model selection: The large model selects the model that matches the model description from the template market according to the user's requirement description.

[0133] Number of models: At least 1 and at most 10 (consistent with the limit on the number of models that can be selected for the analysis task).

[0134] Model application: After the user confirms the selected model and clicks "Confirm", the Agent will automatically apply the selected model to the user.

[0135] (4) Model Planning.

[0136] After the model planning is completed, as Figure 7 shown, the Agent sends one or more of the following relevant model information to the user for selection and confirmation.

[0137] Model capabilities: Currently matching the model capabilities in the template market, for example, models such as "Compound Label Model" and "Classification Label Model" can be supported.

[0138] Model name: The user can directly click on the "Model Name" to jump to the model management in the management terminal to view the model details.

[0139] Model description: The model description corresponding to the model

[0140] "Remove Model": The user can click on the "X" in the upper right corner of the model card to remove the model from the model card list.

[0141] "Add Model": The user clicks on buttons such as "+", adds and selects other models. After clicking, a pop-up window for adding models will appear. The added models can be various types of models that have not been selected by the large model in the template market. The user can click to add the models they have added from the model list, and after clicking "Add", the model card of the added model will be added to the model information list sent by the Agent. If the model description cannot be fully displayed in the list, it will be omitted, and the complete model description will be displayed when hovering the mouse (hover).

[0142] "Confirmation": After the user clicks "Confirm", the Agent replies to the user: "The model confirmation is successful. Now, data analysis and prediction will start, which will take some time. I will send you a message to inform you after the prediction is completed. Please wait patiently for me."

[0143] In addition, after the model confirmation is successful, a data prediction task can be created automatically or manually in "Analysis Endpoint" - Data Prediction. After the task creation is completed, the data prediction list in the "Analysis Endpoint" can synchronously add the prediction task created by the Agent.

[0144] (V) Data Prediction.

[0145] After the Agent creates a prediction task, data prediction is automatically started.

[0146] ○ If the data prediction fails, the Agent replies to the user: "Sorry, the data prediction task was not successful. You can go to 'Analysis Endpoint - Data Prediction' to create it manually again." The user can directly click on 'Analysis Endpoint - Data Prediction' in the message to jump to the data prediction page for manual attempt.

[0147] ○ If the data prediction is completed, the Agent replies to the user: "The data analysis and prediction are completed. Please 'View Prediction Results'. If you need to view the BI report of the data prediction results, you can generate the report and view it in 'Analysis Endpoint - Intelligent Analysis'."

[0148] ■ When the user clicks "View Prediction Results", they are redirected to "Analysis Endpoint - Data Management" to view the data prediction results.

[0149] ■ When the user clicks "Analysis Endpoint - Intelligent Analysis", they are redirected to "Analysis Endpoint - Intelligent Analysis", and the dropdown box for selecting data is default filled

[0150] (VI) An example of the task end logic is as follows:

[0151] When the following four states occur, it is considered that the current task has ended. The Agent returns to the initial state and replies to the user: "This data analysis task has ended. If you have any further data analysis needs in the future, you can come to me again."

[0152] (1) When the Agent has no output result.

[0153] (2) When the user's conversation input indicates an intention to end the task.

[0154] (3) When the user has no operation or reply within 24 hours.

[0155] Note: Support for the user to manually end the current task. During the task process, a "Reset" button is displayed above the dialog box.

[0156] When the mouse hovers, display the prompt text: "Click the button to end the current data analysis task."

[0157] When the user clicks the button, the current task ends and the Agent returns to its initial state.

[0158] The Agent creates the data results of the prediction task, and the user can directly send a request to generate a report.

[0159] (VII) Data Flywheel

[0160] Examples of the Agent's dialogue effects are as follows:

[0161] 1. Evaluation feedback: After each label design task ends, the Agent pushes a request for the user to provide evaluation feedback and stores the user's evaluation feedback data.

[0162] (1) For example, the reply text can include: Are you satisfied with my service this time? Please evaluate my work. I will continuously optimize my capabilities to provide you with better service.

[0163] (2) For example, the evaluation feedback card can include one or more of the following:

[0164] Comprehensive rating: Support the user to select from one star to five stars.

