A mind map adaptive generation method, device and equipment and storage medium

Abstract

The application discloses a mind map adaptive generation method and device, equipment and storage medium. Its characteristics include: in response to the mind map generation request submitted by the user, the mind map data corresponding to the mind map generation request is identified; the mind map data is routed and decided based on the routing decision engine, the mind map generation path corresponding to the mind map data is determined; if the mind map generation path is a parallel generation path, the mind map data is parallel task decomposed and merged, and the target mind map is determined. The embodiment of the application can effectively process all types of data and super-long documents, has the characteristics of strong universality and high information retention, and the parallel architecture can significantly improve the generation efficiency of the multi-source scene, and ensure that the mind map result always meets the optimal experience of the visualization rendering and user reading of the mind map.

Description

Technical Field

[0001] This invention relates to the field of large language model technology, and in particular to a method, apparatus, device and storage medium for adaptive generation of mind maps. Background Technology

[0002] Mind maps are visualization tools that display concepts and their relationships in a radial, tree-like structure, widely used in note-taking, knowledge management, and learning assistance. Traditional mind maps rely on manual creation, which is inefficient. In recent years, with the rapid improvement of large language model capabilities, it has become possible to automatically analyze document content and generate mind maps using LLM (Large Language Model), and various AI knowledge management platforms have introduced this function. However, the content sources in enterprise-level knowledge management scenarios are far more complex than academic presentations: users may simultaneously select dozens of document files from the knowledge base, uploaded PDF reports, technical articles scraped from web pages, and video subtitle files, requiring the system to uniformly generate a logically consistent mind map from these heterogeneous sources. Furthermore, a single document may contain hundreds of thousands of words, far exceeding the context window limitations of current mainstream LLMs. Existing solutions cannot meet these enterprise-level needs, and there is an urgent need for a method that can adaptively adjust the generation strategy based on content characteristics. Existing technologies typically process ultra-long documents through a process of splitting the document, parallel summarizing, and merging summaries, and support the generation of structured output based on the summaries; or they directly encapsulate them into a Prompt, with the LLM outputting a mind map structure; or they build a summary index for the document, and can combine the structured output capabilities of the LLM to convert the summary into business formats such as mind maps and JSON. Existing technologies suffer from the following problems: lack of adaptive routing capabilities that are aware of content features; simplistic strategies for processing ultra-long documents; severe damage to semantic structure; and a lack of effective tree structures for parallel generation from multiple sources. Summary of the Invention

[0003] This invention provides a method, apparatus, device, and storage medium for adaptive mind map generation, in order to solve the technical problem that existing technologies cannot support the rapid and accurate generation of mind maps from all types of mind map data.

[0004] According to one aspect of the present invention, an adaptive mind map generation method is provided, comprising:

[0005] In response to a user's mind map generation request, identify the mind map data corresponding to the mind map generation request;

[0006] The mind map data is routed using a routing decision engine to determine the mind map generation path corresponding to the mind map data.

[0007] If the mind map generation path is a parallel generation path, then the mind map data is decomposed and merged in parallel to determine the target mind map.

[0008] According to another aspect of the present invention, an adaptive mind map generation device is provided, comprising:

[0009] The request and response module is used to respond to the mind map generation request submitted by the user and identify the mind map data corresponding to the mind map generation request.

[0010] The hierarchical routing module is used to make routing decisions on the mind map data based on the routing decision engine, and determine the mind map generation path corresponding to the mind map data.

[0011] The parallel generation module is used to perform parallel task decomposition and merging on the mind map data to determine the target mind map if the mind map generation path is a parallel generation path.

[0012] According to another aspect of the present invention, an electronic device is provided, the electronic device comprising:

[0013] At least one processor; and

[0014] A memory communicatively connected to the at least one processor; wherein,

[0015] The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to execute the mind map adaptive generation method according to any embodiment of the present invention.

[0016] According to another aspect of the present invention, a computer-readable storage medium is provided, the computer-readable storage medium storing computer instructions for causing a processor to execute and implement the mind map adaptive generation method according to any embodiment of the present invention.

[0017] The technical solution of this invention is adaptable to input data from multiple sources and of all types in large language models. Through a hierarchical routing decision mechanism, it can select the optimal generation path, fundamentally improving the generation stability and consistency in multi-source scenarios. For document data of different lengths, it can reduce token usage while preserving the document's semantic structure to the maximum extent. Through parallel generation mechanisms and scheduling, it can stably generate complete mind maps that meet structural specifications in multi-source scenarios, balancing parallel efficiency and result consistency. This solves the technical problem of existing technologies being unable to support the rapid and accurate generation of mind maps from all types of mind map data. It ensures that the mind map results always meet the beneficial effects of optimal visualization rendering and user reading experience.

[0018] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description. Attached Figure Description

[0019] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are 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.

[0020] Figure 1 A flowchart of a mind map adaptive generation method is provided in this embodiment of the invention;

[0021] Figure 2 A flowchart of another adaptive mind map generation method provided in an embodiment of the present invention;

[0022] Figure 3 A flowchart of another adaptive mind map generation method provided in an embodiment of the present invention;

[0023] Figure 4 This is a schematic diagram of the structure of a mind map adaptive generation device provided in an embodiment of the present invention;

[0024] Figure 5 A schematic diagram of the structure of an electronic device 10 that can be used to implement an embodiment of the present invention is shown. Detailed Implementation

[0025] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0026] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0027] Figure 1 This invention provides a flowchart of an adaptive mind map generation method. This embodiment is applicable to scenarios where mind maps are automatically generated using large language models, supporting multiple sources, analyzing and processing input data of different data types, and quickly generating mind maps. This method can be executed by an adaptive mind map generation device, which can be implemented in hardware and / or software and can be configured in an electronic device. Figure 1 As shown, the method includes:

[0028] S110. Respond to the mind map generation request submitted by the user and identify the mind map data corresponding to the mind map generation request.

[0029] Optionally, a mind map generation request can be a request from a user, when using any large language model, to submit at least one piece of data to the large language model, requesting the large language model to generate a mind map. For example, a user operates the front-end interface of the large language model, selecting three PDFs, two webpage excerpts, and a meeting minutes, and requests the large language model to generate a mind map. This mind map generation request includes the data: the three PDFs, two webpage excerpts, and the meeting minutes.

[0030] Optionally, mind map data can be multidimensional feature data obtained by extracting multidimensional content features from data from different sources and at least one data type submitted by users to a large language model. It should be noted that mind map data consists of multiple dimensions of feature data, namely: source quantity feature, file quantity feature, content length feature, word size feature, structural feature, and source type feature.

[0031] Optionally, in this invention, when a user submits a mind map generation request, they can submit a list of source identifiers in the mind map generation request, perform batch parsing of the source identifier list, obtain metadata for each source from the database, and then extract multidimensional feature data for each source based on the metadata of each data source to obtain mind map data.

