A method and system for generating intelligent applications based on knowledge graphs and intent-driven approaches

Abstract

This invention discloses a method and system for generating intelligent applications based on knowledge graphs and intent-driven approaches, belonging to the field of artificial intelligence technology. The method includes: acquiring natural language intents; obtaining initial intents and initial entities based on the natural language intents; anchoring the initial intents and initial entities to obtain declarative questions; searching the domain knowledge graph using a pre-built AI application planner to obtain operators; obtaining an execution plan based on the target state; transforming the execution plan to obtain a task dependency graph; embedding the task dependency graph into a complete application graph; performing application correction on the complete application graph using a multimodal correction method to obtain an initial application graph; acquiring user modification instructions; and obtaining the optimal execution plan and updating the application graph based on the modification instructions. This invention ensures the deterministic logical verification and reliability of the output results, reduces the probability of risk occurrence, improves intent understanding capabilities, and enables application generation without manual design.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence technology, specifically relating to a method and system for generating intelligent applications based on knowledge graphs and intent-driven approaches. Background Technology

[0002] In recent years, with the continuous development of artificial intelligence technology, how to automatically generate complex intelligent applications based on the user's advanced intent has become one of the core goals pursued by the industry. In existing technologies, the following approaches are mainly explored: (1) Visual Workflow Systems with manual orchestration: represented by AWS Step Functions, n8n, or Zapier, where developers manually design a Directed Acyclic Graph (DAG) consisting of multiple tasks through a graphical interface; (2) Simple task generation based on large language models: represented by various AI programming assistants, which understand the user's natural language instructions and call one or more tools to complete a relatively simple task; (3) Declarative query engines: represented by SQL, where the user "declares" what they want to do, and the system plans "how to do it"; (4) "Process-driven" framework based on multi-agent collaboration: represented by Microsoft AutoGen, CrewAI, and LangChain. LangGraph, representing an explicit state graph control flow technique, is a typical example. Developers predefine a set of agents with different roles (such as planners, coders, and testers) and manually design the collaboration process between different roles. When a user makes an intention, the system runs through this fixed collaboration process, and multiple agents complete the task step by step through multiple rounds of interaction.

[0003] However, the above methods all have corresponding shortcomings, such as: (1) relying entirely on manual design and lacking the ability to understand intent: the system cannot understand high-level business objectives and autonomously plan the technical steps required to achieve the objectives based on the business objectives; (2) lacking in-depth process planning capabilities and reliability: it cannot solve the complex process planning problem of "how the whole system should go", and is prone to illusion when facing complex instructions, and the generated results lack certainty; (3) strong limitations: it deals with structured and unambiguous language, and the intent is limited to data query; (4) lacking deterministic logical verification and unable to guarantee reliability: the behavior of the entire system is dominated by the preset dialogue process or state diagram, and cannot start from the final goal declared by the user, dynamically and autonomously plan a new and logically optimal execution path, and the logic and code generated in the collaboration process lack a mandatory verification link, and cannot guarantee the reliability and logical completeness of the output results, which poses corresponding risks.

[0004] Therefore, given the aforementioned shortcomings, how to provide an effective technical solution to address the problems of lack of intent understanding, deep process planning capabilities and reliability, deterministic logic verification, inability to guarantee reliability, and strong limitations in existing technologies has become an urgent problem to be solved in existing technologies. Summary of the Invention

[0005] The purpose of this invention is to provide a method and system for generating intelligent applications based on knowledge graphs and intent-driven approaches, in order to solve the aforementioned problems existing in the prior art.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] In a first aspect, the present invention provides a method for generating intelligent applications based on knowledge graphs and intent-driven approaches, comprising the following steps:

[0008] Step 1: Obtain the user's natural language intent, identify and extract entities from the natural language intent to obtain the initial intent and initial entities;

[0009] Step 2: Anchor the initial intent and initial entity based on the pre-built domain knowledge graph to obtain a declarative question, which includes an initial state and a target state;

[0010] Step 3: Using a pre-built AI application planner, the domain knowledge graph is searched based on the target state to obtain operators. Based on the operators and the initial state, the target state is deduced in a reverse chain to obtain an execution plan, which includes an operator execution sequence.