[0165] Evaluation feedback: A text input box that supports the user to input evaluation feedback within 500 words, for example.

[0166] "Submit evaluation": Click the button to submit the evaluation. After the evaluation is successful, the card updates to display a "successful evaluation" prompt.

[0167] 2. Interaction record: For each sentence of the Agent's reply to the user, "like / dislike" buttons can be added, and all the dialogue data of the user's interaction with the Agent is recorded, along with the like / dislike information.

[0168] For example, when the mouse hovers over the Agent's reply content, "like / dislike" floating buttons are displayed, and the user can click to operate, and click again to cancel.

[0169] II. Overall process. Figure 8 It is a schematic diagram of the intelligent agent work process according to an embodiment of the present disclosure. In one implementation, it may include:

[0170] S801. The user (the client or front end of the user, hereinafter referred to as the user) queries historical information through the intelligent agent microservice (voc-agent).

[0171] Create a WebSocket connection between the user and the voc-agent. There are model training-related interfaces between the user and the voc-agent.

[0172] The user uploads the data to be predicted to the voc-agent.

[0173] The user sends a requirement description to the voc-agent.

[0174] The voc-agent requests the planner to make a plan.

[0175] The planner asynchronously schedules tools.

[0176] One or more of the above steps may have corresponding response steps. For example, the response information can be carried by a return message.

[0177] III. Project Layers

[0178] Figure 9 It is a schematic diagram of the project layer structure according to an embodiment of the present disclosure. A reasonable project layer can avoid circular dependencies in the project and improve the maintainability of the code. The project layers executed by the data analysis agent may include: the WS message processing and sending layer 901, the access layer 902 (which may include a Controller), the agent business logic processing layer 903 (which may include a Service), the execution layer 904 (which may include modules such as a Planner and an Action), and the database DAO (Data Access Object) layer 905.

[0179] The services in the upper layer depend on the services in the lower layer, and the infrastructure layer (WS message processing or database dao layer) can be depended on by all layers.

[0180] The WebSocket (ws) message processing and sending layer 901 can accept the input data and requirement information from the client, and can also send the task processing progress, task processing results, and user operation guidance information to the client. The access layer 902 can connect the ws message processing and sending layer 901, the agent business logic processing layer 903, and the database DAO layer 905. The access layer 902 can forward the requirement information received by the ws message processing and sending layer 901 to the agent business logic processing layer 903. The access layer 902 can send the received input data to the database DAO layer 905, and can also obtain the data traced back according to the label from the database DAO layer 905. The agent business logic processing layer 903 can process the requirement information, call the execution layer 904 according to the processing results of the requirement information, and obtain the execution results of the execution layer 904 (such as the template recommendation information obtained by planning).

[0181] IV. Session Management

[0182] 1. For the details of the ws protocol between the front end and the back end, see Figure 10 。

[0183] S1001. The ws client sends an initialization message to the ws server. The ws server determines whether to enter the prediction process or the training process. The ws server can return a response to the ws client.

[0184] S1002. The ws client sends a chat message to the ws server to enter a chat conversation. The ws server can return a response to the ws client.

[0185] In the disclosed embodiments, the session management can solve the problem of which session the current conversation belongs to. For the session management in the data analysis agent, there are the following problems:

[0186] 1: When to add a record in the agent's session (agent_session)

[0187] In the interface for querying the mapping relationship between the agent type (agent_type) and the session identifier (sessionId). If the current sessionId does not exist, a new record is added to the agent_session.

[0188] 2: How to determine whether to enter the training process or the prediction process

[0189] It is judged through the data analysis sub-session (agent_analysis_sub_session) and the data analysis report (agent_analysis_report) table. An example of the judgment logic is Figure 11As shown

[0190] After the user enters the data analysis agent, which task (training task and prediction task) should be entered? The judgment process example is as follows:

[0191] S1101. Query whether there is a record of the data analysis agent (agent) in agent_session. If there is no record, execute S1102. If there is a record, execute S1103.

[0192] S1102. Query whether there is such a record in the data analysis agent report (agent_analysis_report) table (query according to the latest labelsub_session_id).

