Information interaction method and device based on large model, electronic equipment and intelligent agent

Abstract

The present disclosure provides a large model-based information interaction method and device, electronic equipment and agent, relates to the technical field of artificial intelligence, especially to the technical field of computer vision, deep learning, large model, etc., and can be applied to the scene of AIGC content generation based on artificial intelligence. The method comprises: calling a large model to process input information to generate at least one first task information to be executed, wherein the at least one first task information is used to represent a first intention of the input information; displaying the at least one first task information; in response to an editing operation on the at least one first task information, determining at least one second task information, wherein the at least one second task information is used to represent a second intention of the input information; calling the large model to generate output information matched with the second intention according to the at least one second task information.

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence technology, particularly to computer vision, deep learning, and large-scale models, and can be applied to scenarios such as AIGC-based content generation. Specifically, it relates to an information interaction method, device, electronic device, and intelligent agent based on a large-scale model. Background Technology

[0002] With the continuous development of artificial intelligence technology, large-scale modeling technology has been applied to various fields. For example, large-scale models are used to achieve data querying and content generation.

[0003] However, when a large model performs a task and outputs information, the accuracy of the output information is positively correlated with the extent to which the large model understands the input information. If the large model's understanding of the input information is flawed, the output information will not meet the user's needs, resulting in a poor user experience. Summary of the Invention

[0004] This disclosure provides an information interaction method, device, electronic device, and intelligent agent based on a large model.

[0005] According to one aspect of this disclosure, a method for information interaction based on a large model is provided, comprising: invoking a large model to process input information to generate at least one first task information to be executed, wherein the at least one first task information is used to characterize a first intent of the input information; displaying the at least one first task information; in response to an editing operation on the at least one first task information, determining at least one second task information, wherein the at least one second task information is used to characterize a second intent of the input information; and invoking a large model to generate output information matching the second intent based on the at least one second task information.

[0006] According to another aspect of this disclosure, a large-model-based information interaction device is provided, comprising: a first generation module, configured to invoke the large model to process input information and generate at least one first task information to be executed, wherein the at least one first task information is used to represent a first intent of the input information; a first display module, configured to display the at least one first task information; a determination module, configured to determine at least one second task information in response to an editing operation on the at least one first task information, wherein the at least one second task information is used to represent a second intent of the input information; and a second generation module, configured to invoke the large model to generate output information matching the second intent based on the at least one second task information.

[0007] According to another aspect of this disclosure, an intelligent agent is provided, comprising: an input module for receiving input information; a processing module for determining a target task based on the input information received by the input module, determining a large model based on the target task, and obtaining output information by calling the large model to execute the information interaction method based on the large model as described above; and an output module for outputting the output information obtained by the processing module.

[0008] According to another aspect of this disclosure, an electronic device is provided, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method described above.

[0009] According to another aspect of this disclosure, a non-transitory computer-readable storage medium is provided storing computer instructions, wherein the computer instructions are used to cause the computer to perform the method described above.

[0010] According to another aspect of this disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements the method described above.

[0011] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0012] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein:

[0013] Figure 1 This illustration schematically shows an exemplary system architecture applicable to information interaction methods and apparatus based on large models, according to embodiments of the present disclosure;

[0014] Figure 2 A flowchart illustrating an information interaction method based on a large model according to an embodiment of the present disclosure is shown schematically.

[0015] Figure 3 This schematic diagram illustrates a scenario of displaying at least one first task information according to an embodiment of the present disclosure;

[0016] Figure 4A This diagram illustrates an application scenario for determining second task information according to an embodiment of the present disclosure.

[0017] Figure 4B This diagram illustrates an application scenario for determining second task information according to another embodiment of the present disclosure.

[0018] Figure 5 This illustration schematically depicts a scenario where at least one first task information is generated and edited to obtain at least one second task information according to an embodiment of this disclosure.

[0019] Figure 6 A flowchart illustrating the determination of at least one field of information and parameter information for each field of information according to an embodiment of the present disclosure is shown.

[0020] Figure 7A The illustration shows a scenario in which an information interaction method based on a large model generates output results according to an embodiment of the present disclosure;

[0021] Figure 7B The illustration shows a scenario in which the output result is generated by the information interaction method based on a large model according to another embodiment of the present disclosure;

[0022] Figure 7C The illustration shows a scenario in which the output result is generated by the information interaction method based on a large model according to another embodiment of the present disclosure;

[0023] Figure 8 A block diagram of a large-model-based information interaction device according to an embodiment of the present disclosure is shown schematically.

[0024] Figure 9 A schematic diagram illustrating the structure of an intelligent agent of artificial intelligence according to embodiments of the present disclosure; and

[0025] Figure 10 A block diagram of an electronic device suitable for implementing a large-model-based information interaction method according to an embodiment of the present disclosure is shown schematically. Detailed Implementation

[0026] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0027] In the realm of large models, the accuracy of their output is directly correlated with their understanding of the input information. For example, when the intent of the input is unclear or ambiguous, the output generated by the large model may not match the user's actual needs. Conversely, even when the intent of the input is clear, the large model's understanding of the input may be flawed, leading to outputs that do not meet the user's expectations. Current methods for intent verification include intent rewriting based on Natural Language Processing (NLP) or function parameter filling based on FunctionCall. Generally, NLP intent rewriting is used for unstructured data, while function filling is used for structured data. However, these methods also struggle to demonstrate and modify the thought process of large models.

[0028] Therefore, this disclosure provides an information interaction method based on a large model, comprising: invoking the large model to process input information and generate at least one first task information to be executed, wherein the at least one first task information is used to represent a first intent of the input information; displaying the at least one first task information; in response to an editing operation on the at least one first task information, determining at least one second task information, wherein the at least one second task information is used to represent a second intent of the input information; and invoking the large model to generate output information matching the second intent based on the at least one second task information. Thus, by displaying at least one first task information, embodiments of this disclosure can demonstrate the large model's thought process, and by interactively modifying the thought process to obtain at least one second task information that conforms to the user's intent, the large model can generate output results that meet the user's needs based on at least one second task information, thereby improving the user experience.

[0029] Figure 1 The illustration schematically depicts an exemplary system architecture applicable to information interaction methods and apparatus based on large models, according to embodiments of the present disclosure.

[0030] It is important to note that Figure 1 The examples shown are merely examples of system architectures applicable to embodiments of this disclosure, intended to help those skilled in the art understand the technical content of this disclosure. However, they do not imply that embodiments of this disclosure cannot be used in other devices, systems, environments, or scenarios. For instance, in another embodiment, an exemplary system architecture applicable to the information interaction method and apparatus based on a large model may include a terminal device. However, the terminal device may implement the information interaction method and apparatus based on a large model provided by embodiments of this disclosure without needing to interact with a server.

[0031] like Figure 1As shown, the system architecture 100 according to this embodiment may include terminal devices 101, 102, and 103, a network 104, and a server 105. The network 104 serves as a medium for providing a communication link between the terminal devices 101, 102, and 103 and the server 105. The network 104 may include various connection types, such as wired and / or wireless communication links, etc.

[0032] Users can use terminal devices 101, 102, and 103 to interact with server 105 via network 104 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 101, 102, and 103, such as knowledge reading applications, web browser applications, search applications, instant messaging tools, email clients, and / or social platform software, etc. (for example only).

[0033] Terminal devices 101, 102, and 103 can be various electronic devices with displays and web browsing capabilities, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0034] Server 105 can be a server that provides various services, such as a backend management server that supports the content browsed by users using terminal devices 101, 102, and 103 (for example only). The backend management server can analyze and process data such as received user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal devices.

[0035] A server can be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system. It solves the shortcomings of traditional physical hosts and VPS services ("Virtual Private Server", or simply "VPS"), such as high management difficulty and weak business scalability. A server can also be a server for a distributed system or a server that incorporates blockchain technology.

[0036] It should be noted that the information interaction method based on a large model provided in this disclosure can generally be executed by terminal devices 101, 102, or 103. Accordingly, the information interaction device based on a large model provided in this disclosure can also be disposed in terminal devices 101, 102, or 103.

[0037] Alternatively, the large-model-based information interaction method provided in this embodiment can generally be executed by server 105. Correspondingly, the large-model-based information interaction device provided in this embodiment can generally be located in server 105. The large-model-based information interaction method provided in this embodiment can also be executed by a server or server cluster that is different from server 105 and capable of communicating with terminal devices 101, 102, 103 and / or server 105. Correspondingly, the large-model-based information interaction device provided in this embodiment can also be located in a server or server cluster that is different from server 105 and capable of communicating with terminal devices 101, 102, 103 and / or server 105. Alternatively, the large-model-based information interaction method provided in this embodiment can generally be executed by terminal devices 101, 102, or 103. Correspondingly, the large-model-based information interaction device provided in this embodiment can generally be located in terminal devices 101, 102, or 103.