[0032] Optionally, each data source includes several basic attributes: source type, file list, content, and summary. The source type specifically includes: knowledge base, uploaded file, pasted text, webpage, and video. The file list only applies to knowledge base sources, i.e., the list of knowledge base files associated with the knowledge base, with each file containing a file identifier, filename, MIME type, and content. The content represents the content string directly carried when using a single file source. The summary is a pre-generated summary content used to replace the data submitted in the mind map generation request as input for mind map generation.

[0033] Optionally, for data originating from a knowledge base, this invention prioritizes using completed summary content (asynchronously generated and cached via a summary service) to reduce token consumption for LLM calls. If the summary is not yet ready, the system uses an asynchronous waiting mechanism with an exponential backoff strategy to poll the summary status; the maximum waiting time is configurable. If it is still not ready after a timeout, it falls back to using a mind map to generate the data to be submitted.

[0034] Optionally, in this invention, the source quantity characteristic of mind map data is obtained by counting the length of the source identifier list. When the source quantity characteristic is greater than 1, the mind map data is a multi-source data set.

[0035] The file quantity feature is used to count the sources of knowledge base types and the number of associated files. When the number of associated files for a single source is greater than 1, it is a multi-file scenario.

[0036] The content length feature is calculated in units of characters to determine the total length of the content. When the total length of the content exceeds a preset character threshold, content truncation is required, triggering a content truncation strategy to cut the content within the preset character threshold. For example, the preset character threshold is typically set to 50,000 characters, and paragraph boundaries are prioritized as cutting positions near the truncation point to avoid truncation in the middle of sentences.

[0037] The lexical size feature is used to characterize the total number of lexical units in the content. A lightweight bilingual token estimation method is typically used to quickly estimate the total number of lexical units. If the estimated lexical size feature exceeds a preset lexical threshold, the content needs to be compressed to a safe lexical range before mind map generation. For example, a typical preset lexical threshold value is 80,000, and the lightweight bilingual token estimation method is as follows: CJK characters are estimated at 1.5 tokens / character; ASCII letters and numbers are estimated at 0.25 tokens / character; and other Unicode characters are estimated at 0.5 tokens / character.

[0038] Structural features involve scanning the processed summary content for structural information using regular expressions to detect the presence of section break tags. Since the summary service takes precedence over routing decisions, this feature can only be detected after the summary is ready, eliminating circular dependencies. Content with such tags is considered to have an explicit section structure and can be processed in parallel at the section level. The section break tags are automatically inserted during summary preprocessing when generating section summaries from very long documents; they are a system-defined structured format, not something that relies on manual user input.

[0039] Source type features are used to characterize the source of data, and are obtained by identifying the type enumeration value for each source.

[0040] Specifically, in response to a user's submitted mind map generation request, the system identifies the mind map data corresponding to the mind map generation request.

[0041] S120. Based on the routing decision engine, perform routing decisions on the mind map data to determine the mind map generation path corresponding to the mind map data.

[0042] Optionally, the mind map generation path can be to process the mind map data into prompts for a large language model, and then generate the mind map using the large language model. It should be noted that, in this invention, the mind map generation path includes a direct generation path, a parallel generation path, and a preprocessing path.

[0043] Optionally, the direct generation path can be understood as the large language model being able to directly use mind map data as the data basis for mind maps; the parallel generation path is used to generate mind maps from multiple sources and / or multiple files; the preprocessing path can be used when the mind map data exceeds the processing range of the large language model, requiring data processing of the mind map data before generating mind maps based on the processed mind map data.

[0044] Optionally, the routing decision engine can be a pre-configured functional module for routing decisions on mind map data. It should be noted that the routing decision engine makes routing decisions based on multi-dimensional feature data in the mind map data, matching a corresponding mind map generation path for each mind map data. For example, the routing decision engine extracts the source quantity feature of the mind map data. If the source quantity feature indicates multiple sources or multiple files within a single source, the mind map data is matched as a parallel generation path. If the source quantity feature indicates a single data source, and the word size feature of the mind map data exceeds a preset word threshold, the mind map data is matched as a preprocessing path. If the source quantity feature indicates a single data source, but the content length feature of the mind map data exceeds a preset character threshold, the content will be truncated, and the mind map data will be matched as a direct generation path. If the routing decision engine does not meet any of the above conditions, the mind map data is directly matched as a direct generation path.

[0045] Specifically, the routing decision engine performs routing decisions on the mind map data to determine the mind map generation path corresponding to the mind map data.

[0046] S130. If the mind map generation path is a parallel generation path, then the mind map data is decomposed and merged in parallel to determine the target mind map.

[0047] Optionally, the target mind map can be generated by a large language model using mind map data to generate a mind map based on a mind map generation request.

[0048] Optionally, if the mind map generation path of the mind map data is a parallel generation path, then the mind map data is decomposed and merged into parallel tasks based on the adaptive parallel generation scheduler corresponding to the parallel generation path to determine the target mind map. The adaptive parallel generation scheduler can be a functional module that performs adaptive parallel processing on the mind map data.

[0049] Specifically, if the mind map generation path is a parallel generation path, then the mind map data is decomposed and merged in parallel to determine the target mind map.

[0050] The technical solution of this invention is adaptable to input data from multiple sources and of all types in large language models. Through a hierarchical routing decision mechanism, it can select the optimal generation path, fundamentally improving the generation stability and consistency in multi-source scenarios. For document data of different lengths, it can reduce token usage while preserving the document's semantic structure to the maximum extent. Through parallel generation mechanisms and scheduling, it can stably generate complete mind maps that meet structural specifications in multi-source scenarios, balancing parallel efficiency and result consistency. This solves the technical problem of existing technologies being unable to support the rapid and accurate generation of mind maps from all types of mind map data. It ensures that the mind map results always meet the beneficial effects of optimal visualization rendering and user reading experience.

[0051] Figure 2 This is a flowchart illustrating another adaptive mind map generation method provided by an embodiment of the present invention. The relationship between this embodiment and the above embodiments is that this embodiment specifically describes the process of processing mind map data through parallel generation paths to generate a target mind map. Figure 2 As shown, the method includes:

[0052] S210. Respond to the mind map generation request submitted by the user and identify the mind map data corresponding to the mind map generation request.

[0053] S220. Based on the routing decision engine, perform routing decisions on the mind map data to determine the mind map generation path corresponding to the mind map data.

[0054] S230. If the mind map generation path is a parallel generation path, then the mind map data is adaptively generated in parallel to determine the set of parallel generation tasks.

[0055] The set of parallel generation tasks includes at least one parallel generation task; the set of parallel generation tasks is a data set composed of parallel generation tasks. A parallel generation task can be a unit task that processes mind map data using a parallel generation path; it should be noted that a parallel generation task is used to generate a local subtree.