[0193] If there is no record, execute the prediction process. For example, insert data into agent_analysis_sub_session (prediction task), and then return a reply to enter the data prediction task to the front end.

[0194] If there is a record, execute the training process. For example, insert data into the agent_analysis_sub_session table (training task), and then push relevant replies for model training.

[0195] S1103. Query the data analysis agent's sub-session (agent_analysis_sub_session) table.

[0196] Judge the current task_type field in this table. If it is training, it means that the current is in the training task and enter the training process; if it is predict, it means that the current is a prediction task and enter the prediction process.

[0197] V. Engineering implementation

[0198] Considering that the label design agent and the data analysis agent need to perform historical data queries and establish websockets. And subsequent agents also have this function, but the corresponding logic in each agent is different. The strategy pattern can be considered for implementation. The example is as follows:

[0199] AgentContext: Holds a reference to a strategy class for client use.

[0200] AgentStrategy: The agent strategy class, defined as an interface.

[0201] LabelAgentStrategy and AnalysisAgentStrategy: Specific strategy classes that hold wsPushService and are used to push ws messages.

[0202] The method definitions are as follows:

[0203] initSubSession: Initialize the session.

[0204] finishSubSession: Terminate the session.

[0205] At the same time, decouple message reception and message sending to avoid circular dependencies.

[0206] Compared with the previous implementation, the following benefits are brought: clearly define the agent interface, which is convenient for upper-layer calls. Improve the maintainability of the code.

[0207] An example of a ws message processing logic is as follows, which can solve the following problems:

[0208] 1. There is a coupling between message reception and sending, and there is a circular dependency problem in the code..

[0209] 2. The processing logics of message reception and sending are different, and the entity objects are different. From an object-oriented perspective, they should be distinguished.

[0210] The processing of ws messages can be split, and the split example is as follows:

[0211] For example, after splitting, WsHandleMessage: The message reception processing class, including three methods, namely: handleInitMessage, handleInfoMessage, handleChatMessage, which are used to process init, info, and chat type messages respectively.

[0212] ○ init: Session initialization message.

[0213] ○ info: Control class message, such as terminating the current session and setting the current query as read.

[0214] ○ chat: Session message.

[0215] WsPushMessage: The user sends ws messages, including two public methods, namely: sendInfoMessage, sendChatMessage. They are used to send info and chat type messages respectively.

[0216] ○ Classes that need to send messages externally all depend on the WsPushMesage class.

[0217] VI. Model Prediction Process

[0218] As Figure 12 shown, an example of a model prediction process may include:

[0219] S1201. Create an initial websocket connection. The voc-agent needs to determine which task to enter currently based on the status in the backend database.

[0220] The welcome message returned by the websocket returns different contents according to different statuses.

[0221] S1202. Enter the model prediction process, and the user uploads data. If it is a table import, the data needs to be uploaded to the analysis end.

[0222] S1203. After the data upload is completed, the system sends a prompt message; the user uploads a description of the requirements and sends a request.

[0223] S1204. The voc-agent calls the planner, and the planner outputs the tools to be called.

[0224] a. Currently, there can be multiple types of tools, such as: template list tool and prediction progress query tool.

[0225] b. The planner can reply content to the front-end user.

[0226] S1205. The tool performs asynchronous execution. Taking the template list tool as an example.

[0227] S1206. The tool first obtains the template list, and then assembles the prompt according to the user's needs. Call the large model to obtain the recommended template list.

[0228] Then, hand over the execution result of this tool to the planner for the next round of planning. The output result of the planner is filtered (for example, filtered by template name) to obtain the final template list.

[0229] S1207. Push the template list to the user.

[0230] Currently, the execution of all tools adopts asynchronous execution, and a thread pool is used for the execution of the template selection and progress query tools.

[0231] 6.1 planner (analysisPlanner) of the data analysis agent

[0232] Goal: Plan the tools to be used according to the request, including template selection tool and progress query tool. The Planner adopts the ReAct mode, and the process is as follows Figure 13 as shown.

[0233] S1301. The user can send a query request (query) to the Planner.

[0234] S1302. The Planner plans the execution asynchronously. This process can call the Tool.