[0038] For example, a user can edit input information through interaction with the interfaces of terminal devices 101, 102, and 103. Terminal devices 101, 102, and 103 send the acquired input information to server 105, which then calls a large model to process the input information and generate at least one first task information to be executed. This first task information represents the first intent of the input information. Terminal devices 101, 102, and 103 can receive and display this first task information from server 105. The user can still edit the at least one first task information through the interfaces of terminal devices 101, 102, and 103. In response to the editing of the at least one first task information, terminal devices 101, 102, and 103 determine at least one second task information, which represents the second intent of the input information. Terminal devices 101, 102, and 103 send the at least one second task information to server 105, which then calls a large model to generate output information matching the second intent based on the at least one second task information. Alternatively, the target content can be analyzed by a server or server cluster capable of communicating with terminal devices 101, 102, 103 and / or server 105, ultimately extracting content of interest to the user. Alternatively, a large model can be set up within terminal devices 101, 102, 103, allowing them to directly call the local large model to generate at least one first task information and output information.

[0039] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0040] In the embodiments disclosed herein, the collection, storage, use, processing, transmission, provision, disclosure, and application of user personal information comply with relevant laws and regulations, necessary confidentiality measures have been taken, and they do not violate public order and good morals.

[0041] In the technical solution disclosed herein, the user's authorization or consent is obtained before acquiring or collecting the user's personal information.

[0042] Figure 2 A flowchart illustrating an embodiment of the information interaction method based on a large model according to this disclosure is shown schematically. Figure 2 As shown, Embodiment 200 includes operations S210 to S240.

[0043] In operation S210, the large model is invoked to process the input information and generate at least one first task information to be executed, wherein the at least one first task information is used to characterize the first intent of the input information.

[0044] The input information can be at least one of the following: text information, image information. Accordingly, the large model can be a large language model (LMM) for processing text information, or a large multimodal model (LMM) for processing multimodal information.

[0045] The large model can be a pre-trained large model, which can process the input information and generate at least one first task information.

[0046] The first task information can be understood as the task information planned by the large model after understanding the input information, and it serves as the benchmark for the large model to execute subsequent tasks. At least one piece of first task information can represent the initial intent of the input information as a whole. It should be noted that one or more pieces of first task information can be one or more steps of the same task, or one or more steps of multiple tasks.

[0047] For example, the input information could be "Please help me calculate the number of tasks processed this year." After the large model processes the input information, it generates three primary tasks to be executed: "Query the number of tasks processed in the first half of the year," "Query the number of tasks processed in the second half of the year," and "Sum the number of tasks processed in the first half of the year and the number of tasks processed in the second half of the year." The first two primary tasks are query tasks, while the last one is a calculation task. Through these three primary tasks, the final result is the primary intent of the input information: "annual task processing volume."

[0048] During operation S220, at least one first task information is displayed.

[0049] The first task information can be displayed in various forms. For example, at least one piece of first task information can be displayed in natural language, such as "query the number of tasks processed in the first half of the year", "query the number of tasks processed in the second half of the year", or "sum the number of tasks processed in the first half of the year and the number of tasks processed in the second half of the year". Alternatively, it can be displayed in natural language combined with editable controls, such as "query the number of tasks processed from July to December of this year", where "July" and "December" are in the form of editable controls.

[0050] In operation S230, in response to an editing operation on at least one first task information, at least one second task information is determined, wherein the at least one second task information is used to characterize a second intent of the input information.

[0051] Editing operations can include at least one of the following: adding, deleting, and modifying. At least one editing operation can be performed on at least one first task piece of information to obtain at least one second task piece of information.

[0052] The number of second task information items can be the same as or different from the number of first task information items. For example, after adding, deleting, or modifying some information in a first task information item, the edited task information becomes the second task information, and the number of second task information items is the same as the number of first task information items. Alternatively, when the editing operation on at least one first task information item is to add first task information, the number of second task information items is greater than the number of first task information items. Or, when the editing operation on at least one first task information item is to delete a first task information item, the number of second task information items is less than the number of first task information items.

[0053] For example, consider the following three first task information entries A1~A3: "Query the number of tasks processed from January to June of this year", "Query the number of tasks processed from July to December of this year", and "Sum the number of tasks processed in the first half of the year and the number of tasks processed in the second half of the year". Without editing the first and third first task information entries, modify the "December" entry in the second first task information entry. This results in three second task information entries B1~B3: "Query the number of tasks processed from January to June of this year", "Query the number of tasks processed from July to November of this year", and "Sum the number of tasks processed in the first half of the year and the number of tasks processed in the second half of the year".

[0054] At least one second task information, as a whole, is used to characterize the second intent of the input information. The second intent may be the same as or different from the first intent of the input information characterized by at least one first task information.

[0055] For example, consider the input message "Please help me calculate the number of tasks processed this year," three primary task information entries A1-A3, and three secondary task information entries B1-B3. The primary intent represented by the three primary task information entries A1-A3 is "annual task processing volume," and the secondary intent represented by the three secondary task information entries B1-B3 is also "annual task processing volume." The primary and secondary intents are the same. Alternatively, if we delete the primary task information entries A1 and A3, and modify the primary task information entry A2 to "Query the number of tasks processed in July of this year," then the secondary task information entry C1 becomes "Query the number of tasks processed in July of this year." In this case, the secondary intent represented by the secondary task information entry C1 is "monthly task processing volume," and the primary and secondary intents are different.

[0056] It is understandable that if no editing operation is performed on at least one first task information, then at least one first task information can be regarded as at least one second task information for subsequent operations.

[0057] In operation S240, the large model is invoked to generate output information that matches the second intent based on at least one second task information.

[0058] Output information can include various forms such as text output, image output, and links. It is understood that the form of the output information corresponds to the input information; for example, the output information is generated in the form specified in the input information.

[0059] The large model can contain multiple model components, each used to perform tasks such as classification and recognition; it can also contain multiple plugins, each used for data processing, rendering, and other functions. In this embodiment, the large model can call upon its internal model components or plugins to generate output information based on at least one second task information, according to the input information.

[0060] When calling the large model, output information matching the second intent is directly generated based on at least one second task information, without having to reinterpret and replan the task information as new input information.

[0061] In the embodiments of this disclosure, by calling a large model to process the input information, at least one first task information to be executed is generated. This first task information represents the first intent of the input information. Displaying at least one first task information demonstrates the large model's thought process in understanding the input information, allowing the user to intuitively see this process. In response to editing operations on at least one first task information, at least one second task information is determined, enabling the user to modify the displayed thought process and obtain at least one second task information that matches the user's intent. Therefore, the output information generated by calling the large model based on at least one second task information is more in line with the user's actual needs, improving the user experience. Furthermore, since calling the large model directly generates output information matching the second intent based on at least one second task information, there is no need to reinterpret and plan the task information as new input information. Therefore, the embodiments of this disclosure avoid the intent deviation caused by the large model's reinterpretation due to the introduction of new information, further ensuring that the generated output information better meets the user's actual needs and provides a better user experience.

[0062] According to embodiments of this disclosure, the first task information includes at least one parameter information from the input information; displaying at least one first task information includes: displaying at least one parameter information for each first task information, wherein each parameter information is associated with a first control for editing the parameter information.

[0063] At least one parameter in the first task information can be parameter information of at least one indicator under the same dimension, such as parameter information of year, month, and day under the time dimension. Alternatively, at least one parameter in the first task information can also be parameter information of at least one indicator under at least one dimension. For example, the first task information includes parameter information of year under the time dimension, and also parameter information of data volume under the data dimension.

[0064] Parameter information can be extracted directly from the input information or determined based on the semantics of the input information. For example, for the input information "Please query the number of tasks processed in 20XX", the parameter information can be directly extracted from the input information: "20XX" and "number of tasks processed". The parameter information determined based on semantics can be a specific month of the year, such as "January", "June", "July", and "December". Thus, at least one first task information displayed could be "Query the number of tasks processed from January to June 20XX", "Query the number of tasks processed from July to December 20XX", or "Summarize the number of tasks processed in the first half of the year and the number of tasks processed in the second half of the year".

[0065] In one embodiment, when displaying at least one first task information, a first control associated with each parameter information may be displayed simultaneously. For example, the first control may be a dropdown list displayed near the parameter information.