[0011] Step 4: Transform the execution plan to obtain a task dependency graph, embed the task dependency graph into the pre-built complete application graph, and perform application correction on the complete application graph based on the multimodal correction method to obtain the initial application graph;

[0012] Step 5: Obtain the user's modification instructions, and execute steps 3 and 4 based on the modification instructions to obtain the optimal execution plan and update the application graph.

[0013] In one possible design, natural language intent is identified and entity extraction is performed to obtain initial intent and initial entities, including:

[0014] The natural language intent is cleaned based on the preset cleaning rules to obtain the cleaned natural language intent.

[0015] The cleaned natural language intent is denoised to obtain the denoised natural language intent, and the denoised natural language intent is standardized to obtain the preprocessed natural language intent.

[0016] The preprocessed natural language intent is segmented to obtain natural language words;

[0017] LLM is used to perform intent recognition and entity extraction on natural language words to obtain the initial intent and initial entities.

[0018] In one possible design, the initial intent and initial entities are anchored based on a pre-built domain knowledge graph, resulting in declarative questions, including:

[0019] Acquire knowledge graph data within the domain, the knowledge graph data including a capability operator library, operator reduction, and entity and logical constraints, identify the capability operator library, operator reduction, and entity and logical constraints, and obtain graph entities and connection relationships;

[0020] Constructing domain knowledge graphs based on graph entities and connections;

[0021] The initial intent and initial entities are matched with the domain knowledge graph to obtain declarative questions.

[0022] In one possible design, the multimodal correction method includes natural language processing, graphical operations, and / or code editing operations; the application correction of the complete application graph based on the multimodal correction method to obtain the initial application graph includes at least one of the following steps:

[0023] The complete application graph is modified based on natural language processing to obtain the first initial application graph.

[0024] The complete application graph is modified based on graphical operations to obtain the second initial application graph;

[0025] The complete application graph is modified based on code editing operations to obtain the third initial application graph.

[0026] In one possible design, steps three and four are executed based on the modified instructions, including:

[0027] The modification instructions are cleaned and denoised to obtain denoised modification instructions.

[0028] The denoised modification instructions are standardized to obtain standardized modification instructions.

[0029] The standardized modification instructions are segmented and grammatically corrected to obtain preprocessed modification instructions.

[0030] The preprocessed modification instructions are translated using a unified translator to obtain change data, which includes changes to the target state and changes to formal constraints.

[0031] Perform steps three and four based on the changed data.

[0032] In one possible design, the AI ​​application planner is built based on an AI planning algorithm.

[0033] In one possible design, the execution plan is obtained by performing a reverse chain derivation of the target state based on the operator and the initial state, including:

[0034] By reverse-engineering the target state, the first operator is obtained;

[0035] The domain knowledge graph is searched based on the preconditions of the first operator to obtain new sub-targets;

[0036] By reverse-engineering the new sub-objective, the second operator is obtained;

[0037] Based on the preconditions of the second operator, the domain knowledge graph is searched to obtain a new sub-target;

[0038] Repeat the above steps until the iteration termination condition is reached to obtain the execution plan. The iteration termination condition is the prerequisite condition that the conditions in the initial state satisfy the target state.

[0039] Secondly, the present invention provides an intelligent application generation system based on knowledge graphs and intent-driven approaches, comprising:

[0040] The intent understanding module is used to obtain the user's natural language intent, identify the natural language intent and extract entities to obtain the initial intent and initial entities;

[0041] The knowledge graph anchoring module is used to anchor the initial intent and initial entity based on the pre-built domain knowledge graph to obtain a declarative question, which includes an initial state and a target state.

[0042] The AI ​​application planning module is used to use a pre-built AI application planner to search the domain knowledge graph based on the target state to obtain operators, and to perform a reverse chain derivation of the target state based on the operators and the initial state to obtain an execution plan, wherein the execution plan includes an operator execution sequence.