[0056] Optionally, the mind map data is adaptively and parallelly generated using an adaptive parallel generation scheduler. During the processing of the adaptive parallel generation scheduler, the mind map data is decomposed at the file level and / or at the section level to obtain the parallel generation tasks corresponding to the mind map data and construct a set of parallel generation tasks.

[0057] Optionally, the total number of tasks in the parallel generation task set is equal to the set of task numbers for all parallel generation tasks; if the mind map data contains files with a segmented structure, the total number of tasks can be the product of the number of files from all sources and the number of chapters; if the mind map data does not contain files with a segmented structure, the total number of tasks is the sum of the number of source files.

[0058] Specifically, if the mind map generation path is a parallel generation path, then the mind map data is adaptively generated in parallel to determine the set of parallel generation tasks.

[0059] Optionally, in this invention, the adaptive parallel generation of the mind map data and the determination of the set of parallel generation tasks include:

[0060] For each data source in the mind map data, if there are multiple associated files in the associated file list of the data source, a file-level task is set for each associated file; for each file-level task, if the file-level task has a segmented structure, the file-level task is decomposed into a set of segmented subtasks based on the segmented structure; the parallel generation task set is constructed based on all the set of segmented subtasks and all the file-level tasks.

[0061] In this context, associated files can be files existing in the associated file list; file-level tasks can be subtasks obtained by the adaptive parallel generation scheduler decomposing mind map data at the file-level decomposition granularity. It should be noted that file-level tasks include three attributes: filename, content, and task type; the filename is used as a candidate label for the subtree root node, and the task type is typically file-by-file generation; the content of the file-level task is either the content of the file itself or a preprocessed summary.

[0062] Optionally, the segmentation structure can be a set of chapter / segmentation markers for the file content of a file-level task. The set of segmented subtasks can be a set of subtasks obtained by an adaptive parallel generation scheduler decomposing the content of a file-level task at the segmentation-level decomposition granularity.

[0063] Optionally, if the adaptive parallel generation scheduler detects that a file-level task has a segmented structure, it will split the content of the file-level task into multiple segmented subtasks, i.e., a set of segmented subtasks, according to chapter segmentation markers. The decomposition process is as follows: the adaptive parallel generation scheduler uses regular expressions to extract all chapter segmentation markers, obtaining a list of chapter titles and corresponding content boundaries; an independent generation task is created for each chapter, resulting in segmented subtasks; the original file-level task is replaced with the set of subtasks. The segmented subtasks include chapter titles and chapter content.

[0064] Specifically, for each data source in the mind map data, if there are multiple associated files in the associated file list of the data source, a file-level task is set for each associated file; for each file-level task, if the file-level task has a segmented structure, the file-level task is decomposed into a set of segmented subtasks based on the segmented structure; and a set of parallel generation tasks is constructed based on the complete set of segmented subtasks and the complete set of file-level tasks.

[0065] Optionally, in this invention, the method further includes: for each parallel generation task in the set of parallel generation tasks, performing a routing decision on the parallel generation task based on the routing decision engine to determine the large model input data corresponding to each parallel generation task.

[0066] The input data for the large model can be the text content used to input into the large language model to generate mind maps. It should be noted that the input data, as the content data input into the large language model, is combined with the prompt word template to assemble complete contextual data for processing by the large language model. For example, the prompt word template includes at least the task role, the input data for the large model, the output format, and the output requirements.

[0067] Optionally, the large language model can be a pre-built agent. A pre-trained large language model is obtained by training massive amounts of text as training data. Based on this pre-trained large language model, a mind map generation agent and a mind map merging agent are built. The large language model can be at least one of the mind map generation agent and the mind map merging agent.

[0068] Optionally, large model input data can be generated based on the content fields of each parallel generation task, and the large model input data can be input into the mind map generation agent to obtain the target mind map.

[0069] Optionally, an adaptive parallel generation scheduler can be used to decompose multiple sources or multiple files into individual parallel generation tasks. The routing decision engine can independently perform scale detection on each parallel generation task to prevent the content fields of the parallel generation tasks from having excessively long content that far exceeds the upper limit of the LLM single context window, thus posing a risk of context overflow.

[0070] Optionally, the routing decision engine can perform scale detection independently for each parallel generation task, or it can perform routing decisions on the mind map data in the same way, that is, it can use the routing decision engine again to perform routing decisions on each parallel generation task.

[0071] Specifically, for each parallel generation task in the set of parallel generation tasks, a routing decision engine is used to make routing decisions for the parallel generation tasks to determine the large model input data corresponding to each parallel generation task.

[0072] Optionally, in another optional embodiment of the present invention, the step of performing routing decisions on the parallel generation tasks based on the routing decision engine to determine the large model input data corresponding to each parallel generation task includes: If the word size feature of the content field of the parallel generation task is greater than the preset word threshold, the content field of the parallel generation task is compressed into a mind map concise summary based on the three-modal preprocessor, and the mind map concise summary is determined as the input data of the large model. If the word size feature of the content field of the parallel generation task is not greater than the preset word threshold, and the content length feature of the content field of the parallel generation task is greater than the preset character threshold, then paragraph boundary truncation processing is performed on the content field of the parallel generation task to obtain a refined summary of the mind map, and the refined summary of the mind map is determined as the input data of the large model. If the lexical size feature of the content field of the parallel generation task is not greater than a preset lexical threshold, and if the content length feature of the content field of the parallel generation task is not greater than a preset character threshold, then the content field of the parallel generation task is determined as the input data of the large model.

[0073] Optionally, the routing decision engine of this invention makes routing decisions for each parallel generation task as follows: If the lexical size feature of the content field of the parallel generation task is greater than a preset lexical threshold, the content field of the parallel generation task is compressed into a mind map concise summary by a trimodal preprocessor, and the mind map concise summary is used as the content field of the parallel generation task; if the content length feature of the content field of the parallel generation task is greater than a preset character threshold, and the lexical size feature is not greater than the preset lexical threshold, paragraph boundary truncation is performed on the content field of the parallel generation task to obtain the mind map concise summary, which is used as the processed content field; otherwise, the content field of the parallel generation task is not processed. The mind map concise summary is a structured summary text used as input to the large language model; the trimodal preprocessor is a pre-built trimodal Map-Reduce preprocessor. This mechanism ensures that the processing logic for excessively long content is completely consistent between the parallel path and the direct path—regardless of whether the content comes from multi-source aggregation or single-source direct generation, each LLM call is subject to the same scale safety guarantee before actual triggering, eliminating the risk of LLM context overflow caused by excessively long factor task content.