[0235] S1303. The Planner returns the planned response to the user.

[0236] S1304. The Planner receives the execution result of the tool.

[0237] S1305. The Planner sends the final result obtained from the execution result of the tool to the user.

[0238] In the above process, the Plannerprompt template of the data analysis agent can be obtained through the Plannerprompt template link.

[0239] 6.2 Template selection tool (modelSelectTool)

[0240] Function: Call the large model interface to obtain the recommended template list in the template center; assemble the front-end response. The example process is as follows Figure 14 as shown.

[0241] S1401. The Planner calls the template selection tool to send a request to obtain the template list data to the microservice dataset (voc-java-datatset).

[0242] S1402. The template selection tool receives the result returned by the voc-java-datatset.

[0243] S1403. The template selection tool sends a request to the large model to ask for template recommendations.

[0244] S1404. The template selection tool receives the recommended model list returned by the large model.

[0245] S1405. The template selection tool sends a template filtering request to the template filtering module.

[0246] S1406. The template selection tool returns the filtering result to the template filtering module. And returns the filtered template recommendation list to the Planner.

[0247] Among them, the example of the large model prompt content can include:

[0248] User requirement description;

[0249] Template information: template name, template usage scenario.

[0250] If the candidate template list is too long, it may cause the length of the large model prompt window to exceed the limit; at the current stage, consider truncating the prompt window. Subsequently, the small model can be used for screening first, and then the large model can be requested.

[0251] The model can be published as a template, and there can be no restrictions on the content filled in by users. To optimize the template recommendation results of the large model, an approval process can be introduced.

[0252] 6.3 Progress query tool (predictProcessTool)

[0253] Function: Query the progress of the prediction task, and the process is as Figure 15 shown.

[0254] S1501. The user sends a task progress intention to the planner.

[0255] S1502. The planner returns a response to the user.

[0256] S1503. The planner calls the query tool, and this step can be executed asynchronously.

[0257] S1504. The progress query tool sends a request to query the progress information to the prediction microservice (voc-java-predict).

[0258] S1505. The progress query tool receives the progress information returned by voc-java-predict.

[0259] S1506. The progress query tool returns the progress information to the planner.

[0260] S1507. The planner returns the progress information to the user.

[0261] 6.4 Start the model prediction task

[0262] See Figure 16 , and the process of starting the model prediction task can include:

[0263] S1601. The front end sends information to confirm model prediction to voc-agent.

[0264] S1602. If there are templates that need to take effect, voc-agent instructs voc-java-predict to take effect the templates.

[0265] S1603. The voc-agent instructs voc-java-predict to start a prediction task.

[0266] An example of the parameters of a prediction task is as follows:

[0267] Task name: Random generation.

[0268] Data type: The data type uploaded by the user.

[0269] Model list: The model list selected by the user.

[0270] Automated prediction: Off.

[0271] Upload time: The start time is set to the data upload time, and the end time is set to the data upload completion time.

[0272] S1604. The voc-agent polls the prediction task results and progress from voc-java-predict at regular intervals.

[0273] S1605. After the task is completed, voc-java-predict sends callback information to the voc-agent.

[0274] S1606. The voc-agent returns (return) to the front end one or more of the following: the prediction task ends, the current sub-session ends, and the evaluation content is pushed.

[0275] One or more steps before the above S1606 may also have corresponding return steps.

[0276] VII. Data Flywheel

[0277] The flywheel data collected may include one or more of the following:

[0278] · The execution plan output by the Planner.

[0279] · The response guidance output by the Planner.

[0280] · The output of the ModelSelectTool.

[0281] · The output of the PredictProcessTool.

[0282] The flywheel data of the data analysis agent can be recorded in ES. An ES index can be implemented in JSON, and an example of an index can include one or more of the following:

[0283] VIII. Database

[0284] 8.1 Add the agent_analysis_sub_session data table.

[0285] Table 1. agent_analysis_sub_session: Data analysis agent prediction session table

[0286]

[0287]

[0288] 8.2 For the State field status process, see Figure 17 :

[0289] For example, initially, the State field can be pending (1701). When executing the plan, the State field can be plan (1702). If the plan fails, modify the State field to failed (1703). If the plan is successful, modify the State field to action (1704) and call the tool to execute the task. If the tool execution fails, modify the State field to failed (1703). If the tool execution is successful, modify the State field to done (1705).