[0043] The application graph generation module is used to transform the execution plan to obtain a task dependency graph, embed the task dependency graph into a pre-built complete application graph, and perform application correction on the complete application graph based on a multimodal correction method to obtain an initial application graph.

[0044] The replanning correction module is used to obtain the user's modification instructions. Based on the understanding of the execution intent of the modification instructions, knowledge graph anchoring, AI application planning and application graph generation steps, the optimal execution plan and updated application graph are obtained.

[0045] Thirdly, the present invention provides a computer device comprising a memory, a processor, and a transceiver that are sequentially and communicatively connected, wherein the memory is used to store a computer program, the transceiver is used to send and receive messages, and the processor is used to read the computer program and execute the intelligent application generation method based on knowledge graph and intent-driven as described in any of the above.

[0046] Fourthly, the present invention provides a computer program product containing instructions that, when the instructions are executed on a computer, cause the computer to perform the intelligent application generation method based on knowledge graphs and intent-driven methods as described in any of the preceding claims.

[0047] The beneficial effects of this invention are as follows:

[0048] (1) This invention discloses a method and system for generating intelligent applications based on knowledge graphs and intent-driven approaches. By performing preliminary identification and entity extraction of the user's natural language intent and using an AI application planner to obtain an execution plan, the intent understanding ability is improved, and the application can be generated without manual design.

[0049] (2) This invention processes declarative problems through an AI application planner and performs application correction through a multimodal correction method, thereby improving the certainty of the generated results and avoiding errors caused by illusions.

[0050] (3) The natural language intent obtained in this invention is unstructured and ambiguous natural language, and there are no limitations on the type of natural language intent obtained.

[0051] (4) In this invention, a domain knowledge graph is used to perform mandatory verification on the generated declarative questions, ensuring the logical verification and reliability of the output results, reducing the probability of risk occurrence, and facilitating application and promotion. Attached Figure Description

[0052] Figure 1 A flowchart of the intelligent application generation method provided in the first aspect of the present invention;

[0053] Figure 2 A block diagram of the intelligent application generation system provided in the second aspect of the present invention. Detailed Implementation

[0054] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the present invention will be briefly introduced below in conjunction with the accompanying drawings and descriptions of the embodiments or the prior art. Obviously, the following description of the structure of the accompanying drawings is only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. It should be noted that the description of these embodiments is for the purpose of helping to understand the present invention, but does not constitute a limitation of the present invention.

[0055] It should be understood that although the terms first, second, etc., may be used herein to describe various units, these units should not be limited by these terms. These terms are only used to distinguish one unit from another. For example, a first unit may be referred to as a second unit, and similarly, a second unit may be referred to as a first unit, without departing from the scope of the exemplary embodiments of the invention.

[0056] It should be understood that the term "and / or" that may appear in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can mean: A exists alone, B exists alone, and A and B exist simultaneously. The term " / and" that may appear in this document describes another relationship between related objects, indicating that two relationships can exist. For example, A / and B can mean: A exists alone, and A and B exist alone. In addition, the character " / " that may appear in this document generally indicates that the related objects before and after it are in an "or" relationship.

[0057] Example:

[0058] like Figure 1 As shown, the first aspect of this embodiment provides a method for generating intelligent applications based on knowledge graphs and intent-driven approaches. This method can be executed, but is not limited to, by a computer device or virtual machine with certain computing resources, such as a personal computer or smartphone, or by a virtual machine. The intelligent application generation method includes, but is not limited to, the following steps:

[0059] S1. Obtain the user's natural language intent, identify and extract entities from the natural language intent to obtain the initial intent and initial entities;

[0060] For example, if a user's natural language intent is "Analyze this meeting recording, summarize Zhang San's points, and give me a Word document of the minutes," then the initial intents are "analyze the recording," "summarize the points," and "generate minutes"; the initial entities are "Zhang San" and "Word."

[0061] Specifically, in step S1, the natural language intent is identified and entities are extracted to obtain the initial intent and initial entities, including:

[0062] S11. Perform data cleaning on the natural language intent based on preset cleaning rules to obtain the cleaned natural language intent;

[0063] The preset cleaning rules are manually set cleaning rules, including but not limited to handling missing data and handling abnormal data.