[0074] Specifically, if the lexical size of the content field in the mind map adaptive generation parallel generation task exceeds a preset lexical threshold, the content field is compressed into a refined mind map summary using a three-modal preprocessor. This refined summary is then used as input data for the large-scale mind map adaptive generation model. If the lexical size of the content field does not exceed the preset lexical threshold, but the content length exceeds a preset character threshold, paragraph truncation is performed to obtain a refined mind map summary, which is then used as input data for the large-scale mind map adaptive generation model. Finally, if the lexical size of the content field does not exceed the preset lexical threshold, and the content length exceeds the preset character threshold, the content field is used as input data for the large-scale mind map adaptive generation model.

[0075] S240. Merge each of the parallel generation tasks in the set of parallel generation tasks to determine the target mind map.

[0076] Optionally, in this invention, each parallel generation task generates a local subtree, and the complete target mind map is obtained by merging these local subtrees. The merging process can utilize a programmatic hierarchical tree merging algorithm, eliminating the need for additional LLM calls.

[0077] Specifically, the parallel generation tasks of the parallel generation task set are merged to determine the target mind map.

[0078] Optionally, in another optional embodiment of the present invention, merging the parallel generation tasks of the parallel generation task set to determine the target mind map includes:

[0079] The mind map generation agent uses an asynchronous concurrent scheduler based on semaphores to generate local subtrees for a set of parallel tasks, determining the local subtree corresponding to each parallel task; and then merges the local subtrees based on a programmatic hierarchical tree merging algorithm to determine the target mind map.

[0080] Optionally, a local subtree can be understood as a local mind map output by the mind map generating agent based on parallel generation tasks, and the local subtree is a JSON file.

[0081] Optionally, the mind map generation agent can be a pre-configured agent for mind map generation based on a large language model. It should be noted that, under the parallel generation path, the prompt word project of the mind map generation agent consists of a pre-defined tree structure specification and the content fields of the parallel generation task. The agent's role is also defined as generating mind maps, with the output format being JSON.

[0082] Optionally, the preset tree structure specification can be any predefined tree structure constraint parameters. For example, the preset tree structure specification is represented by Table 1, as shown in Table 1:

[0083] Table 1

[0084]

[0085] Optionally, the asynchronous concurrent scheduler can be a pre-configured module for asynchronous concurrent scheduling of a set of parallel tasks. It should be noted that the asynchronous concurrent scheduler is designed with a maximum number of concurrent slots, which defaults to 64 and can be specified through configuration options. This controls the number of parallel generation tasks that simultaneously hold semaphores and initiate LLM calls. Each parallel generation task has a defined execution flow: after acquiring a semaphore, each task independently initiates an LLM call, schedules the mind map generation agent to generate a local subtree, and releases the semaphore upon completion. Multiple tasks alternately occupy concurrent slots in a parallel queuing and upper-slot execution manner. For all parallel generation tasks, semantic concurrent aggregation is used, allowing all tasks to wait but permitting partial failures. The failure of any single task does not interrupt the execution of other tasks. A fault tolerance mechanism is also set up: for failed parallel generation tasks, error logs are recorded, and the local subtree of the failed parallel generation task is marked as empty. The success rate of parallel task generation is calculated out of the total number of tasks. If the success rate is lower than a preset tolerance rate for the total number of tasks, the local subtree generation of the current parallel task set is considered a failure. If the success rate is not lower than the preset tolerance rate for the total number of tasks, the failed parallel task generation is recorded, and the local subtrees corresponding to the successful parallel task generation are retained. The maximum concurrency of the semaphore is the maximum number of concurrent slots.

[0086] Optionally, an asynchronous concurrent scheduler schedules a mind map generation agent, passing in the data source and content field Prompt of the parallel generation task, combining it with tree structure reduction to generate mind maps, and outputting local subtree JSON.

[0087] Optionally, the programmatic hierarchical tree merging algorithm merges the various local subtrees to determine the target mind map, including:

[0088] The local subtrees are merged using a programmatic hierarchical tree merging algorithm to determine the programmatic merging result. If the direct child nodes of the programmatic merging result are greater than a preset child node count constraint, a mind map merging agent is invoked to merge the programmatic merging result and determine the target mind map. If the direct child nodes of the programmatic merging result are not greater than the preset child node count constraint, the programmatic merging result is determined as the target mind map.

[0089] Optionally, the programmatic merging result is a complete tree structure JSON file obtained after merging all the local subtrees. The programmatic merging result is obtained by a programmatic hierarchical tree merging algorithm. The specific steps of the programmatic hierarchical tree merging algorithm are as follows: Step 1: Root node generation: Identify the data source of each local subtree. If the data source of a local subtree is only one file, the root node label after merging is the file name or data source name; if the data sources of each local subtree are multiple, the root node label after merging is the data source name or the mind map name specified by the user; Step 2: Subtree access: For each local subtree, the root node of the local subtree is accessed as a direct child node of the mind map. If the local subtree corresponds to a segmented subtask, the file subtask corresponding to the local subtree is accessed as a file child node of the mind map, and the local subtree connects to the file child nodes. The third step is to rename the global node identifiers. If there are node identifier conflicts in local subtrees, all child nodes are traversed in a depth-first manner, and the node identifiers are reassigned as globally unique identifiers. The identifier fields inside all nodes are also updated. The fourth step is to impose structural normative constraints. After merging, three-dimensional constraints are applied to the final tree structure: Depth constraint: Maximum depth of 10 levels. Nodes exceeding this depth are truncated, and their child nodes are promoted to the parent node level; Number of nodes constraint: Maximum total number of nodes of 10,000. If this number is exceeded, nodes at higher levels, i.e., those closer to the root node, are retained first; Number of child nodes constraint: Each node can have a maximum of 12 direct child nodes generated by the agent or 8 child nodes parsed by Markdown. If this number is exceeded, semantically similar child nodes are merged to obtain a programmatic merging result.

[0090] Optionally, the mind map merging agent can be a pre-set agent based on a large language model for merging mind map nodes. It should be noted that the prompt word project of the mind map generating agent mainly consists of a preset tree structure specification and a programmatic merging result. It also specifies that the agent's role is to merge mind maps and the output format is a JSON mind map.

[0091] Optionally, the mind map merging agent takes the programmatic merging result as input, performs a more natural merging of multiple subtrees based on semantics, and uses a more relaxed tree structure reduction during the merging process to output the target mind map. It should be noted that the mind map output by the mind map merging agent is a more compact, summary-type mind map.

[0092] Specifically, the programmatic hierarchical tree merging algorithm is used to merge the various local subtrees to determine the programmatic merging result. If the direct child nodes of the programmatic merging result are greater than a preset child node count constraint, the mind map merging agent is invoked to merge the programmatic merging result and determine the target mind map. If the direct child nodes of the programmatic merging result are not greater than the preset child node count constraint, the programmatic merging result is determined as the target mind map.