[0290] 8.3 Conversation information (session_message) table.

[0291] The content in the dialog box needs to record whether the user can perform operations. For example, it is represented by the action flag.

[0292] · true: Indicates that the user cannot perform operations. Such as clicking to create a prediction task and other operations.

[0293] · false: The user can perform operations.

[0294] 8.4 Add message types.

[0295] For the display of messages in the data analysis agent, according to the requirements document, a new card type can be added. Expand the card_type field in the session_message table.

[0296] The operation of parsing the card will be called in multiple places in the code, and the factory pattern design pattern can be used to implement it. An example is as follows:

[0297] · BaseCard is an abstract base class, and other card types inherit from the base class.

[0298] · The CardFactory is a factory class that provides the createCardParam method for converting a string into specific card parameters. The card parameters can include quey, label, review, template, predictSuccess, trainFailed, trainInform, questionnaire, modelDetail, etc.

[0299] IX. Interface Examples

[0300] 9.1 Query the mapping relationship between sessionId and agentType: It is necessary to add the content of the data analysis agent.

[0301] 9.2 Initialize the websocket session: It is necessary to add a welcome message for the data analysis agent.

[0302] 9.3 Create a prediction task

[0303] It can include information such as Uniform Resource Locator (URL), Method, request, response, and backend logic. Among them, the method can be POST.

[0304] The request and response can be implemented using the JSON method. An example of a request can include:

[0305] "sessionId": Session identifier;

[0306] "subSessionId": Sub-session identifier of the current data analysis agent;

[0307] "queryId": Message identifier

[0308] "templates": Template information.

[0309] Among them, the template information can include:

[0310] "id": Template identifier;

[0311] "modelType": Model type;

[0312] "modelCategory": Model category, large model or small model;

[0313] "name": Template name;

[0314] "desc": Template description;

[0315] "isApplied": A boolean value indicating whether it is applied. Yes: Applied; No: Not applied.

[0316] "modelPath": The model path.

[0317] "enableSentiment": A boolean value indicating whether sentiment analysis is enabled. Yes: Enabled.

[0318] "industryAndScenario": Industry and scenario information.

[0319] Industry and scenario information may include:

[0320] "industry": The industry.

[0321] "showIndustry": Chinese name of the industry.

[0322] "scenario": The scenario.

[0323] "showScenario": Chinese name of the scenario.

[0324] An example of a response includes: Returning via a general code message (codemessage).

[0325] An example of a possible backend logic includes:

[0326] (1) Verify the request parameters.

[0327] (2) Update the request parameters to session_message.

[0328] (3) For the content in the current templates field, if it has not been applied, call the voc-java-dataset service interface for template application (in the dimensions of agentId and userId).

[0329] a. If it fails.

[0330] b. Send a failure message via the ws protocol.

[0331] c. Terminate the current conversation, set sub_session_id in session_message to empty, insert a failure message into session_message (where cardType is textLink), and send the evaluation content.

[0332] d. The http returns the reason for the creation failure.

[0333] (4) Call the create prediction task interface of the voc-java-predict service.

[0334] a. If it fails.

[0335] b. The same as the failure handling logic for calling voc-java-dataset

[0336] (5) If successful, save a message to session_message.

[0337] (6) Send via the ws protocol "The model confirmation is successful. Now, data analysis and prediction will start, which will take some time. After the prediction is completed, I will send you a message to inform you. Please be patient. If you need to query the prediction progress, you can also ask me 'Where is the current prediction?'".

[0338] 9.4. Query historical session records: Need to handle for the newly added card types.

[0339] 9.5 Query agent unread message information

[0340] It can include information such as URL, method, request, and response. Among them, the method can be GET. The request can be empty.

[0341] The request and response can be implemented using the JSON method. An example of a possible response is as follows:

[0342] "data": data.