[0066] In another embodiment, only at least one first task information may be displayed. By manipulating the parameter information within the first task information, a first control corresponding to that parameter information is activated to edit it. For example, double-clicking a parameter information SS1 can initiate editing of that parameter information SS1; or, right-clicking a parameter information SS2 displays candidate parameter information for that parameter information SS2, and selecting a candidate parameter information to replace the parameter information SS2 allows for editing of that parameter information SS2.

[0067] In the embodiments of this disclosure, for each first task information, at least one parameter information determined from the input information is displayed in an editable form. This not only helps users to intuitively view and understand the thought process of the large model, but also makes it convenient for users to modify the parameter information, making the operation simple and convenient.

[0068] Figure 3 The illustration shows a scenario diagram of displaying at least one first task information according to an embodiment of the present disclosure.

[0069] like Figure 3 As shown, Example 300 can be a large model analyzing the input information "Please query this year's annual expenditure" and generating three first task information, corresponding to steps 1, 2, and 3 respectively. Step 1 includes five parameter information for three indicators under two dimensions. The two dimensions are the time dimension and the data dimension. The three indicators are year and month under the time dimension and indicator B under the data dimension. The five parameter information are "2024", "4", "202XX", "X", and "indicator B".

[0070] Taking parameter information 301 as an example, such as "2024", the first control associated with parameter information 301 is input box 302. Users can start the editing function and begin editing new parameter information by clicking on the area of ​​input box 302 or parameter information 301, such as entering a new year "20XX".

[0071] According to embodiments of this disclosure, in response to an editing operation on at least one first task information, determining at least one second task information includes: for each first task information, in response to an operation on a first control associated with target parameter information in at least one parameter information, displaying at least one candidate parameter information; and in response to a selection operation on at least one candidate parameter information, replacing the target parameter information with the selected candidate parameter information to obtain the edited second task information.

[0072] Each parameter information may include at least one candidate parameter information. The parameter information and at least one candidate parameter information can be understood as optional parameter information under the same indicator.

[0073] The target parameter information can be understood as the parameter information that is manipulated in at least one of the parameter information in the first task information, or the parameter information to be edited. In this embodiment, at least one candidate parameter information of the target parameter information can be displayed by operating on the first control associated with the target operation information.

[0074] For example, the first control could be a dropdown list. By operating the dropdown list associated with the target parameter information, at least one candidate parameter information could be displayed within the dropdown list's display area. For instance, for the target parameter information "20XX", the candidate parameter information could be a 40-year period that fluctuates around "20XX" by 20 years.

[0075] For the at least one candidate parameter information displayed, the user can select one candidate parameter information as the required parameter information, and replace the target parameter information with the selected candidate parameter information to modify the target parameter information, thereby obtaining the edited second task information.

[0076] In the embodiments of this disclosure, at least one candidate parameter is displayed by selecting a first control associated with the target parameter information. In response to the selection of the at least one candidate parameter, the selected candidate parameter replaces the target parameter, resulting in modified second task information. Since the user can edit the target parameter information by selecting the automatically displayed candidate parameter, the editing of the first task information is simple and convenient, providing a good user experience. Furthermore, the user determines the second task information through the displayed candidate parameter information, meaning the entire generation process of the output information is achieved by artificial intelligence, eliminating the need for the user to think about and construct new parameter information, further enhancing the user experience.

[0077] Figure 4A The diagram illustrates an application scenario for determining second task information according to an embodiment of the present disclosure.

[0078] like Figure 4A As shown, Example 400A can be three first task information generated by the large model after analyzing the input information "Please query this year's annual expenditure", corresponding to steps 1, 2 and 3 respectively.

[0079] Step 1 includes five parameters for three indicators across two dimensions: time and data. The three indicators are year and month in the time dimension and indicator B in the data dimension. The five parameters are “2024”, “4”, “202XX”, “X”, and “Indicator B”.

[0080] The first control associated with the parameter information "2024", "4", "202XX", and "X" is a dropdown list. Taking parameter information 401, such as "4", as an example, the first control 402 is the first control 402 marked with a dropdown list. In addition, the first control 403 associated with the parameter information "Indicator B" can be an input box.

[0081] When the target parameter information is parameter information 401, after the user clicks the first control 402, at least one candidate parameter information 404 is displayed, such as "5, 6, 7". It should be noted that the at least one candidate parameter information displayed by the first control 402 may also include other candidate parameter information. The display area of ​​the first control 402 can be scrolled to display other candidate parameter information, such as scrolling down to display "8, 9, 10...", and scrolling up to display "4, 3, 2...".

[0082] According to embodiments of this disclosure, the first task information includes multiple parameter information that have dependencies. In response to an editing operation on at least one first task information, determining at least one second task information includes: for each first task information, in response to an operation on a first control associated with a target parameter information among the at least one parameter information, displaying at least one candidate parameter information; in response to a selection operation on at least one candidate parameter information, replacing the target parameter information with the selected candidate parameter information, and deleting at least one parameter information that has a dependency on the target parameter information; and in response to an editing operation on at least one parameter information that has a dependency on the target parameter information, obtaining the second task information.

[0083] Dependency can be a relationship between parameters, where the specific information of a parameter is affected by one or more other parameters with which it depends. For example, in a chained dependency relationship, later parameters are affected by earlier ones. Taking the parameters "City XX", "District XY", and "Street Y" as examples, "District XY" is dependent on "City XX" and changes as "City XX" changes; similarly, "Street Y" is dependent on "District XY" and changes as "District XY" changes.

[0084] Regarding the target parameter information, after replacing the target parameter information with the selected candidate parameter information, the target parameter information will change, and at least one parameter information that depends on the target parameter information will also change accordingly. Therefore, after obtaining the new target parameter information, at least one parameter information that depends on the target parameter information can be deleted to avoid forgetting to modify at least one dependent parameter information. Correspondingly, the user can input new parameter information through editing operations. The specific implementation of displaying at least one candidate parameter information in response to operations on the first control associated with the target parameter information among the at least one parameter information, and replacing the target parameter information with the selected candidate parameter information in response to selection operations on the at least one candidate parameter information, is described above and will not be repeated here.

[0085] In another embodiment, instead of deleting at least one parameter that is dependent on the target parameter information, the editing operation on at least one parameter that is dependent on the target parameter information can be omitted, and the information obtained after directly replacing the target parameter information and deleting at least one parameter that is dependent on the target parameter information can be used as the second task information.

[0086] In another embodiment, in response to an operation on a first control associated with at least one parameter information that is dependent on the target parameter information, at least one candidate operation information of the aforementioned parameter information is displayed, and a new candidate operation information is selected by a selection operation to replace the parameter information, thereby updating at least one parameter information that is dependent on the target parameter information to obtain the second task information.

[0087] For example, using the parameters "XX City", "XY District", and "Y Street" as examples, the target parameter could be "XX City", while "XY District" and "Y Street" are parameters that depend on the target parameter. After selecting "BB City" from at least one candidate parameter, "BB City" replaces "XX City", and "XY District" and "Y Street" are deleted. The user can then edit the input to enter "BY District" and "X Street" to obtain the second task information. Alternatively, after replacing "XX City" with "BB City" and deleting "XY District" and "Y Street", the user can click the first control associated with "XY District" to select "BY District" from the displayed candidate parameters, and click the first control associated with "X Street" to select "Y Street" from the displayed candidate parameters to obtain the second task information.

[0088] In the embodiments of this disclosure, for multiple parameter information that have dependencies, while replacing the target parameter information with the selected candidate parameter information, at least one parameter information that has a dependency on the target parameter information is deleted; and in response to the editing operation of at least one parameter information that has a dependency on the target parameter information, second task information is obtained, so that the user can intuitively understand the dependencies between multiple parameter information in the first task information and promptly modify at least one parameter information that has a dependency on the target parameter information, resulting in a better user experience, while avoiding deviations in output results due to forgetting to modify at least one parameter information that has a dependency.

[0089] According to embodiments of this disclosure, the method further includes: determining the field information to which the target parameter information belongs; and determining at least one parameter information other than the target parameter information in the field information as at least one candidate parameter information.

[0090] Field information is used to characterize a field. Field information may include the field name, abbreviation, short name, field description, etc.

[0091] Target parameter information can be understood as the specific parameter values ​​of the field information to which it belongs. Each field information includes parameter information with a specific meaning or form. Therefore, the field information to which the target parameter information belongs can be determined based on the form or meaning of the target parameter information.

[0092] Based on the stored data of the field information, at least one parameter information other than the target parameter information can be determined under the field information, and the above-mentioned at least one parameter information other than the target parameter information is determined as at least one candidate parameter information, so that the user can select a candidate parameter information to replace the target parameter information.

[0093] For example, stored data of field information can be retrieved from a database to determine the stored data of the field information. In one embodiment, the field information can be a field name. The first control associated with the target parameter information can be a control associated with the field name. Thus, when the user clicks the first control associated with the target parameter information, at least one parameter information other than the target parameter information can be determined and displayed based on the stored information of the current field name in the database.