[0093] The technical solution of this invention is adaptable to input data from multiple sources and of all types in large language models. Through a hierarchical routing decision mechanism, it can select the optimal generation path, fundamentally improving the generation stability and consistency in multi-source scenarios. For document data of different lengths, it can reduce token usage while preserving the document's semantic structure to the maximum extent. Through parallel generation mechanisms and scheduling, it can stably generate complete mind maps that meet structural specifications in multi-source scenarios, balancing parallel efficiency and result consistency. This solves the technical problem of existing technologies being unable to support the rapid and accurate generation of mind maps from all types of mind map data. It ensures that the mind map results always meet the beneficial effects of optimal visualization rendering and user reading experience.

[0094] Figure 3 This is a flowchart illustrating another adaptive mind map generation method provided by an embodiment of the present invention. The relationship between this embodiment and the above embodiments is that this embodiment specifically describes the process of preprocessing mind map data to generate the target mind map. For example... Figure 3 As shown, the method includes:

[0095] S310. Respond to the mind map generation request submitted by the user and identify the mind map data corresponding to the mind map generation request.

[0096] S320. Based on the routing decision engine, perform routing decisions on the mind map data to determine the mind map generation path corresponding to the mind map data.

[0097] S330. If the mind map generation path is a preprocessing path, then the mind map data is processed by a three-modal preprocessor to determine a refined summary of the mind map.

[0098] The refined summary of the mind map can be a high-fidelity intermediate text obtained by processing the mind map data. It should be noted that the refined summary is obtained by compressing the mind map data, retaining the key content of the mind map data, and achieving a high-fidelity replacement of the mind map data.

[0099] Specifically, if the mind map generation path is a preprocessing path, then the mind map data is processed by a three-modal preprocessor to determine at least one refined summary of the mind map.

[0100] Optionally, before processing the mind map data, the trimodal preprocessor needs to perform result probing on the mind map data, identify the block structure availability attribute and entity density attribute of the mind map data, and then perform further data processing on the mind map data based on the block structure availability attribute and entity density attribute.

[0101] Optionally, the block structure availability attribute is the chunked data of the mind map data. The process of identifying the block structure availability attribute of the mind map data is as follows: First, identify whether the mind map data is a knowledge base document that has been indexed by the knowledge base. If it is not a knowledge base document that has been indexed by the knowledge base, it does not have the block structure availability attribute. If it is a knowledge base document that has been indexed by the knowledge base, then further obtain the list of document chunks stored in the knowledge base as the block structure availability attribute.

[0102] Optionally, the entity density attribute is the density value of named entities in the mind map data. The method for identifying the entity density attribute of mind map data is as follows: quickly scan the mind map data and count the density of named entities; if the density of named entities exceeds a preset threshold, the entity density attribute is considered entity-dense content; if the density of named entities does not exceed the preset threshold, it does not have an entity density attribute. For example, the density of named entities is calculated as the number of named entities per thousand characters; the preset threshold is 5 entities per thousand characters.

[0103] Optionally, in this invention, the step of processing the mind map data using a three-modal preprocessor to determine at least one refined mind map summary includes:

[0104] If the mind map data is a knowledge base document and there is a document block list, then the mind map data is processed based on the hierarchical perception model to determine the refined summary of the mind map.

[0105] If the mind map data is not a knowledge base document, and the entity density of the mind map data is greater than a preset threshold, then the mind map data is processed based on the entity perception mode to determine the refined summary of the mind map.

[0106] If the mind map data is not a knowledge base document, and the entity density of the mind map data is not greater than a preset threshold, then the mind map data is processed based on a flat pattern to determine the refined summary of the mind map.

[0107] Optionally, in this invention, the hierarchical awareness mode is named the hierarchical awareness Map-Reduce mode. This mode is used to process mind map data that is a knowledge base document and has a document chunk list. The hierarchical awareness mode utilizes the inherent hierarchical structure of the document for hierarchical summarization, preserving the logical organization of the document to the greatest extent possible. The processing of the hierarchical awareness mode consists of a hierarchical grouping stage, a hierarchical Map stage, and a hierarchical Reduce stage; each stage is specifically as follows:

[0108] Hierarchical grouping stage: The document chunk list of the knowledge base documents is grouped based on the top-level path of the hierarchical path to form the main branches of the document hierarchy tree. For example, the chunks of a technical white paper might be grouped as follows:

[0109] ["Chapter 1: Background Introduction"] contains all sub-blocks under this chapter; ["Chapter 2: Technical Solution"] contains all sub-blocks under this chapter; and so on.

[0110] Hierarchical Mapping Phase: Within each hierarchical group, blocks are arranged sequentially based on their block numbers in the document chunk list. Semantic boundary segmentation is used to further segment hierarchical groups exceeding the lexical limit. LLM is invoked in parallel for multiple batches of content split within each hierarchical group or each hierarchical group, generating a structured summary for that group. The summary retains the hierarchical path information of that group as metadata tags.

[0111] Hierarchical Reduce Phase: Utilizing the parent block ID relationship of each hierarchical group, recursively reduce upwards from the leaf node level: 1. First, summarize the summaries of all child hierarchical groups under the same parent block; call LLM to generate the parent summary, and explicitly include the summaries and hierarchical paths of the child hierarchical groups in the Prompt; recursively move upwards to the root node to generate the final summary, which is the refined summary of the mind map data.

[0112] Optionally, in this invention, the flat mode is named the flat Map-Reduce mode. The flat mode is used to process plain text without internal structure, i.e., mind map data without internal structure. For example, mind map data without internal structure can be articles and web page text pasted into an LLM file.

[0113] Optionally, the flattened mode consists of a text segmentation stage, a parallel summarization stage, and a reduction and merging stage, with each stage specifically as follows:

[0114] Text Segmentation Stage: The segmenter in the text segmentation stage uses bilingual token estimation to perform semantic boundary-prioritized paragraph segmentation, resulting in text segments. The segmentation method is as follows: 1. Prioritize identifying paragraph boundaries based on blank lines; 2. If the number of tokens in a single batch still exceeds the limit (default 20,000 tokens), backtrack to the newline boundary; 3. Further backtrack to the space boundary for hard segmentation; 4. Maintain 100-200 tokens of content overlap between adjacent segments to preserve semantic coherence. Content overlap is understood as taking the last paragraph of the previous segment as the context prefix of the next segment.

[0115] Parallel Digest Phase: The asynchronous concurrent scheduler is invoked to perform LLM (Limited Least Metric) calls in parallel on text segments based on semaphores. Specifically, in the asynchronous concurrent scheduler, an LLM call is initiated independently for each text segment after acquiring a semaphore to perform digest extraction and generate a segment digest for each text segment. The asynchronous concurrent scheduler also employs a fault-tolerance mechanism for text segments. The success rate of successfully acquired text segments is considered. If this success rate is lower than a preset tolerance rate for the total number of text segments, the entire MapReduce process is terminated and an error is reported. If the success rate is not lower than the preset tolerance rate, the segment digests corresponding to the successful text segments are retained.