[0343] The data can include:

[0344] "agentType": agent type;

[0345] "sessionId": session identifier

[0346] "count": unread quantity, long integer;

[0347] "latestMessage": content of the latest message;

[0348] "createdAt": creation time of the latest message.

[0349] 9.6 Set the messages in the session as read

[0350] It can include information such as URL, method, request, and response. Among them, the method can be GET. The request can include the session identifier (sessionId), and the response can be empty.

[0351] 9.7 Data analysis agent uploads file data

[0352] It may include information such as URLs, methods, requests, and responses. For example, the method may be POST.

[0353] 9.8 Query Template List

[0354] It may include information such as URLs, methods, requests, and responses. Among them, the method may be GET. The request may include the data analysis agent sub-session ID.

[0355] The response may include a template list. The template list may include specific information about each template. For example, list, name, desc, modelpath, isApplied, enableSentiment, enableSentiment, enableSentiment, etc. of the template.

[0356] The list may include id, industryAndScenario, industry, showIndustry, scenario, showScenario, etc.

[0357] Figure 18 It is a schematic structural diagram of a model prediction device 1800 according to an embodiment of the present disclosure, including:

[0358] A first acquisition module 1801, configured to acquire model recommendation information according to demand information;

[0359] A second acquisition module 1802, configured to acquire a target model in response to a selection command for the model recommendation information;

[0360] An input module 1803, configured to input data to be processed into the target model for processing to obtain an output result.

[0361] Figure 19 It is a schematic structural diagram of a model prediction device 1900 according to another embodiment of the present disclosure. The device 1900 may include: a first acquisition module 1901, a second acquisition module 1902, and an input module 1903. The functions of the above modules may refer to the functions of the respective modules of the agent-based model prediction device 1800 in the above embodiment. In one implementation, the first acquisition module 1901 includes:

[0362] A planning sub-module 19011, configured to call the planner of the agent to plan the demand information to obtain a model selection tool to be called;

[0363] A selection sub-module 19012, configured to call the model selection tool to obtain recommended model information that meets the demand information.

[0364] In one embodiment, the selection sub-module 19012 obtains recommendation model information that meets the requirement information, including: sending a request to obtain a model list to the model data set; receiving a first model list returned by the model data set; assembling a prompt for the large model based on the first model list and the requirement information, where the prompt is used to ask the large model for model recommendation results; inputting the prompt into the large model to obtain a second model list of recommended models returned by the large model.

[0365] In one embodiment, the selection sub-module 19012 obtaining recommendation model information that meets the requirement information further includes: sending a model filtering request to the model filtering module, where the model filtering request is used to request filtering of the second model list; receiving a third model list recommended after filtering the second model list by the model filtering module.

[0366] In one embodiment, as Figure 19 shown, the input module 1903 includes:

[0367] A start sub-module 19031, configured to start a prediction task in response to a confirmation command for prediction task start information;

[0368] An input sub-module 19032, configured to input the data to be processed into the target model for prediction in response to the prediction task.

[0369] In one embodiment, as Figure 19 shown, the device further includes:

[0370] A planning module 1904, configured to call the planner of the intelligent agent to plan the progress query intention to obtain the progress query tool to be called;

[0371] A query module 1905, configured to call the progress query tool to query the prediction task progress information.

[0372] In one embodiment, the query module 1905 querying the prediction task progress information includes: sending a prediction task progress query request to the prediction service; receiving the prediction task progress information returned by the prediction service.

[0373] In one embodiment, as Figure 19 shown, the device further includes:

[0374] A third acquisition module 1906, configured to acquire first prompt information; the first prompt information is used to prompt to provide the data to be processed;

[0375] A fourth acquisition module 1907, configured to acquire the data to be processed provided according to the first prompt information in response to a data provision command.

[0376] In one embodiment, as Figure 19 shown, the apparatus further includes:

[0377] A fifth acquisition module 1908, configured to acquire second prompt information; the second prompt information is used to prompt the client to provide the requirement information;

[0378] A sixth acquisition module 1909, configured to acquire the requirement information provided according to the second prompt information in response to a requirement provision command.