[0094] In the embodiments of this disclosure, by operating the first control associated with the target parameter information, at least one parameter information other than the target parameter information in the field information is determined as at least one candidate parameter information and displayed according to the field information to which the target parameter information belongs, so that the user can replace the target parameter information with the candidate parameter information displayed by the actual stored data, and further ensure that the generated output results are more in line with the user's needs.

[0095] According to embodiments of this disclosure, each first task information is associated with a second control and / or a third control. The second control is used to add and / or delete at least one parameter information in the first task information; the third control is used to delete the first task information and / or add new first task information after the first task information; in response to an editing operation on at least one first task information, determining at least one second task information includes: in response to an operation on the second control associated with target task information in at least one first task information, adding / deleting at least one parameter information in the target task information to obtain at least one second task information; and / or in response to an operation on the third control associated with the target task information, deleting the target task information, and / or adding new first task information after the first task information to obtain at least one second task information.

[0096] Similar to parameter information, each first task information can be associated with at least one second control and / or at least one third control. For example, if first task information 1 includes 3 parameter information items, then 3 second controls can be associated to add and / or delete parameter information. For instance, clicking a second control adds new parameter information to the current position of the second control, or double-clicking a second control deletes the parameter information preceding the current second control. Alternatively, the second controls can include a first add sub-control and a first delete sub-control, used for adding and deleting parameter information respectively. For example, clicking the first add sub-control adds new parameter information to the current position of the second control; clicking the first delete sub-control deletes the parameter information preceding the current second control.

[0097] For example, the first task information 2 may include a third control. Clicking the third control once adds a new first task information after the current first task information 2, or double-clicking the third control deletes the first task information 2 associated with the current third control. Alternatively, the third control may include a second add sub-control and a second delete sub-control, used to add a new first task information after the first task information and delete the first task information, respectively. For example, clicking the second add sub-control adds a new first task information after the first task information; clicking the second delete sub-control deletes the first task information 2 associated with the current third control.

[0098] When using large models, various situations may arise that affect the quality of the output information, such as: the input information may not include parameter information that expresses the user's intent; upon viewing the first task information, it is discovered that parameter information needs to be added; the task information is found to include illusory output; redundant parameter information in the input information leads to redundancy in the first task information; and the first task information considered by the large model is too complex.

[0099] Therefore, embodiments of this disclosure provide a second control that can add and / or delete parameter information, and a third control that can add and / or delete first task information. This allows users to add parameter information not present in the input information, add new first task information not involved in the large model's thinking process, or delete redundant parameter information or first task information when viewing at least one first task information. This results in corrected second task information, ensuring the accuracy of the output information generated by the large model based on at least one second task information, and better meeting user needs.

[0100] Figure 4B The diagram illustrates an application scenario for determining second task information according to another embodiment of this disclosure. Figure 4B As shown, still in Figure 4A For example, at least one first task information.

[0101] A second control is set after the parameter "4" in the first task information, and a third control is set after the first task information. For example... Figure 4B As shown, the second control includes a first add sub-control 4051 and a first delete sub-control 4052, and the third control includes a second add sub-control 4061 and a second delete sub-control 4062.

[0102] Taking step 1 as an example, after clicking the first add sub-control 4051, new parameter information, such as parameter "1" for the "day" field, can be added after the month "4". Adaptively, new parameter information, such as parameter "31" for the day, can be added by clicking the first add sub-control (not shown in the figure) after the month "X". After clicking the first delete sub-control 4052, parameter information "4" can be deleted, and the description "month" associated with this parameter information can be deleted accordingly. Clicking the second add sub-control 4061 allows new first task information to be added after step 1, such as "Query indicator A from April 2024 to month X of 20XX". Alternatively, clicking the second delete sub-control 4062 deletes step 1.

[0103] In the scenario of adding the first task information, click the second add sub-control 4061 after step 1 and enter "Query indicator A from April 2024 to month X of 20XX" to complete the add operation. After the add operation is completed, step 2 will be automatically displayed before "Query indicator A from April 2024 to month X of 20XX". Figure 4B The original steps 2 and 3 are changed to steps 3 and 4, and the original step 3 is changed from "adding the data of step 2 to the data of step 1" to "adding the data of step 3 to the data of step 1".

[0104] Furthermore, for the modified "Step 4: Add the data from Step 3 to the data from Step 1", the user can also manipulate its associated second add sub-control to add new task information, such as Step 5 "Subtract the data from Step 1". Thus, the second intent represented by the five second task information includes the intents corresponding to "add" and "subtract", respectively. Compared to the "add" intent represented by the first task information generated based on the input information, the modification here adds a new intent "subtract".

[0105] According to embodiments of this disclosure, invoking a large model to process input information and generate at least one first task information to be executed includes: invoking the large model to perform intent analysis on the input information to obtain a first intent; generating third task information in machine language form based on the first intent and the input information; and converting the third task information into at least one first task information in natural language form.

[0106] Machine language can be understood as a domain-specific language (DSL) that supports complex human semantics and structured queries. In this embodiment, task information in machine language, such as third task information, can express multi-step computational operations within the input information, such as arithmetic operations.

[0107] Larger models can use existing intent recognition models to analyze the input information and obtain the primary intent. For example, the primary intent can be determined based on the input information through a classification task, which will not be elaborated on here.

[0108] Generating third task information in machine language form based on first intent information and input information can be achieved by using a slot-filling method, where information for each slot under the first intent is obtained from the input information, and then the third task information in machine language form is generated.

[0109] Understandably, based on the third task information in machine language form, the computer can directly execute the task. For example, for third task information used for querying, the computer can directly query the database based on the parameters in the third task information.

[0110] In this embodiment, the third task information can be converted into at least one first task information in natural language form through language conversion, so as to display the first task information that humans can understand. Natural language form can also be understood as human language form. It should be noted that machine language form can convert complex arithmetic formulas into natural language form, and can also convert arithmetic operations described in natural language form into task information in machine language form.

[0111] Both the third task information in machine language and the first task information in natural language include descriptive information and parameter information. The descriptive information is expressed differently in the two language forms, while the parameter information can be expressed the same or different in the two language forms.

[0112] Machine language, as an intermediate language for intent confirmation, directly determines the problem capabilities that large models can support, based on its design level and the complexity it can support. If the capabilities supported by machine language are limited, the fluency, readability, and editability of the subsequent translation into human language during the intent confirmation stage will be restricted, thus affecting the accuracy of intent confirmation.

[0113] In the embodiments of this disclosure, a novel machine language form is used to visualize the thought process by which a large model understands input information. Furthermore, in this novel machine language form, the third task information can include complex multi-step calculations, allowing for the extraction of one or more first task information pieces when converting the third task information into natural language form. Therefore, the embodiments of this disclosure can support various complex computational operations, particularly complex computational operations edited by the user based on the first task information, thereby expanding the applicability of the output information and improving the user experience.

[0114] Taking a specific scenario as an example, this explains the process of generating and displaying the first task information, and editing the first task information to obtain the second task information. For a user-defined new indicator M, indicator M is obtained by adding known indicators B and C. In this case, the input information could be "Please query indicator M for Q2 of 20XX, which is calculated from indicators B and C." Calling the large model for intent recognition yields a first intent that is multi-query, such as querying indicators B and C. After generating the third task information in machine language form based on the first intent and input information, the third task information, which includes multiple operations (querying indicators B and C), can be converted into three first task information steps, as shown in steps 1-3:

[0115] Step 1: Query the data for indicator B from April to June 20XX;

[0116] Step 2: Query the data for indicator C from April to June 20XX;

[0117] Step 3: Calculate the data for index M using the data from Step 2 and Step 1.

[0118] In this scenario, the large model cannot understand how to calculate indicator M based on indicators B and C, given the entirely new indicator M. After displaying the three initial task information entries, the second task information can be obtained by editing the text: "Step 3: Calculate the data for indicator M using the data from step 2 and step 1" to "Step 3: Add the data from step 2 and step 1 to obtain the data for indicator M." Thus, the large model can generate output information based on this second task information, and this output information matches the user's actual need to calculate indicator M.

[0119] Existing intent confirmation schemes cannot define and directly calculate entirely new metrics during the thought process. They can only refine the new metrics as much as possible from the input information and determine whether the output is correct through multiple rounds of input and output. If incorrect, the input information needs to be continuously optimized based on the previous round of output until the calculation is correct, resulting in a poor user experience. In contrast, the method disclosed in this paper, which demonstrates the thought process and modifies intent through information interaction, can confirm the intent before task execution (such as determining the calculation method of the new metric M), allowing a single calculation using a large model to obtain an output that meets the user's needs.