[0116] Reduce and merge phase: The LLM algorithm is invoked to concatenate all segmented summaries. If the total length after concatenation still exceeds a preset character threshold, a new round of flat MapReduce is recursively executed. If the total length after concatenation does not exceed the preset character threshold, the concatenated full-text execution summary is output as the refined summary of the mind map data. The maximum recursion depth is 3 levels; the full-text execution summary is the summary synthesized in each single LLM call.

[0117] Optionally, the entity-aware mode in this invention is named the Entity-Aware Map-Reduce mode, which is used to process entity-intensive content. For example, entity-intensive content can be biographies, corporate reports, and academic literature.

[0118] Optionally, the entity-aware model includes a text segmentation stage, a parallel summarization stage, an entity deduplication and fusion stage, and a reduction and merging stage; each stage is specifically as follows:

[0119] Text segmentation stage: The segmenter in the text segmentation stage uses bilingual token estimation to perform semantic boundary-priority paragraph segmentation of the text, resulting in text segments; the segmentation method is as follows: 1. Identify paragraph boundaries first by blank lines; 2. If the number of tokens in a single batch still exceeds the limit (default 20,000 tokens) within a paragraph, backtrack to the newline boundary; 3. Further backtrack to the space boundary and perform hard segmentation; 4. Preserve 100-200 tokens of content overlap between adjacent segments to maintain semantic coherence.

[0120] Parallel Summarizing Phase: The parallel summarizing phase in the entity-aware mode is an enhanced parallel summarizing phase. It invokes an asynchronous concurrent scheduler, using semaphores to perform parallel LLM calls on text segments to execute summary extraction and named entity extraction. Within the asynchronous concurrent scheduler, each text segment independently initiates an LLM call after acquiring a semaphore to perform summary extraction and named entity extraction, generating a segment summary and named entity for each text segment. The asynchronous concurrent scheduler also employs a fault-tolerant mechanism for summarizing and extracting text segments. The success rate of successfully extracted text segments out of the total number of tasks is considered. If this success rate is lower than a preset tolerance rate for the total number of text segments, the entire entity-aware MapReduce process is terminated and an error is reported. If the success rate is not lower than the preset tolerance rate, the segment summaries and named entities corresponding to the successful text segments are retained. Named entities include entity text, entity type, context fragment of occurrence, and importance score in the range of 0-1; entity types include the following types: PERSON: person's name; ORGANIZATION: organization / institution name; DATE_TIME: time node; TECH_TERM: technical term / product name; LOCATION: location; CONCEPT: core concept.

[0121] Entity deduplication and fusion stage: Global deduplication and fusion are performed on all named entities extracted from all text segments (i.e., the named entity list). An entity summary table is then constructed based on this globally deduplicated and fused list. Specifically: The text similarity of each named entity in the named entity list for each text segment is calculated. Named entities with high text similarity are deduplicated and fused to obtain a globally deduplicated and fused list of named entities. These entities are then sorted based on their importance score, and the top 50 most important named entities are selected to construct the entity summary table. The entity summary table is structured text, listing the name, entity type, and context of each named entity.

[0122] Reduce and merge phase: The LLM is invoked to concatenate all segmented summaries and append an entity summary table. The LLM is instructed to focus on named entities and relationships in the entity summary table when generating the final summary. If the total length after concatenation still exceeds a preset character threshold, a new round of flat Map-Reduce mode is recursively executed. If the total length after concatenation does not exceed the preset character threshold, the concatenated full-text execution summary is output as the refined summary of the mind map data. The maximum recursion depth is 3 levels; the full-text execution summary is the summary synthesized in each single LLM call.

[0123] Specifically, when the mind map data is a knowledge base document and there is a document block list, the mind map data is processed based on the hierarchical perception model to determine the concise summary of the mind map.

[0124] If the mind map data is not a knowledge base document, and the entity density of the mind map data exceeds a preset threshold, then data processing is performed based on an entity-aware model to determine a refined mind map summary. If the mind map data is not a knowledge base document, and the entity density of the mind map data does not exceed a preset threshold, then data processing is performed based on a flattening model to determine a refined mind map summary. The refined mind map summary retains the hierarchical information of the document, providing a natural hierarchical input for subsequent mind map generation, which helps LLM generate mind maps that correspond to the original document's organizational structure.

[0125] S340. The mind map generating agent generates a mind map based on the refined summary of the mind map, and determines the target mind map.

[0126] Optionally, in the preprocessing path, the prompt word project for the mind map generation agent consists of a preset tree structure specification and a refined summary of the mind map, defining the agent's role as generating mind maps and outputting mind maps in JSON format.

[0127] Specifically, the mind map generation agent generates mind maps based on the refined summary of the mind map, and determines the target mind map.

[0128] Optionally, in this invention, if the mind map generation path is a direct processing path, then the mind map generation agent generates the mind map based on the mind map data to determine the target mind map.

[0129] Optionally, in the direct processing path, the prompt word project of the mind map generation agent consists of a preset tree structure specification and mind map data, specifying the agent's role as generating mind maps and outputting mind maps in JSON format.

[0130] Specifically, if the mind map generation path is a direct processing path, then the mind map generation agent generates the mind map based on the mind map data and determines the target mind map.

[0131] The technical solution of this invention is adaptable to input data from multiple sources and of all types in large language models. Through a hierarchical routing decision mechanism, it can select the optimal generation path, fundamentally improving the generation stability and consistency in multi-source scenarios. For document data of different lengths, it automatically selects from three preprocessing modes—flat mode, entity-aware mode, and hierarchical-aware mode—based on the availability of content block structure and entity density. This compresses ultra-long documents to a vocabulary range that the large language model can process before generating mind maps, solving the technical problem of existing technologies that cannot support the rapid and accurate generation of mind maps from all types of data. This ensures that the mind map results always meet the beneficial effects of optimal visualization rendering and user reading experience.

[0132] Figure 4 This is a schematic diagram of the structure of a mind map adaptive generation device provided in an embodiment of the present invention. Figure 4 As shown, the device includes: a request-response module 410, a hierarchical routing module 420, and a parallel generation module 430; wherein,

[0133] The request response module 410 is used to respond to the mind map generation request submitted by the user and identify the mind map data corresponding to the mind map generation request.

[0134] The hierarchical routing module 420 is used to make routing decisions on the mind map data based on the routing decision engine, and determine the mind map generation path corresponding to the mind map data;

[0135] The parallel generation module 430 is used to perform parallel task decomposition and merging on the mind map data to determine the target mind map if the mind map generation path is a parallel generation path.

[0136] The technical solution of this invention is adaptable to input data from multiple sources and of all types in large language models. Through a hierarchical routing decision mechanism, it can select the optimal generation path, fundamentally improving the generation stability and consistency in multi-source scenarios. For document data of different lengths, it can reduce token usage while preserving the document's semantic structure to the maximum extent. Through parallel generation mechanisms and scheduling, it can stably generate complete mind maps that meet structural specifications in multi-source scenarios, balancing parallel efficiency and result consistency. This solves the technical problem of existing technologies being unable to support the rapid and accurate generation of mind maps from all types of mind map data. It ensures that the mind map results always meet the beneficial effects of optimal visualization rendering and user reading experience.

