Active time-sensitive knowledge structure evolution method and system, electronic device and storage medium

Abstract

The application discloses an active time-sensitive knowledge structure evolution method and system, an electronic device and a storage medium, and aims to solve the "cold start" problem and the prediction risk control problem of a time-sensitive retrieval system when responding to a new sudden event. The method comprises the following steps: predicting a future potential hot topic; in response to the predicted topic, creating a temporary sub-tree with a to-be-verified state which is isolated from a main retrieval tree; verifying the reliability of the temporary sub-tree by using real-time data flow to generate a verification result; and based on the verification result, automatically integrating the temporary sub-tree with reliable verification into the main retrieval tree, or deleting the temporary sub-tree with unreliable verification. Through the "prediction-verification" closed-loop mechanism, the traditional passive response is changed into predictive service, which helps to reduce the "cold start" delay, and meanwhile, through isolation and verification, the waste of resources and the pollution of the knowledge base caused by false prediction are avoided, and balance between forward-looking layout and risk control can be achieved.

Description

Technical Field

[0001] This application relates to the fields of artificial intelligence and information retrieval technology, and in particular to an active time-sensitive knowledge structure evolution method, system, electronic device and storage medium. Background Technology

[0002] In the field of time-sensitive information retrieval, such as news retrieval, the core challenge lies in balancing the depth, breadth, and response speed of search results. Traditional static indexing methods have slow knowledge structure updates, making it difficult to adapt to rapidly changing news hotspots, resulting in low recall rates and poor information organization capabilities for new events.

[0003] To address this issue, existing technologies have developed multi-agent retrieval methods based on large language models. These methods collaboratively construct reusable "retrieval trees" in an offline phase, thus mitigating to some extent the contradiction between the high latency of pure agent retrieval and the poor adaptability of static indexes. However, these methods are essentially still passive and reactive frameworks, and their adaptive adjustments depend on the evaluation of existing corpora.

[0004] Therefore, for a completely new, sudden, and historically unprecedented hot topic, the system cannot effectively respond in the early stages due to the sparse relevant corpus and the lack of corresponding knowledge structure in the main search tree. This results in a significant "cold start" problem, preventing users from obtaining structured and in-depth search results. Furthermore, blindly expanding the main search tree through simple predictions will lead to a huge waste of computational resources due to the inherent uncertainty of predictions. Incorrect predictions will directly "pollute" the main search tree, reducing its overall reliability and search performance, ultimately resulting in more harm than good. Summary of the Invention

[0005] The purpose of this application is to provide an active time-sensitive knowledge structure evolution method, system, electronic device and storage medium, which aims to solve the "cold start" problem of existing time-sensitive retrieval systems when dealing with new emergencies, and the technical challenge of how to balance forward-looking layout and control of prediction failure risk (i.e. waste of computing resources and pollution of knowledge structure) when introducing prediction mechanism.

[0006] To achieve the above objectives, this application provides a proactive time-sensitive knowledge structure evolution method, comprising: a prediction step: predicting potential future hot topics; an isolation construction step: in response to the predicted hot topics, creating a temporary subtree isolated from the main search tree and having a state to be verified; a verification step: verifying the reliability of the temporary subtree using real-time data streams and generating verification results; and an evolution step: based on the verification results, automatically merging the verified reliable temporary subtrees into the main search tree and automatically deleting the verified unreliable temporary subtrees.

[0007] Optionally, the isolation construction step further includes at least one of the following: making the temporary subtree independent of the main search tree in logical storage space and / or physical storage space, blocking direct write operations of the temporary subtree to the main search tree; limiting the construction resource consumption of the temporary subtree, the limiting method including limiting the maximum depth and / or limiting the number of external application interface calls.

[0008] Optionally, the method further includes a state management step: maintaining a lifecycle state machine for each temporary subtree, the lifecycle state machine including at least a creation state, a verification state, a promotion state, and a pruning state.

[0009] Optionally, the verification step includes: calculating the correlation score between the temporary subtree and the real-time data stream; comparing the correlation score with a preset promotion threshold and a preset pruning threshold, and determining whether a preset time window condition is met.

[0010] Optionally, the verification step includes: when the relevance score is consistently higher than the promotion threshold and meets the preset time window condition, merging the temporary subtree into the main search tree and changing its state to promotion state; when the relevance score is consistently lower than the pruning threshold or the lifespan of the temporary subtree exceeds the preset maximum lifespan, deleting the temporary subtree and changing its state to pruning state.

[0011] Optionally, the prediction step includes: analyzing multimodal data; generating confidence scores for predicted hot topics; and performing the isolation construction step only when the confidence scores exceed a preset trigger threshold.

[0012] To achieve the above objectives, this application also provides an active time-sensitive knowledge structure evolution system for implementing the method described in any of the preceding claims, comprising: a main retrieval tree for storing confirmed knowledge structures; a prediction module for executing the prediction step; an isolation construction module for executing the isolation construction step; a verification module for executing the verification step; and an evolution module for executing the evolution step.

[0013] Optionally, the system includes at least one of the following: (1) The isolation construction module further includes an isolation implementation unit and / or a resource limiting unit; the isolation implementation unit is used to make the temporary subtree independent of the main search tree in the logical storage space and / or physical storage space, and to block the direct write operation of the temporary subtree to the main search tree; the resource limiting unit includes a depth counter and / or an API call counter, the depth counter is used to limit the maximum depth of the temporary subtree, and to forcibly terminate the construction when the depth of the temporary subtree reaches a preset threshold, and the API call counter is used to limit the maximum number of times the external application interface is called during the construction process, and to forcibly terminate the construction when the number of calls reaches a preset threshold; (2) a state management module, used to perform state management steps.