[0120] In one embodiment, the intent in the data question-answering domain can be broadly categorized as: detailed query, year-on-year / month-on-month comparison, pairwise entity query comparison, and single entity multi-indicator comparison. Specifically, the broad category of intent can further include subcategories of intent. For example, detailed query can include single query, multiple query, query + sorting, and pairwise comparison; year-on-year / month-on-month comparison can include year-on-year change, month-on-month change, and month-on-month change; pairwise entity query comparison can include pairwise query + comparison; and single entity multi-indicator comparison can include column comparison. In the data question-answering domain, the first intent can be one of the above intents.

[0121] Figure 5 The illustration shows a scenario of generating at least one first task information and editing it to obtain at least one second task information according to an embodiment of the present disclosure.

[0122] like Figure 5As shown, the input information 501 of the interactive interface can be "Please query the year-on-year and month-on-month comparisons of Indicator 1 in Q2 of 20XX". After calling the large model to perform intent analysis on the input information 501, a first intent 502 is obtained. Based on the first intent 502 and the input information 501, a third task information 503 in machine language form is generated. After language conversion of the third task information 503, at least one first task information in natural language form is obtained, such as first task information 1 504-1, first task information 2 504-2, etc. Then, at least one first task information is displayed on the interactive interface, such as at least one first task information 5051. Users can directly edit the displayed at least one first task information 5051 on the interactive page, such as changing "Indicator A" in step 1 to "Indicator C", and changing "Indicator B" in step 2 to "Indicator B'", to obtain at least one second task information 5052.

[0123] According to a specific embodiment of this disclosure, generating third task information in machine language form based on a first intent and input information includes: determining at least one field information and parameter information for each field information based on the input information; and generating third task information in machine language form based on the first intent and the parameter information for each field information.

[0124] The input information may include at least one parameter. Therefore, the field information to which each parameter belongs can be determined based on the parameter information in the input information, thereby obtaining the parameter information for each field. When generating the third task information in machine language form, the parameter information for each field is filled into the corresponding slot according to the correspondence between the slots and field information under the first intent, so as to generate the third task information.

[0125] In one embodiment, each first intent can correspond to a task template in machine language form, with the aforementioned slots reserved in the task template. Thus, the parameter information of each field can be filled into the corresponding slot to directly obtain the third task information. In another embodiment, a task template based on natural language can be used to fill the parameter information of each field into the corresponding slot, obtaining intermediate task information in natural language form; then, the intermediate task information in natural language form is converted into third task information in machine language form.

[0126] In the embodiments of this disclosure, at least one field of information and parameter information for each field of information are determined using input information. Third task information in machine language form is then generated based on the first intent and the parameter information for each field of information, ensuring that the parameter information in the third task information is in a field form that the machine can understand and manipulate. Therefore, in scenarios supporting complex computational operations, complex calculations can be performed quickly based on specific field information, improving the user experience.

[0127] Figure 6 A flowchart illustrating the determination of at least one field of information and parameter information for each field of information according to an embodiment of this disclosure is shown. Figure 6 As shown, Embodiment 600 includes operations S610 to S630. Embodiment 600 can be considered a specific embodiment for generating third task information in machine language form based on the first intent and parameter information of each field.

[0128] In operation S610, determine the field description information that is similar to each parameter information in the input information.

[0129] Field description information describes the meaning and / or form of the parameters included in the field information. This description information is in natural language. For example, for the field "name", the description could be "name", "team name", "task name", "task name in the XX series", etc.; for the field "table", the description could be "timetable", "detail table", etc.

[0130] The similarity between the existing field descriptions of each parameter in the input information can be calculated, and the field descriptions with similarity higher than a certain threshold can be regarded as field descriptions similar to the parameter information.

[0131] In operation S620, the field information corresponding to each field description information is determined based on the attribute information of the first dimension field.

[0132] The first dimension of field attribute information includes field information and field description information, as well as the correspondence between the two. Based on the correspondence between the two, the field information corresponding to each field description information can be directly determined.

[0133] For example, the relationship between field information and field description information can be one-to-one or one-to-many. For a one-to-many relationship, by assigning multiple natural language field descriptions to the field information, the field information corresponding to the parameter information can be more accurately determined, that is, the parameter information of each field information.

[0134] In one embodiment, the first-dimensional field attribute information can be stored in the form of a low-dimensional data set, containing low-dimensional field information, low-dimensional field description information, and the correspondence between them. For example, it can be a 1-dimensional data set.

[0135] When operating S630, the display format of each field information is determined based on the attribute information of the second dimension field.

[0136] The second-dimensional field attribute information includes the high-dimensional attributes of each field, which can be used to determine the display format of the field information. For example, the second-dimensional field attribute information may include multiple pre-determined high-dimensional attributes, such as data format, number of editable data, correlation information between high-dimensional attributes, similar field information, sorting, display format, etc.

[0137] For example, data format is used to limit field information to "structured data or unstructured data", editable quantity is used to limit field information to "whether multiple selections are allowed", correlation information between high-dimensional attributes is used to limit field information to "whether it is correlated with other high-dimensional attributes", similarity field information is used to limit field information to "related synonym arrays", sorting is used to limit field information to "whether it needs to be sorted by default within the entity", and display format is used to limit field information to "whether it is forced to be output".

[0138] For example, taking the field information "month" as an example, its high-dimensional attributes can be structured data, not selectable by multiple selections, correlated with "year", related synonyms include "quarter" and "Q", default sorting within the entity, and non-forced output.

[0139] Alternatively, the input information could be "Please query the 20XX annual summary of XX department." The parameter "annual summary" in the input information is actually a complex semantic parameter, similar to multiple field descriptions, such as annual task volume, completed task volume, legacy task volume, and long-term task volume, etc., and these multiple field descriptions correspond to multiple field information. Therefore, based on the first-dimensional field attribute information, multiple low-dimensional field information can be determined, and the display format of each field information can be determined based on the second-dimensional field attribute information. For example, field information 1 and field information 2 corresponding to annual task volume and completed task volume can be linked, such as annual task volume - number of completed tasks = number of legacy tasks.

[0140] In the embodiments of this disclosure, by abstracting the real physical world, different attributes of different entities in the physical world are separated and modeled, thus establishing a correspondence between field description information in natural language form and field information in machine language form. Furthermore, different dimensions of separation modeling and storage are performed for the same field information to obtain first-dimensional field attribute information and second-dimensional field attribute information. Therefore, using the first-dimensional and second-dimensional field attribute information, each field information and its display form in the low-dimensional complex semantics of the input information can be determined, thereby quickly generating third-task information in machine language form. Since each field information and its display form have been determined, the process of generating third-task information has flexible generalization while ensuring data security, and requires no other complex sample training; sample annotation is simple and training costs are low.

[0141] According to embodiments of this disclosure, converting third task information into at least one first task information in natural language form includes: converting third task information in machine language form into fourth task information in natural language form; and splitting the fourth task information into at least one first task information according to the complexity of the fourth task information.

[0142] The third task information, in machine language form, can be converted into the fourth task information, in natural language form, by using a language conversion tool. Such a language conversion tool could be a translation tool or a pre-trained translation model.

[0143] Complexity can be represented by the number of parameter information or the number of operations to be executed in the fourth task information. The more parameter information or the more operations to be executed in the fourth task information, the higher the complexity, and the more first task information is split; conversely, the fewer the parameter information or the more operations to be executed in the fourth task information, the fewer the first task information is split.

[0144] For example, the input information could be "In Q4 of 20XX, which departments had a higher processing volume than department XX?" The fourth task information would include querying the processing volume of department XX, querying the processing volume of other departments, and comparing the processing volume of department XX with the processing volume of other departments. The complexity can be considered as 3, and it can be broken down into 3 first task information.

[0145] In the embodiments of this disclosure, by converting the third task information in machine language form into the fourth task information in natural language form, and by splitting the fourth task information into at least one first task information according to the complexity of the fourth task information, complex input information can be broken down into simpler and clearer first task information, so as to facilitate understanding and modification, resulting in a better user experience.

[0146] According to an embodiment of this disclosure, for operation S240, calling the large model to generate output information matching the second intent based on at least one second task information includes: converting at least one second task information into at least one fifth task information in machine language form; converting at least one fifth task information into at least one task vector information in vector form; and calling the large model to generate output information based on at least one task vector information.

[0147] For at least one edited second task information, before calling the large model for processing, the at least one second task information is converted back to machine language form to obtain at least one fifth task information. At this point, the second and fifth task information can have a one-to-one or many-to-one relationship. For example, in DSL language, a single fifth task information can encompass both query and computation tasks, and the second and fifth task information can have a many-to-one relationship.