[0064] S12. Denoise the cleaned natural language intent to obtain the denoised natural language intent, and standardize the denoised natural language intent to obtain the preprocessed natural language intent.

[0065] The denoising process includes preprocessing the natural language intent to remove irrelevant punctuation and modal particles, and standardizing the denoised natural language intent to reduce the impact of large discrepancies on the processing results in subsequent processing.

[0066] S13. Perform word segmentation on the preprocessed natural language intent to obtain natural language words;

[0067] S14. Use LLM (Large Language Model) to perform intent recognition and entity extraction on natural language words to obtain initial intent and initial entities.

[0068] S2. Anchor the initial intent and initial entity based on the pre-built domain knowledge graph to obtain a declarative question, which includes an initial state and a target state;

[0069] In this embodiment, the process of anchoring the initial intent and initial entity based on the pre-built domain knowledge graph includes, for example, matching operators such as ASR (speech-to-text) and SpeakerDiarization in the domain knowledge graph based on "analyzing the recording"; matching the TopicSummary operator in the domain knowledge graph based on "summarizing the viewpoint"; and matching the FormatConversion operator in the domain knowledge graph based on "generating the minutes", with the target format being docx.

[0070] The initial state of the resulting declarative problem is: { has(file: "meeting.wav")}; the target state is: { has(file: "*.docx"), contains(file, content_from(TopicSummary)),summary_target("Zhang San")}.

[0071] Specifically, in step S2, the initial intent and initial entities are anchored based on the pre-built domain knowledge graph to obtain declarative questions, including:

[0072] S21. Obtain knowledge graph data within the domain, wherein the knowledge graph data includes a capability operator library, operator reduction and entity and logical constraints, identify the capability operator library, operator reduction and entity and logical constraints, and obtain graph entities and connection relationships;

[0073] S22. Construct a domain knowledge graph based on graph entities and connections;

[0074] S23. Match the initial intent and initial entities with the domain knowledge graph to obtain the declarative question.

[0075] It should be noted that each capability operator in the knowledge graph data of this embodiment has a formalized operator specification. This operator specification strictly defines the preconditions and effects of the target, providing a strict "physical law" for each step of the subsequent AI application planner's deduction, namely the operator specification.

[0076] Furthermore, in this embodiment, the process of anchoring the initial intent and the initial entity is not a simple semantic matching, but rather uses the deterministic rules of the domain knowledge graph to perform mandatory verification on the fuzzy and potentially illusory preliminary results generated by intent recognition and entity extraction. When a mismatch or logical contradiction is found, it will be corrected or refined to ensure the logical completeness and accuracy of the declarative question.

[0077] S3. Using a pre-built AI application planner, the domain knowledge graph is searched based on the target state to obtain operators. Based on the operators and the initial state, the target state is deduced in a reverse chain to obtain an execution plan, which includes an operator execution sequence.

[0078] Furthermore, the AI ​​application planner is built on AI planning algorithms, and the AI ​​application planners used include, but are not limited to, planners based on Planning Domain Definition Language (PDDL) or Hierarchical Task Network (HTN) planners.

[0079] In step S3, the target state is derived by a reverse chain derivation based on the operator and the initial state to obtain the execution plan, which includes the following steps:

[0080] S31. The first operator is obtained by reverse derivation of the target state;

[0081] S32. Based on the preconditions of the first operator, search the domain knowledge graph to obtain new sub-targets;

[0082] S33. The second operator is obtained by reverse derivation of the new sub-objective;

[0083] S34. Based on the preconditions of the second operator, search the domain knowledge graph to obtain a new sub-target;

[0084] S35. Repeat the above steps until the iteration termination condition is reached to obtain the execution plan. The iteration termination condition is the prerequisite condition that the conditions in the initial state satisfy the target state.