[0137] Optionally, the parallel generation module 430 is specifically used for:

[0138] The mind map data is adaptively generated in parallel to determine a set of parallel generation tasks; wherein the set of parallel generation tasks includes at least one parallel generation task.

[0139] The parallel generation tasks in the set of parallel generation tasks are merged to determine the target mind map.

[0140] Optionally, the parallel generation module 430 is also specifically used for:

[0141] For each data source in the mind map data, if there are multiple associated files in the associated file list of the data source, then set a file-level task for each associated file;

[0142] For each file-level task, if the file-level task has a segmentation structure, then the file-level task is decomposed into a set of segmented subtasks based on the segmentation structure.

[0143] The parallel generation task set is constructed based on the entire set of segmented subtasks and the entire set of file-level tasks.

[0144] Optionally, the parallel generation module 430 is also specifically used for:

[0145] For each parallel generation task in the set of parallel generation tasks, a routing decision is made based on the routing decision engine to determine the large model input data corresponding to each parallel generation task.

[0146] Optionally, the parallel generation module 430 is also specifically used for: If the word size feature of the content field of the parallel generation task is greater than the preset word threshold, the content field of the parallel generation task is compressed into a mind map concise summary based on the three-modal preprocessor, and the mind map concise summary is determined as the input data of the large model. If the word size feature of the content field of the parallel generation task is not greater than the preset word threshold, and the content length feature of the content field of the parallel generation task is greater than the preset character threshold, then paragraph boundary truncation processing is performed on the content field of the parallel generation task to obtain a refined summary of the mind map, and the refined summary of the mind map is determined as the input data of the large model. If the lexical size feature of the content field of the parallel generation task is not greater than a preset lexical threshold, and if the content length feature of the content field of the parallel generation task is not greater than a preset character threshold, then the content field of the parallel generation task is determined as the input data of the large model.

[0147] Optionally, the parallel generation module 430 is also specifically used for:

[0148] The mind map-generated intelligent agent uses an asynchronous concurrent scheduler based on semaphores to generate local subtrees for a set of parallel tasks, and determines the local subtree corresponding to each of the parallel generated tasks.

[0149] The target mind map is determined by merging the various local subtrees using a programmatic hierarchical tree merging algorithm.

[0150] Optionally, the parallel generation module 430 is also specifically used for:

[0151] The local subtrees are merged based on a programmatic hierarchical tree merging algorithm to determine the programmatic merging result;

[0152] If the direct child nodes of the programmatic merging result are greater than the preset child node number constraint, then the mind map merging agent is invoked to merge the programmatic merging result and determine the target mind map.

[0153] If the direct child nodes of the programmatic merge result are not greater than the preset child node number constraint, then the programmatic merge result is determined as the target mind map.

[0154] Optionally, the apparatus further includes a preprocessing generation module, which is used for:

[0155] If the mind map generation path is a preprocessing path, then the mind map data is processed by a three-modal preprocessor to determine a refined summary of the mind map.

[0156] The mind map generating agent generates a mind map based on the refined summary of the mind map, and determines the target mind map.

[0157] Optionally, the preprocessing generation module is specifically used for:

[0158] If the mind map data is a knowledge base document and there is a document block list, then the mind map data is processed based on the hierarchical perception model to determine the refined summary of the mind map.

[0159] If the mind map data is not a knowledge base document, and the entity density of the mind map data is greater than a preset threshold, then the mind map data is processed based on the entity perception mode to determine the refined summary of the mind map.

[0160] If the mind map data is not a knowledge base document, and the entity density of the mind map data is not greater than a preset threshold, then the mind map data is processed based on a flat pattern to determine the refined summary of the mind map.

[0161] Optionally, the apparatus further includes a direct generation module, which is used for:

[0162] If the mind map generation path is a direct processing path, then the mind map generation agent generates a mind map based on the mind map data to determine the target mind map.

[0163] The mind map adaptive generation device provided in the embodiments of the present invention can execute the mind map adaptive generation method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0164] Figure 5 A schematic diagram of an electronic device 10, which can be used to implement embodiments of the present invention, is shown. 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 can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their patterns are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0165] like Figure 5 As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0166] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of monitors, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer grids such as the Internet and / or various telecommunications grids.

[0167] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 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 processors running machine learning model algorithms, digital signal processors (DSPs), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as the adaptive mind map generation method.

[0168] In some embodiments, the adaptive mind map generation method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or mounted on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the adaptive mind map generation method described above may be performed. Alternatively, in other embodiments, processor 11 may be configured to perform the adaptive mind map generation method by any other suitable means (e.g., by means of firmware).

[0169] 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), payload-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.

[0170] Computer programs used to implement the methods of the present invention can be written in any combination of one or more programming languages. These computer programs can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the patterns / operations specified in the flowcharts and / or block diagrams to be implemented. The computer programs can 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.

[0171] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. 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 fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0172] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. 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).

[0173] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or grid browsers through which users 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 grid). Examples of communication grids include local area networks (LANs), wide area networks (WANs), blockchain grids, and the Internet.

[0174] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact through communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system to address the shortcomings of traditional physical hosts and VPS servers, such as high management difficulty and weak business scalability.

[0175] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0176] This embodiment provides a computer-readable storage medium storing a computer program thereon. When executed by a processor, the program implements the steps of the adaptive mind map generation method provided in any embodiment of the present invention. The method includes:

[0177] In response to a user's mind map generation request, identify the mind map data corresponding to the mind map generation request;

[0178] The mind map data is routed using a routing decision engine to determine the mind map generation path corresponding to the mind map data.

[0179] If the mind map generation path is a parallel generation path, then the mind map data is decomposed and merged in parallel to determine the target mind map.

[0180] The computer storage medium of this invention can be any combination of one or more computer-readable media. A computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable computer disk, a hard disk, 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 device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0181] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of sending, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device.

[0182] Program code contained on a computer-readable medium may be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0183] Computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages—such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of mesh, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0184] Those skilled in the art will understand that the modules or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a grid of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby allowing them to be stored in a storage device for execution by a computing device, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0185] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0186] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. 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 invention should be included within the scope of protection of this invention.