[0148] At least one fifth task information can be converted into at least one task vector information in vector form using an encoder-decoder. Then, the larger model can be invoked to generate output information based on this task vector information. For example, the task vector information could be a token sequence of the fifth task information; the larger model can then directly generate output information based on this token sequence.

[0149] In the embodiments of this disclosure, the second task information in natural language form is converted into the fifth task information in machine language form. Then, the fifth task information in machine language form, which can support complex calculations, is converted into vector form and the large model is called to generate the output result. This enables the large model to better understand and execute the task from the perspective of machine language, and generate output information that better meets the user's needs.

[0150] Figure 7A The illustration shows a scenario in which an information interaction method based on a large model generates output results according to an embodiment of the present disclosure.

[0151] like Figure 7A As shown, after the large model performs intent analysis on the input information 701 in natural language form, the third task information 702 in machine language form can be obtained. After converting the third task information 702 in machine language form into fourth task information 703 in natural language form, the fourth task information 703 is further split to obtain at least one first task information 704. The at least one first task information 704 in natural language form can be directly displayed to the user, and the user can also edit it to obtain at least one second task information 705. This process is also the intent correction and confirmation stage. Afterwards, the at least one second task information 705 can be converted into at least one fifth task information 706 in machine language form, and then the at least one fifth task information 706 can be converted into at least one task vector information 707 in vector form. The large model then generates and displays the output result 708 in natural language form based on the at least one task vector information 707.

[0152] In another embodiment, the method further includes: converting at least one fifth task information into intent description information in natural language form; displaying the intent description information; and in response to an intent confirmation operation on the intent description information, invoking a large model to generate output information matching a second intent based on at least one fifth task information.

[0153] The intent description information is used to describe the second intent. For the edited second task information, to ensure accurate understanding and execution of subsequent tasks, at least one second task information is converted into at least one fifth task information in machine language form. Furthermore, to ensure the accuracy of the second intent, at least one fifth task information is converted into intent description information in natural language form, and the intent description information is displayed for user confirmation.

[0154] For example, intent description information can be displayed on the interactive interface for inputting information. The user's confirmation action on the intent description information can be a click, a double click, a reply in natural language, or an action on an intent confirmation control, such as a "confirmation control".

[0155] The user's intent confirmation operation on the intent description information indicates that the second intent is an intent that meets the user's needs. Therefore, the large model can be invoked to generate output information that matches the second intent based on at least one fifth task information.

[0156] In the embodiments of this disclosure, before invoking the large model, the second intent represented by at least one second task information is displayed to the user through intent description information, and the large model is invoked to generate output results upon user confirmation, which can further ensure the degree of conformity between the generated output results and the user's actual needs.

[0157] Figure 7B The illustration shows a scenario in which an information interaction method based on a large model generates output results according to another embodiment of the present disclosure.

[0158] like Figure 7B As shown, after converting at least one second task information 705 into at least one fifth task information 706 in machine language form, the at least one fifth task information 706 in machine language form is organized into a complete intent description information 709, and this intent description information 709 in natural language form is displayed to the user for confirmation. After the user confirms the intent description information 709, the at least one fifth task information 706 in machine language form can be converted into task vector information 707, and the large model is called to generate an output result 708 based on the task vector information 707. The meanings of other parameter information are the same as... Figure 7A Similarities or identical items will not be elaborated upon here.

[0159] According to embodiments of this disclosure, in response to an intent editing operation on intent description information, at least one second task information is displayed; in response to an editing operation on at least one second task information, at least one sixth task information is determined, wherein the at least one sixth task information is used to characterize a third intent of the input information; and a large model is invoked to generate output information matching the third intent based on the at least one sixth task information.

[0160] After displaying the intent description information, if the intent description information still differs from the user's actual needs, the user can use intent editing controls, such as the "back control," to return to at least one of the previously edited second task information so that the second task information can be edited again.

[0161] In response to the editing operation of at least one second task information, at least one sixth task information is determined. Similar to the operation S230 above, the large model is invoked, and output information matching the third intent is generated based on at least one sixth task information. Similar to the operation S240 above, it will not be described again here.

[0162] In this embodiment, the third intent may be the same as or different from the first intent, or not completely the same as the second intent.

[0163] In the embodiments of this disclosure, before calling the large model to generate output information, two or more editing operations can be supported to obtain sixth task information, so that the large model can be called to generate output information based on at least one sixth task information. The embodiments of this disclosure support multiple editing of task information until the sixth task information that meets the user's actual needs is obtained. This eliminates the need to call the large model to reinterpret new input information and generate output results if the output does not meet the user's needs, resulting in high generation efficiency and a good user experience.

[0164] Figure 7C The illustration shows a scenario where the output result is generated according to another embodiment of the information interaction method based on a large model in accordance with the present disclosure. Still using... Figure 7B For example, in the embodiments, such as Figure 7C As shown, the user edits to obtain at least one second task information 705 in natural language form. The at least one second task information 705 is converted into at least one fifth task information 706 in machine language form, and the at least one fifth task information 706 is converted into intent description information 709 for user confirmation. The user can perform intent editing to re-display the at least one second task information 705. Further, the at least one second task information 705 is edited to obtain at least one sixth task information 710. The at least one sixth task information 710 in natural language form is converted into at least one seventh task information 711 in machine language form, and then the at least one seventh task information 711 is converted into task vector information 707. The large model is then called to generate output result 708 based on the task vector information 707. The meanings of other parameter information are... Figure 7B Similarities or identical items will not be elaborated upon here.

[0165] Figure 8A block diagram of a large-model-based information interaction device according to an embodiment of the present disclosure is shown schematically.

[0166] like Figure 8 As shown, the information interaction device 800 based on a large model includes a first generation module 810, a first display module 820, a determination module 830, and a second generation module 840.

[0167] The first generation module 810 is used to call the large model to process the input information and generate at least one first task information to be executed, wherein the at least one first task information is used to characterize the first intent of the input information.

[0168] The first display module 820 is used to display at least one first task information.

[0169] The determining module 830 is configured to determine at least one second task information in response to an editing operation on at least one first task information, wherein the at least one second task information is used to characterize a second intent of the input information.

[0170] The second generation module 840 is used to call the large model to generate output information that matches the second intent based on at least one second task information.

[0171] According to an embodiment of this disclosure, the first task information includes at least one parameter information in the input information; the first display module 820 includes: a first display submodule, configured to display at least one parameter information for each first task information, wherein each parameter information is associated with a first control for editing the parameter information.

[0172] According to embodiments of this disclosure, the determining module 830 includes: for each first task information, a second display submodule, configured to display at least one candidate parameter information in response to an operation on a first control associated with target parameter information in at least one parameter information; and

[0173] The first operation submodule is used to replace the target parameter information with the selected candidate parameter information in response to the selection operation of at least one candidate parameter information, so as to obtain the edited second task information.

[0174] According to embodiments of this disclosure, the first task information includes multiple parameter information with dependencies, and the determining module 830 includes: for each piece of the first task information,

[0175] The third display submodule is used to display at least one candidate parameter information in response to an operation of the first control associated with the target parameter information in at least one parameter information.

[0176] The second operation submodule is used to, in response to the selection operation of at least one candidate parameter information, replace the target parameter information with the selected candidate parameter information, and delete at least one parameter information that is dependent on the target parameter information.

[0177] The third operation submodule is used to obtain the second task information in response to the addition operation of at least one parameter information that is dependent on the target parameter information.

[0178] According to embodiments of this disclosure, the determining module 830 further includes:

[0179] The first determination submodule is used to determine the field information to which the target parameter information belongs.

[0180] The second determination submodule is used to determine at least one parameter information other than the target parameter information in the field information as at least one candidate parameter information.

[0181] According to embodiments of this disclosure, each first task information is associated with a second control and / or a third control, wherein the second control is used to add and / or delete at least one parameter information in the first task information; the third control is used to delete the first task information and / or add new first task information after the first task information; the determining module 830 includes:

[0182] A fourth operation submodule is configured to, in response to an operation on a second control associated with target task information in at least one first task information, add / delete at least one parameter information in the target task information to obtain at least one second task information. And / or a fifth operation submodule is configured to, in response to an operation on a third control associated with the target task information, delete the target task information, and / or add new first task information after the first task information to obtain at least one second task information.

[0183] According to embodiments of this disclosure, the first generation module includes:

[0184] The intent analysis submodule is used to call the large model to perform intent analysis on the input information and obtain the first intent.

[0185] The first generation submodule is used to generate third task information in machine language form based on the first intent and input information.