[0085] For example, if the target state is has("*.docx"), performing a reverse chain of reasoning on the target state yields the first operator, FormatConversion. The precondition for FormatConversion is input in Text format, which is a new sub-target. Then, based on this new sub-target, we find the operator TopicSummary. TopicSummary generates input in Text format, and its precondition is that it requires text labeled with the speaker's name (LabeledText), thus creating a new sub-target. Based on this new sub-target, we find the operator SpeakerDiarization. SpeakerDiarization generates LabeledText, and its precondition is that it requires input in Audio format and the complete text (FullText).

[0086] At this point, the initial state includes Audio, and FullText can be generated by the ASR operator. The prerequisite for the ASR operator is Audio, which satisfies the iteration termination condition. The execution plan is [ASR, SpeakerDiarization, TopicSummary, FormatConversion].

[0087] It should be noted that by continuously analyzing the preconditions of the target and searching for operators in the domain knowledge graph that satisfy these conditions, the AI ​​application planner makes the logical foundation of "reverse chain deduction" more solid, and at the same time makes the core position of the domain knowledge graph as a "world rule manual" more prominent.

[0088] S4. Transform the execution plan to obtain a task dependency graph, embed the task dependency graph into the pre-built complete application graph, and perform application correction on the complete application graph based on the multimodal correction method to obtain the initial application graph;

[0089] The task dependency graph is an executable DAG graph, and the executable DAG graph is serialized into a JSON file; the multimodal correction method includes at least one of natural language processing, graphical operation and / or code editing operation;

[0090] Specifically, in step S4, the complete application graph is modified based on the multimodal modification method to obtain the initial application graph, which includes at least one of the following steps:

[0091] S41. Apply natural language processing to the complete application graph to obtain the first initial application graph;

[0092] S42. Apply the corrective action to the complete application graph using graphical operations to obtain the second initial application graph;

[0093] S43. Based on the code editing operation, the complete application graph is modified to obtain the third initial application graph.

[0094] S5. Obtain the user's modification instructions, and execute steps S3 to S4 based on the modification instructions to obtain the optimal execution plan and update the application graph.

[0095] In this embodiment, the modification instructions include, but are not limited to, natural language instructions, graphical interface operation instructions, or code editing instructions. For example, the modification instruction is "before summarizing, perform spelling and grammar correction on the text."

[0096] Specifically, in step S5, steps S3 to S4 are executed based on the modification instruction, including:

[0097] S51. Perform data cleaning and noise reduction on the modification instructions to obtain the noise-reduced modification instructions;

[0098] S52. Standardize the noise-reduced modification instructions to obtain standardized modification instructions;

[0099] S53. Perform word segmentation and syntax correction on the standardized modification instructions to obtain preprocessed modification instructions;

[0100] S54. Use a unified translator to translate the preprocessed modification instructions to obtain change data, which includes changes to the target state and changes to formal constraints;

[0101] For example, the formal constraint changes to {execute(TextCorrection) after execute(SpeakerDiarization), execute(TextCorrection) before execute(TopicSummary)}.

[0102] S55. Perform steps three to four based on the changed data.

[0103] In this embodiment, after receiving the changed data, the AI ​​application planner re-plans the path. The knowledge graph contains the `TextCorrection` operator, whose precondition is `Text` and whose effect is also `Text`. It inserts this operator into the plan, generating a new optimal execution plan: [ASR, SpeakerDiarization, TextCorrection, TopicSummary, FormatConversion], and generates an updated application graph based on the new optimal execution plan.

[0104] It should be noted that the correction method proposed in this embodiment is "replanning," which is not a manual or "patching" modification of the already generated application graph. Manual and "patching" modifications are prone to disrupting the overall logical consistency and optimality of the application. In this embodiment, a unified translator is used to translate the modification instructions to obtain changes to the target state or formal constraints, and trigger a completely new end-to-end planning process. This ensures that the modified application graph has complete, self-consistent, and overall optimal logic, which greatly improves the maintainability and reliability of the application throughout its entire lifecycle.

[0105] Based on the above disclosure, the domain knowledge graph in this invention is not a passive data source, but actively and forcibly anchors the identified intents and entities, and provides strict operator specifications for the deduction of the AI ​​application planner, improving robustness, while ensuring the logical reliability of the final output application. This invention enables the capability to leap from orchestration to planning, and its essence is "Planning & Reasoning" to achieve a qualitative change from passive construction to active planning.