[0379] In one embodiment, the apparatus further includes an execution module, and the execution module is configured to perform one of the following:

[0380] When there is a target agent record in the session of the agent and the task type in the sub-session of the target agent is model training, perform model training;

[0381] When there is a target agent record in the session of the agent and the task type in the sub-session of the target agent is model prediction, perform model prediction;

[0382] When there is no target agent record in the session of the agent and there is a target agent record in the agent analysis report, perform model training;

[0383] When there is no target agent record in the session of the agent and there is no target agent record in the agent analysis report, perform model prediction.

[0384] For the specific functions and examples of the modules and sub-modules of the apparatus according to the embodiments of the present disclosure, reference may be made to the relevant descriptions of the corresponding steps in the foregoing method embodiments, which will not be elaborated herein.

[0385] Figure 20 is a schematic structural diagram of an agent 2000 according to an embodiment of the present disclosure. The agent may include:

[0386] An input module 2001, configured to input requirement information and data to be processed;

[0387] A processing module 2002, configured to call a large model to perform model prediction processing according to any embodiment of the present disclosure based on the requirement information and the data to be processed received by the input module to obtain an output result;

[0388] An output module 2003, configured to output the output result obtained by the processing module.

[0389] For the specific functions and examples of the modules of the agent according to the embodiments of the present disclosure, reference may be made to the relevant descriptions of the corresponding steps in the foregoing method embodiments, which will not be elaborated herein.

[0390] In the technical solution of the present disclosure, the acquisition, storage, and application of the user's personal information involved all comply with the provisions of relevant laws and regulations and do not violate public order and good customs.

[0391] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0392] Figure 21 FIG. shows a schematic block diagram of an exemplary electronic device 2100 that can be used to implement the embodiments of the present disclosure. The electronic device is intended to represent various forms of digital computers, such as, for example, a laptop computer, a desktop computer, a workbench, a personal digital assistant, a server, a blade server, a mainframe computer, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as, for example, a personal digital assistant, a cellular phone, a smart phone, a wearable device, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely exemplary and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0393] As Figure 21 shown, the device 2100 includes a computing unit 2101 that can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 2102 or a computer program loaded from a storage unit 2108 into a random access memory (RAM) 2103. In the RAM 2103, various programs and data required for the operation of the device 2100 can also be stored. The computing unit 2101, the ROM 2102, and the RAM 2103 are connected to each other via a bus 2104. An input / output (I / O) interface 2105 is also connected to the bus 2104.

[0394] A plurality of components in the device 2100 are connected to the I / O interface 2105, including: an input unit 2106, such as a keyboard, a mouse, etc.; an output unit 2107, such as various types of displays, speakers, etc.; a storage unit 2108, such as a magnetic disk, an optical disk, etc.; and a communication unit 2109, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 2109 allows the device 2100 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0395] The computing unit 2101 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 2101 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 2101 executes the various methods and processes described above, such as the model prediction method. For example, in some embodiments, the model prediction method can be implemented as a computer software program tangibly embodied in a machine-readable medium, such as the storage unit 2108. In some embodiments, part or all of the computer program can be loaded and / or installed onto the device 2100 via the ROM 2102 and / or the communication unit 2109. When the computer program is loaded into the RAM 2103 and executed by the computing unit 2101, one or more steps of the model prediction method described above can be executed. Alternatively, in other embodiments, the computing unit 2101 can be configured to execute the model prediction method in any other suitable manner (e.g., by means of firmware).

[0396] Various embodiments of the systems and techniques described above in this document can be implemented in digital electronic circuitry, integrated circuit systems, field-programmable gate arrays (FPGA), application-specific integrated circuits (ASIC), application-specific standard products (ASSP), systems-on-a-chip (SOC), complex programmable logic devices (CPLD), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special-purpose or general-purpose programmable processor, receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting the data and instructions to the storage system, the at least one input device, and the at least one output device.

[0397] The program code for implementing the methods of the present disclosure can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when the program codes are executed by the processor or controller, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The program codes can be executed entirely on the machine, partially on the machine, as a stand-alone software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0398] In the context of this disclosure, a machine-readable medium can be a tangible medium that can contain or store a program for use by or in connection with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0399] In order to provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the computer. Other kinds of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0400] The systems and techniques described herein can be implemented in a computing system including backend components (e.g., as a data server), or a computing system including middleware components (e.g., an application server), or a computing system including frontend components (e.g., a user computer having a graphical user interface or a web browser through which the user can interact with an implementation of the systems and techniques described herein), or a computing system including any combination of such backend components, middleware components, or frontend components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), and the Internet.