[0186] The first conversion submodule is used to convert the third task information into at least one first task information in natural language form.

[0187] According to embodiments of this disclosure, the first generation submodule includes:

[0188] The first determining unit is used to determine at least one field of information and parameter information for each field of information based on the input information.

[0189] The first generation unit is used to generate third task information in machine language form based on the first intent and the parameter information of each field.

[0190] According to embodiments of this disclosure, the first determining unit includes:

[0191] The first determining subunit is used to determine field description information similar to each parameter information in the input information.

[0192] The second determining subunit is used to determine the field information corresponding to each field description information based on the attribute information of the first dimension field.

[0193] The third determining sub-unit is used to determine the display format of each field information based on the attribute information of the second dimension field.

[0194] According to embodiments of this disclosure, the first conversion submodule includes:

[0195] The conversion unit is used to convert third task information in machine language form into fourth task information in natural language form.

[0196] The splitting unit is used to split the fourth task information into at least one first task information according to the complexity of the fourth task information.

[0197] According to embodiments of this disclosure, the second generation module 840 includes:

[0198] The second conversion submodule is used to convert at least one fifth task information into at least one task vector information in vector form.

[0199] The second generation submodule is used to call the large model and generate output information based on at least one task vector information.

[0200] According to embodiments of this disclosure, the large-model-based information interaction device 800 further includes:

[0201] The first conversion module is used to convert at least one second task information into at least one fifth task information in machine language form.

[0202] The second conversion module converts at least one fifth task information into intent description information in natural language form.

[0203] The second display module is used to display intent description information.

[0204] The third generation module is used to respond to the intent confirmation operation on the intent description information, call the large model, and generate output information that matches the second intent based on at least one fifth task information.

[0205] According to embodiments of this disclosure, the large-model-based information interaction device 800 further includes:

[0206] The third display module is used to display at least one second task information in response to an intent editing operation on the intent description information.

[0207] A task update module is configured to determine at least one sixth task information in response to an editing operation on at least one second task information, wherein the at least one sixth task information is used to characterize a third intent of the input information.

[0208] The fourth generation module is used to call the large model and generate output information that matches the third intent based on at least one sixth task information.

[0209] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0210] According to an embodiment of the present disclosure, an electronic device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method described above.

[0211] According to embodiments of the present disclosure, a non-transitory computer-readable storage medium stores computer instructions, wherein the computer instructions are used to cause a computer to perform the methods described above.

[0212] According to an embodiment of this disclosure, a computer program product includes a computer program that, when executed by a processor, implements the method described above.

[0213] Figure 9 A schematic block diagram of an artificial intelligence agent according to an embodiment of the present disclosure is shown.

[0214] In embodiments of this disclosure, the von Neumann architecture in modern computer theory is inspired, such as... Figure 9 As shown, the AI ​​agent 900 may include five core modules: input module 910, processing module 920 and output module 930.

[0215] The input module 910 is responsible for receiving or sensing information such as queries, requests, instructions, signals, or data from the outside world (e.g., users or the external environment), and converting it into a format that the AI ​​agent 900 can understand and process. The input module 910 is the primary link for the AI ​​agent 900 to interact with the outside world. It enables the AI ​​agent 900 to efficiently and accurately obtain the necessary "sensory" information from the outside world and respond to this information.

[0216] In the example, input module 910 can input the input information described above.

[0217] In the example, the processing module 920 is the core support for the AI ​​agent 900's ability to handle complex tasks. The processing module 920 is used to determine the target task based on the input information received by the input module 910, determine the large model based on the target task, and execute the information interaction method based on the large model described above by calling the large model.

[0218] In the example, output module 930 can output the output information described above.

[0219] In the example, the processing module 920 may include a control unit 921, a storage unit 922, and a processing unit 923.

[0220] During operation, the control unit 921 will continuously interact with the storage unit 922, the arithmetic unit 923, and / or the output module 930. However, it should be noted that in the embodiments of this disclosure, the control unit 921 initiates communication with the storage unit 922, the arithmetic unit 923, and / or the output module 930 as a single initiator, and there is no communication coupling between the storage unit 922, the arithmetic unit 923, and the output module 930.

[0221] In the example, the performance of the control unit 921 is closely related to the large model on which the AI ​​agent 900 is based. To fully leverage the capabilities of the large model, the internal structure of the control unit 921 can be designed to be highly configurable and scalable to handle various types of tasks and requirements in real-world scenarios.

[0222] Storage unit 922 can be responsible for remembering information such as historical dialogues and event streams. For example, previous prompts, information on the first to sixth tasks can be included in storage unit 922.

[0223] The computation unit 923 can be viewed as a predefined tool library. Controls for vector encoding, display controls, etc., as mentioned above, can be included in the computation unit 923.

[0224] In the example, after acquiring input information, the AI ​​agent 900 can use the input information to determine the target task and, based on the target task, determine a large model. The input information can be stored in the storage unit 922 of the processing module 920. The control unit 921 can call the large model to retrieve the input information from the storage unit 922 and process it to generate at least one first task information to be executed. The control unit 921 calls the display control in the calculation module 923 to display at least one first task information. In response to the editing operation of at least one first task information, the control unit 921 determines at least one second task information and stores it in the storage unit 922. The control unit 921 can retrieve at least one second task information from the storage unit 922 and call the large model to generate output information matching the second intent based on at least one second task information. Then, the control unit 921 transmits the output information to the output module 930.

[0225] The AI ​​agent 900 according to embodiments of this disclosure can simply and effectively improve the level of intelligence, as well as enhance flexibility and versatility.

[0226] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0227] According to an embodiment of the present disclosure, an electronic device includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method described above.

[0228] According to embodiments of the present disclosure, a non-transitory computer-readable storage medium stores computer instructions, wherein the computer instructions are used to cause a computer to perform the method described above.

[0229] According to an embodiment of this disclosure, a computer program product includes a computer program that, when executed by a processor, implements the method described above.

[0230] Figure 10The diagram schematically illustrates an electronic device suitable for implementing a large-model-based information interaction method according to embodiments of the present disclosure. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0231] like Figure 10 As shown, the electronic device 1000 includes a computing unit 1001, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 1002 or a computer program loaded from a storage unit 1008 into a random access memory (RAM) 1003. The RAM 1003 may also store various programs and data required for the operation of the electronic device 1000. The computing unit 1001, ROM 1002, and RAM 1003 are interconnected via a bus 1004. An input / output (I / O) interface 1005 is also connected to the bus 1004.

[0232] Multiple components in electronic device 1000 are connected to input / output (I / O) interface 1005, including: input unit 1006, such as keyboard, mouse, etc.; output unit 1007, such as various types of monitors, speakers, etc.; storage unit 1008, such as disk, optical disk, etc.; and communication unit 1009, such as network card, modem, wireless transceiver, etc. Communication unit 1009 allows electronic device 1000 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0233] The computing unit 1001 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 1001 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose 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 1001 performs the various methods and processes described above, such as the image search method. For example, in some embodiments, the image search method may be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 1008. In some embodiments, part or all of the computer program may be loaded and / or installed on the electronic device 1000 via ROM 1002 and / or communication unit 1009. When the computer program is loaded into RAM 1003 and executed by the computing unit 1001, one or more steps of the image search method described above may be performed. Alternatively, in other embodiments, the computing unit 1001 may be configured to perform the image search method by any other suitable means (e.g., by means of firmware).

[0234] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0235] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0236] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction 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 be, 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 machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0237] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, 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 sound input, voice input, or tactile input).

[0238] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0239] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, distributed system servers, or servers incorporating blockchain technology.

[0240] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0241] The specific embodiments described above do not constitute a limitation on the scope of protection 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 spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. An information interaction method based on a large model, comprising: The large model is invoked to process the input information and generate at least one first task information to be executed, wherein at least one first task information is used to characterize the first intent of the input information; Display at least one piece of information about the first task; In response to an editing operation on at least one of the first task information, at least one second task information is determined, wherein the at least one second task information is used to characterize a second intent of the input information; The large model is invoked to generate output information that matches the second intent based on at least one piece of the second task information.

2. The method according to claim 1, wherein, The first task information includes at least one parameter from the input information; The display of at least one piece of the first task information includes: For each of the first task information, at least one of the parameter information is displayed, wherein each of the parameter information is associated with a first control for editing the parameter information.

3. The method according to claim 2, wherein, The step of determining at least one second task piece of information in response to an editing operation on at least one piece of the first task information includes: for each piece of the first task information... In response to an operation on a first control associated with at least one of the target parameter information, at least one candidate parameter information is displayed; and In response to the selection operation of at least one of the candidate parameter information, the target parameter information is replaced with the selected candidate parameter information to obtain the edited second task information.