[0106] like Figure 2 As shown, the second aspect of this embodiment provides an intelligent application generation system based on knowledge graphs and intent-driven methods, including:

[0107] The intent understanding module is used to obtain the user's natural language intent, identify the natural language intent and extract entities to obtain the initial intent and initial entities;

[0108] The knowledge graph anchoring module is used to anchor the initial intent and initial entity based on the pre-built domain knowledge graph to obtain a declarative question, which includes an initial state and a target state.

[0109] The AI ​​application planning module is used to use a pre-built AI application planner to search the domain knowledge graph based on the target state to obtain operators, and to perform a reverse chain derivation of the target state based on the operators and the initial state to obtain an execution plan, wherein the execution plan includes an operator execution sequence.

[0110] The application graph generation module is used to transform the execution plan to obtain a task dependency graph, embed the task dependency graph into a pre-built complete application graph, and perform application correction on the complete application graph based on a multimodal correction method to obtain an initial application graph.

[0111] The replanning correction module is used to obtain the user's modification instructions. Based on the understanding of the execution intent of the modification instructions, knowledge graph anchoring, AI application planning and application graph generation steps, the optimal execution plan and updated application graph are obtained.

[0112] The working process, working details, and technical effects of the intelligent application generation system provided in the second aspect of this embodiment can be found in the intelligent application generation method described in the first aspect, and will not be repeated here.

[0113] This third aspect of the embodiment provides a computer device, including a memory, a processor, and a transceiver sequentially and communicatively connected. The memory stores a computer program, the transceiver sends and receives messages, and the processor reads the computer program and executes the intelligent application generation method as described in the first aspect. Specifically, the memory may include, but is not limited to, random-access memory (RAM), read-only memory (ROM), flash memory, first-in-first-out (FIFO) memory, and / or first-in-last-out (FILO) memory, etc.; the processor may include, but is not limited to, a microprocessor of the STM32F105 series. Furthermore, the computer device may also include, but is not limited to, a power supply module, a display screen, and other necessary components.

[0114] The working process, working details and technical effects of the aforementioned computer device provided in the third aspect of this embodiment can be found in the intelligent application generation method described in the first aspect, and will not be repeated here.

[0115] The fourth aspect of this embodiment provides a computer program product, including a computer program or instructions, which, when executed by a computer, are used to implement the intelligent application generation method as described in the first aspect.

[0116] The working process, working details, and technical effects of the aforementioned computer program product provided in this embodiment can be found in the intelligent application generation method described in the first aspect, and will not be repeated here.