[0401] A computer system can include a client and a server. The client and the server are generally remote from each other and typically interact through a communication network. The client-server relationship is generated by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, or a server of a distributed system, or a server incorporating a blockchain.

[0402] It should be understood that the various forms of processes shown above can be used, with steps reordered, added or deleted. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and no limitation is imposed herein.

[0403] The above specific embodiments do not constitute a limitation on the protection scope of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub - combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the principles of this disclosure shall be included within the protection scope of this disclosure.

Claims

1. A model prediction method, comprising: Obtain model recommendation information based on demand information; In response to a selection command for the model recommendation information, obtaining a target model; The data to be processed is input into the target model for processing to obtain an output result.

2. The method according to claim 1, wherein: According to the demand information, obtain the model recommendation information, including: Calling the planner of the intelligent agent to plan the demand information and obtain the model selection tool to be called; The model selection tool is called to obtain recommended model information that meets the requirement information.

3. The method according to claim 2, wherein: Obtain recommended model information that meets the requirements, including: Send a request to the model dataset to obtain the model list; Receive a first model list returned by the model data set; Assembling prompt words of a large model according to the first model list and the demand information, wherein the prompt words are used to inquire the large model about a model recommendation result; The prompt word is input into the large model to obtain a recommended second model list returned by the large model.

4. The method according to claim 3, wherein: Obtaining recommended model information that meets the requirement information also includes: Sending a model filtering request to a model filtering module, wherein the model filtering request is used to request filtering of the second model list; Receive a third model list recommended by the model filtering module after filtering the second model list.

5. The method according to any one of claims 1 to 4, wherein: Inputting the data to be processed into the target model for processing includes: In response to a confirmation command of the prediction task start information, start the prediction task; In response to the prediction task, the data to be processed is input into the target model for prediction.

6. The method according to any one of claims 1 to 5, further comprising: Call the agent's planner to plan the progress query intention and obtain the progress query tool that needs to be called; Call the progress query tool to query the prediction task progress information.

7. The method according to claim 6, wherein: Query the forecast task progress information, including: Send a prediction task progress query request to the prediction service; Receive the prediction task progress information returned by the prediction service.

8. The method according to any one of claims 1 to 7, further comprising: Get the first prompt information; The first prompt information is used to prompt to provide data that needs to be processed; In response to the data providing command, data to be processed provided according to the first prompt information is obtained.

9. The method according to any one of claims 1 to 8, further comprising: Get the second prompt information; The second prompt information is used to prompt to provide the required information; In response to a requirement providing command, the requirement information provided according to the second prompt information is acquired.

10. The method according to any one of claims 1 to 9, further comprising one of the following: If there is a target agent record in the agent's session and the task type in the target agent's subsession is model training, perform model training; If there is a target agent record in the agent's session and the task type in the target agent's subsession is model prediction, perform model prediction; Perform model training when there is no target agent record in the agent's session and there is a target agent record in the agent analysis report; Perform model prediction when there is no target agent record in the agent's session and no target agent record in the agent's analysis report.

11. A model prediction device, comprising: The first acquisition module is used to obtain model recommendation information according to demand information; A second acquisition module, configured to acquire a target model in response to a selection command for the model recommendation information; The input module is used to input the data to be processed into the target model for processing to obtain an output result.

12. An intelligent agent, comprising: Input module, used to input demand information and data to be processed; A processing module, configured to call a large model to execute a model prediction method as described in any one of claims 1 to 10 to obtain an output result based on the demand information and the data to be processed received by the input module; An output module is used to output the output result obtained by the processing module.

13. An electronic device, comprising: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 10.

14. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to cause the computer to execute the method according to any one of claims 1-10.

15. A computer program product comprising a computer program, which, when executed by a processor, implements the method according to any one of claims 1 to 10.

Images