4. The method according to claim 2, wherein, The first task information includes multiple parameter information that have dependencies, and determining at least one second task information in response to an editing operation on at least one piece of the first task information includes: for each piece of the first task information, In response to an operation on a first control associated with at least one of the target parameter information, at least one candidate parameter information is displayed; In response to a selection operation on at least one of the candidate parameter information, the target parameter information is replaced with the selected candidate parameter information, and at least one of the parameter information that is dependent on the target parameter information is deleted; and In response to an editing operation on at least one of the parameter information that is dependent on the target parameter information, second task information is obtained.

5. The method according to claim 3 or 4, further comprising: Determine the field information to which the target parameter information belongs; as well as At least one parameter in the field information other than the target parameter information is determined as at least one candidate parameter.

6. The method according to claim 2, wherein, Each piece of the first task information is associated with a second control and / or a third control, wherein the second control is used to add and / or delete at least one of the parameter information in the first task information; and the third control is used to delete the first task information and / or add new first task information after the first task information. The step of determining at least one second task information in response to an editing operation on at least one piece of first task information includes: In response to an operation on at least one of the second controls associated with target task information in the first task information, at least one of the parameter information is added / deleted in the target task information to obtain at least one piece of second task information; and / or In response to an operation on the third control associated with the target task information, the target task information is deleted, and / or a new first task information is added after the first task information to obtain at least one second task information.

7. The method according to claim 1, wherein, The process of calling the large model to process the input information and generate at least one first task information to be executed includes: The large model is invoked to perform intent analysis on the input information to obtain the first intent; Based on the first intent and the input information, generate third task information in machine language form; and Convert the third task information into at least one of the first task information in natural language form.

8. The method according to claim 7, wherein, The step of generating third task information in machine language form based on the first intent and the input information includes: Based on the input information, at least one field of information and parameter information for each field of information are determined; and Based on the first intent and the parameter information of each of the field information, the third task information in machine language form is generated.

9. The method according to claim 8, wherein, The step of determining at least one field information and parameter information for each field information based on the input information includes: Determine field description information that is similar to each parameter information in the input information; Based on the attribute information of the first dimension field, determine the field information corresponding to each of the field description information; and Based on the attribute information of the second dimension field, determine the display format of each field information.

10. The method according to any one of claims 7-9, wherein, The step of converting the third task information into at least one piece of the first task information in natural language form includes: Convert the third task information in machine language form into the fourth task information in natural language form; and Based on the complexity of the fourth task information, the fourth task information is split into at least one piece of the first task information.

11. The method according to any one of claims 1 to 10, wherein, The step of calling the large model to generate output information matching the second intent based on at least one piece of the second task information includes: Convert at least one of the second task information into at least one fifth task information in machine language form; Convert at least one of the fifth task information into at least one task vector information in vector form; The large model is invoked to generate the output information based on at least one of the task vector information.

12. The method of claim 11, further comprising: Convert at least one of the fifth task information into intent description information in natural language form; Display the intent description information; as well as In response to the intent confirmation operation on the intent description information, the large model is invoked to generate output information that matches the second intent based on at least one of the fifth task information.

13. The method of claim 12, further comprising: In response to an intent editing operation on the intent description information, at least one piece of the second task information is displayed; In response to an editing operation on at least one of the second task information, at least one sixth task information is determined, wherein the at least one of the sixth task information is used to characterize a third intent of the input information; as well as The large model is invoked to generate output information that matches the third intent, based on at least one of the sixth task information.

14. An information interaction device based on a large model, comprising: The first generation module is used to call a large model to process the input information and generate at least one first task information to be executed, wherein at least one first task information is used to characterize the first intent of the input information; The first display module is used to display at least one piece of information about the first task. A determining module is configured to determine at least one second task information in response to an editing operation on at least one first task information, wherein the at least one second task information is used to characterize a second intent of the input information; The second generation module is used to call the large model to generate output information that matches the second intent based on at least one of the second task information.

15. The apparatus according to claim 14, wherein, The first task information includes at least one parameter from the input information; The first display module includes: The first display submodule is used to display at least one parameter information for each of the first task information, wherein each parameter information is associated with a first control for editing the parameter information.

16. The apparatus according to claim 15, wherein, The determining module includes: for each piece of the first task information... The second display submodule is configured to display at least one candidate parameter information in response to an operation on a first control associated with at least one target parameter information among the parameter information; and A first operation submodule is configured to, in response to a selection operation on at least one of the candidate parameter information, replace the target parameter information with the selected candidate parameter information to obtain the edited second task information.

17. The apparatus according to claim 15, wherein, The first task information includes multiple parameter information that have dependencies, and the determining module includes: for each piece of the first task information, The third display submodule is used to display at least one candidate parameter information in response to an operation of a first control associated with the target parameter information in at least one of the parameter information; The second operation submodule is configured to, in response to a selection operation on at least one of the candidate parameter information, replace the target parameter information with the selected candidate parameter information, and delete at least one of the parameter information that is dependent on the target parameter information; and The third operation submodule is used to obtain second task information in response to an operation that adds at least one parameter information that is dependent on the target parameter information.

18. The apparatus according to claim 16 or 17, wherein the determining module further comprises: The first determining submodule is used to determine the field information to which the target parameter information belongs; as well as The second determining submodule is used to determine at least one parameter information other than the target parameter information in the field information as at least one candidate parameter information.

19. The apparatus according to claim 15, wherein, Each of the first task information is associated with a second control and / or a third control, wherein the second control is used to add and / or delete at least one of the parameter information in the first task information; The third control is used to delete the first task information and / or add new first task information after the first task information; The determining module includes: The fourth operation submodule is used to respond to an operation on at least one of the second controls associated with the target task information in the first task information, to add / delete at least one of the parameter information in the target task information, so as to obtain at least one piece of the second task information; and / or The fifth operation submodule is used to delete the target task information and / or add new first task information after the first task information in response to the operation of the third control associated with the target task information, so as to obtain at least one second task information.

20. The apparatus according to claim 14, wherein, The first generation module includes: The intent analysis submodule is used to call the large model to perform intent analysis on the input information to obtain the first intent; The first generation submodule is configured to generate third task information in machine language form based on the first intent and the input information; and The first conversion submodule is used to convert the third task information into at least one piece of the first task information in natural language form.

21. The apparatus according to claim 20, wherein, The first generation submodule includes: The first determining unit is configured to determine at least one field information and parameter information for each field information based on the input information; and The first generation unit is configured to generate the third task information in machine language form based on the first intent and the parameter information of each of the field information.

22. The apparatus according to claim 21, wherein, The first determining unit includes: The first determining subunit is used to determine field description information similar to each parameter information in the input information; The second determining subunit is used to determine the field information corresponding to each of the field description information based on the first dimension field attribute information; and The third determining subunit is used to determine the display format of each field information based on the second dimension field attribute information.

23. The apparatus according to any one of claims 20-22, wherein, The first conversion submodule includes: A conversion unit is used to convert third-task information in machine language form into fourth-task information in natural language form; and The splitting unit is used to split the fourth task information into at least one first task information according to the complexity of the fourth task information.

24. The apparatus according to claim 23, wherein, The second generation module includes: The second conversion submodule is used to convert at least one of the fifth task information into at least one task vector information in vector form; The second generation submodule is used to call the large model and generate the output information based on at least one of the task vector information.

25. The apparatus according to any one of claims 14 to 24, further comprising: A first conversion module is used to convert at least one second task information into at least one fifth task information in machine language form; The second conversion module converts at least one of the fifth task information into intention description information in natural language form; The second display module is used to display the intent description information; as well as The third generation module is used to, in response to the intent confirmation operation on the intent description information, invoke the large model and generate output information that matches the second intent based on at least one of the fifth task information.

26. The apparatus of claim 25, further comprising: The third display module is used to display at least one piece of the second task information in response to an intent editing operation on the intent description information; A task update module is configured to determine at least one sixth task information in response to an editing operation on at least one second task information, wherein at least one of the sixth task information is used to characterize a third intent of the input information; as well as The fourth generation module is used to call the large model and generate output information that matches the third intent based on at least one of the sixth task information.

27. An intelligent agent, comprising: The input module is used to receive input information; The processing module is configured to determine a target task based on the input information received by the input module, determine a large model based on the target task, and obtain output information by calling the large model to execute the method as described in any one of claims 1 to 13. An output module is used to output the output information obtained by the processing module.

28. 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 to enable the at least one processor to perform the method of any one of claims 1 to 13.

29. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to any one of claims 1 to 13.

30. A computer program product comprising a computer program that, when executed by a processor, implements the method according to any one of claims 1 to 13.

Images