[0117] Finally, it should be noted that the above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A knowledge graph and intent-driven intelligent application generation method, characterized in that, The method comprises the following steps: Step 1: obtaining a natural language intention of a user, identifying and extracting entities of the natural language intention, obtaining an initial intention and initial entities; Step 2: anchoring the initial intention and the initial entities based on a pre-constructed domain knowledge graph to obtain a declarative question, wherein the declarative question comprises an initial state and a target state; Step 3: searching the domain knowledge graph based on the target state using a pre-constructed AI application planner to obtain an operator, performing reverse chain deduction on the target state based on the operator and the initial state to obtain an execution plan, wherein the execution plan comprises an operator execution sequence; Step 4: converting the execution plan to obtain a task dependency graph, embedding the task dependency graph into a pre-constructed complete application graph, and performing application correction on the complete application graph based on a multi-modal correction method to obtain an initial application graph; Step 5: obtaining a modification instruction of the user, and performing steps 3 to 4 based on the modification instruction to obtain an optimal execution plan and an updated application graph. 2.The knowledge graph and intent driven intelligent application generation method of claim 1, wherein, The method for identifying and extracting entities of the natural language intention comprises the following steps: performing data cleaning on the natural language intention based on pre-set cleaning rules to obtain cleaned natural language intention; performing denoising processing on the cleaned natural language intention to obtain denoised natural language intention, and performing standardization processing on the denoised natural language intention to obtain pre-processed natural language intention; performing word segmentation processing on the pre-processed natural language intention to obtain natural language words; performing intention recognition and entity extraction on the natural language words using an LLM to obtain an initial intention and initial entities. 3.The knowledge graph and intent driven intelligent application generation method of claim 1, wherein, The method for anchoring the initial intention and the initial entities based on the pre-constructed domain knowledge graph to obtain a declarative question comprises the following steps: obtaining knowledge graph data in a domain, wherein the knowledge graph data comprises an ability operator library, operator reduction, and entity and logical constraints, identifying the ability operator library, the operator reduction, and the entity and logical constraints to obtain graph entities and connection relationships; constructing a domain knowledge graph based on the graph entities and the connection relationships; matching the initial intention and the initial entities with the domain knowledge graph to obtain a declarative question. 4.The knowledge graph and intent driven intelligent application generation method of claim 1, wherein, The multi-modal correction method comprises natural language processing, graphical operation and / or code editing operation; and the method for performing application correction on the complete application graph based on the multi-modal correction method to obtain an initial application graph comprises at least one of the following steps: performing application correction on the complete application graph based on natural language processing to obtain a first initial application graph; performing application correction on the complete application graph based on graphical operation to obtain a second initial application graph; performing application correction on the complete application graph based on code editing operation to obtain a third initial application graph. 5.The knowledge graph and intent driven intelligent application generation method of claim 1, wherein, The method for performing steps 3 to 4 based on the modification instruction comprises the following steps: performing data cleaning and denoising processing on the modification instruction to obtain denoised modification instruction; performing standardization processing on the denoised modification instruction to obtain standardized modification instruction; performing word segmentation processing and grammar correction on the standardized modification instruction to obtain pre-processed modification instruction; The modified instructions after preprocessing are translated using a unified translator to obtain change data, the change data including changes of the target state and changes of the formalized constraints; Steps three to four are performed based on the change data. 6.The method of claim 1, wherein, The AI application planner is constructed based on an AI planning algorithm. 7.The method of claim 1, wherein, The target state is reversely chain deduced based on the operator and the initial state to obtain an execution plan, including: The target state is reversely deduced to obtain a first operator; The domain knowledge graph is searched based on the precondition of the first operator to obtain a new sub-target; The new sub-target is reversely deduced to obtain a second operator; The domain knowledge graph is searched based on the precondition of the second operator to obtain a new sub-target again; The above steps are repeated until an iteration termination condition is reached to obtain the execution plan, the iteration termination condition being that the precondition of the target state obtained by the condition in the initial state is satisfied.

8. A knowledge graph and intent driven intelligent application generation system for implementing the method of any one of claims 1-7. The method comprises: An intent understanding module is configured to obtain a natural language intent of a user, identify and extract entities from the natural language intent, and obtain an initial intent and initial entities; A knowledge graph anchoring module is configured to anchor the initial intent and initial entities based on a pre-constructed domain knowledge graph to obtain a declarative question, the declarative question including an initial state and a target state; An AI application planning module is configured to search the domain knowledge graph based on the target state using a pre-constructed AI application planner to obtain an operator, and reversely chain deduce the target state based on the operator and the initial state to obtain an execution plan, the execution plan including an operator execution sequence; An application graph generation module is configured to convert the execution plan to obtain a task dependency graph, embed the task dependency graph into a pre-constructed complete application graph, and apply a multi-modal correction method to the complete application graph to obtain an initial application graph; A re-planning correction module is configured to obtain a modification instruction of a user, perform the intent understanding, knowledge graph anchoring, AI application planning and application graph generation steps based on the modification instruction to obtain an optimal execution plan and an updated application graph.

9. A computer device, comprising: The computer program or the instructions realize the intelligent application generation method driven by the knowledge graph and the intent when executed by a computer.

10. A computer program product comprising computer programs or instructions, characterized in that, The computer program or the instructions realize the intelligent application generation method driven by the knowledge graph and the intent when executed by a computer.

Images