Claims

1. A mind map adaptive generation method, characterized in that, include: In response to a user's mind map generation request, identify the mind map data corresponding to the mind map generation request; The mind map data is routed using a routing decision engine to determine the mind map generation path corresponding to the mind map data. If the mind map generation path is a parallel generation path, then the mind map data is decomposed and merged in parallel to determine the target mind map; The step of performing routing decisions on the mind map data based on the routing decision engine to determine the mind map generation path corresponding to the mind map data includes: Based on the routing decision engine, source quantity characteristics, word size characteristics, and content length characteristics of mind map data are extracted; If there are multiple sources or multiple files in one source, the mind map data will be matched as a parallel generation path. If the source quantity feature is a single data source, and the word size feature is greater than a preset word threshold, then the mind map data is matched as a preprocessing path. If the source quantity characteristic is a single data source, and the content length characteristic exceeds a preset character threshold, then the mind map data is truncated and matched as a direct generation path. If the source quantity feature, word size feature, and content length feature of the mind map data do not meet any of the above conditions, then the mind map data is matched as the direct generation path; If the mind map generation path is a preprocessing path, then the mind map data is processed by a three-modal preprocessor to determine a refined summary of the mind map. The target mind map is determined by generating an intelligent agent based on the refined summary of the mind map. The process of processing the mind map data using a three-modal preprocessor to determine a concise summary of the mind map includes: If the mind map data is a knowledge base document and there is a document block list, then the mind map data is processed based on the hierarchical perception model to determine the refined summary of the mind map. If the mind map data is not a knowledge base document, and the entity density of the mind map data is greater than a preset threshold, then the mind map data is processed based on the entity perception mode to determine the refined summary of the mind map. If the mind map data is not a knowledge base document, and the entity density of the mind map data is not greater than a preset threshold, then the mind map data is processed based on a flat pattern to determine the refined summary of the mind map.

2. The method according to claim 1, characterized in that, The parallel task decomposition and merging of the mind map data to determine the target mind map includes: The mind map data is adaptively generated in parallel to determine a set of parallel generation tasks; wherein the set of parallel generation tasks includes at least one parallel generation task. The parallel generation tasks in the set of parallel generation tasks are merged to determine the target mind map.

3. The method according to claim 2, characterized in that, The adaptive parallel generation of the mind map data, and the determination of the set of parallel generation tasks, include: For each data source in the mind map data, if there are multiple associated files in the associated file list of the data source, then set a file-level task for each associated file; For each file-level task, if the file-level task has a segmentation structure, then the file-level task is decomposed into a set of segmented subtasks based on the segmentation structure. The parallel generation task set is constructed based on the entire set of segmented subtasks and the entire set of file-level tasks.

4. The method according to claim 3, characterized in that, Also includes: For each parallel generation task in the set of parallel generation tasks, a routing decision is made based on the routing decision engine to determine the large model input data corresponding to each parallel generation task.

5. The method according to claim 4, characterized in that, The step of merging the parallel generation tasks in the set of parallel generation tasks to determine the target mind map includes: The mind map-generated intelligent agent uses an asynchronous concurrent scheduler based on semaphores to generate local subtrees for a set of parallel tasks, and determines the local subtree corresponding to each of the parallel generated tasks. The target mind map is determined by merging the various local subtrees using a programmatic hierarchical tree merging algorithm.

6. The method according to claim 5, characterized in that, The programmatic hierarchical tree merging algorithm merges the various local subtrees to determine the target mind map, including: The local subtrees are merged based on a programmatic hierarchical tree merging algorithm to determine the programmatic merging result; If the number of direct child nodes in the programmatic merging result is greater than the preset child node count constraint, then the mind map merging agent is invoked to merge the programmatic merging result and determine the target mind map; If the direct child nodes of the programmatic merge result are not greater than the preset child node number constraint, then the programmatic merge result is determined as the target mind map.

7. The method according to claim 1, characterized in that, Also includes: If the mind map generation path is a direct generation path, then the mind map generation agent generates the mind map based on the mind map data to determine the target mind map.

8. The method according to claim 4, characterized in that, The step of performing routing decisions for the parallel generation tasks based on the routing decision engine, and determining the large model input data corresponding to each parallel generation task, includes: If the word size feature of the content field of the parallel generation task is greater than the preset word threshold, the content field of the parallel generation task is compressed into a mind map concise summary based on the three-modal preprocessor, and the mind map concise summary is determined as the input data of the large model. If the word size feature of the content field of the parallel generation task is not greater than the preset word threshold, and the content length feature of the content field of the parallel generation task is greater than the preset character threshold, then paragraph boundary truncation processing is performed on the content field of the parallel generation task to obtain a refined summary of the mind map, and the refined summary of the mind map is determined as the input data of the large model. If the lexical size feature of the content field of the parallel generation task is not greater than a preset lexical threshold, and if the content length feature of the content field of the parallel generation task is not greater than a preset character threshold, then the content field of the parallel generation task is determined as the input data of the large model.

9. A mind map adaptive generation device, characterized in that, include: The request and response module is used to respond to the mind map generation request submitted by the user and identify the mind map data corresponding to the mind map generation request. The hierarchical routing module is used to make routing decisions on the mind map data based on the routing decision engine, and determine the mind map generation path corresponding to the mind map data. The parallel generation module is used to perform parallel task decomposition and merging on the mind map data to determine the target mind map if the mind map generation path is a parallel generation path. The step of performing routing decisions on the mind map data based on the routing decision engine to determine the mind map generation path corresponding to the mind map data includes: Based on the routing decision engine, source quantity characteristics, word size characteristics, and content length characteristics of mind map data are extracted; If there are multiple sources or multiple files in one source, the mind map data will be matched as a parallel generation path. If the source quantity feature is a single data source, and the word size feature is greater than a preset word threshold, then the mind map data is matched as a preprocessing path. If the source quantity characteristic is a single data source, and the content length characteristic exceeds a preset character threshold, then the mind map data is truncated and matched as a direct generation path. If the source quantity feature, word size feature, and content length feature of the mind map data do not meet any of the above conditions, then the mind map data is matched as the direct generation path; The apparatus further includes a preprocessing generation module, the preprocessing generation module being used for: If the mind map generation path is a preprocessing path, then the mind map data is processed by a three-modal preprocessor to determine a refined summary of the mind map. The target mind map is determined by generating an intelligent agent based on the refined summary of the mind map. The preprocessing generation module is specifically used for: If the mind map data is a knowledge base document and there is a document block list, then the mind map data is processed based on the hierarchical perception model to determine the refined summary of the mind map. If the mind map data is not a knowledge base document, and the entity density of the mind map data is greater than a preset threshold, then the mind map data is processed based on the entity perception mode to determine the refined summary of the mind map. If the mind map data is not a knowledge base document, and the entity density of the mind map data is not greater than a preset threshold, then the mind map data is processed based on a flat pattern to determine the refined summary of the mind map.

10. An electronic device, characterized in that, The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores a computer program that can be executed by the at least one processor, the computer program being executed by the at least one processor to enable the at least one processor to perform the mind map adaptive generation method according to any one of claims 1-8.

11. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that cause a processor to execute the mind map adaptive generation method according to any one of claims 1-8.

Images