[0014] To achieve the above objectives, this application also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the active time-sensitive knowledge structure evolution method described above.

[0015] To achieve the above objectives, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the active time-sensitive knowledge structure evolution method described above.

[0016] The embodiments of this application have the following beneficial effects:

[0017] 1. By predicting potential hotspots in advance through a forecasting step and pre-constructing temporary subtrees through an isolation building step, the system can prepare a deep knowledge structure at the very beginning of a hotspot event or even before it occurs. Once a small amount of relevant data appears and triggers verification, the reliable temporary subtree can be merged into the main search tree through an evolutionary step, thereby providing users with high-quality structured search results. This helps reduce the "cold start" delay of retrieval for sudden events and transforms traditional passive response into predictive service.

[0018] 2. By isolating the construction steps, the uncertain temporary subtrees are logically or physically separated from the stable and reliable main search tree. A verification step then performs data-driven checks on the reliability of the temporary subtrees. Only verified temporary subtrees are incorporated into the main search tree, while those that fail verification are deleted. This risk isolation and verification mechanism avoids the waste of computational resources and knowledge structure pollution caused by blindly expanding the main search tree due to erroneous predictions, helping to achieve a balance between foresight and economy.

[0019] 3. By introducing state management steps and a lifecycle state machine into the temporary subtree, and combining them with automatic promotion and pruning mechanisms in the evolutionary steps, this method constructs a self-evolving closed loop of knowledge structure driven by "prediction-verification". The system can proactively adapt to changes in the information environment and continuously maintain high performance. Simultaneously, the automatic pruning mechanism ensures that the system can self-clean up invalid and outdated speculative knowledge, maintaining the purity and efficiency of the main retrieval tree, and enhancing the system's self-evolution and self-healing capabilities. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a flowchart illustrating the proactive time-sensitive knowledge structure evolution method according to an embodiment of this application.

[0022] Figure 2 This is a schematic diagram of the structure of the proactive time-sensitive knowledge structure evolution system according to an embodiment of this application.

[0023] Figure 3 This is a schematic diagram of the lifecycle state transition of a temporary subtree according to an embodiment of this application.

[0024] Explanation of reference numerals in the attached figures:

[0025] 10. Prediction module; 20. Isolation construction module; 21. Isolation implementation unit; 22. Resource constraint unit; 30. Verification module; 40. Evolution module; 50. Main search tree; 60. Temporary subtree; 106. Decision box; 108. Decision box. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of this application clearer, the application will be further described in detail below with reference to the accompanying drawings. The described embodiments should not be considered as limitations on this application. All other embodiments obtained by those skilled in the art without inventive effort are within the scope of protection of this application. Unless otherwise defined, all technical and scientific terms used in the embodiments of this application have the same meaning as commonly understood by those skilled in the art. The terminology used in the embodiments of this application is for the purpose of describing the embodiments of this application only and is not intended to limit this application.

[0027] Before providing a further detailed description of the embodiments of this application, the nouns and terms involved in the embodiments of this application will be explained, and the nouns and terms involved in the embodiments of this application shall be interpreted as follows.

[0028] (1) Main retrieval tree: refers to a data structure used to store confirmed and persistent knowledge structures, serving as the core knowledge base of an information retrieval system. In an information retrieval system, this structure is usually verified and confirmed, possessing high reliability, and is the foundation for the system to provide stable retrieval services to the outside world.

[0029] (2) Temporary subtree: refers to a new type of knowledge structure unit with a "to be verified" state, which is isolated from the main retrieval tree in logical storage space and / or physical storage space, and is used to carry predictively generated knowledge content whose reliability has not yet been verified. It is the core carrier for realizing forward-looking layout while controlling risks in the embodiments of this application.

[0030] (3) Lifecycle state machine: refers to a model used to manage the entire process of temporary subtrees from creation to destruction, including at least the creation state, the verification state, the promotion state, and the pruning state, and defines the conditions for the transition between states. Through this mechanism, predictive knowledge units can be managed in a refined and automated manner.

[0031] (4) Relevance Score: This is a quantitative indicator used to measure the degree of matching or relevance between the knowledge structure represented by the temporary subtree and the content of the real-time data stream (such as a news stream). This score can be calculated based on standard evaluation indicators in the field of information retrieval, such as Normalized Discounted Cumulative Gain (NDCG), Recall, and Precision. This score provides an objective basis for decision-making in the validation steps.

[0032] (5) Promotion threshold: This refers to a preset value. When the relevance score of a temporary subtree is consistently higher than this threshold, it triggers the evolution operation of merging into the main retrieval tree. This threshold is the key threshold for determining whether predictive knowledge is "converted" into formal knowledge.

[0033] (6) Pruning threshold: This refers to a preset value. When the relevance score of a temporary subtree continues to fall below this threshold, it may trigger an evolutionary operation to delete it. This threshold is used to promptly eliminate predictions with low value or those that have been falsified, preventing waste of resources.

[0034] Please see Figure 1 and Figure 2This application provides a proactive, time-sensitive knowledge structure evolution method and system, aiming to solve the "cold start" problem in existing information retrieval systems when dealing with new and unexpected events, as well as the difficulty in balancing forward-looking planning and risk control when introducing prediction mechanisms. Existing technologies typically respond passively to events that have already occurred, leading to chaotic information organization and poor retrieval results in the early stages of an event. This application achieves proactive and forward-looking evolution of knowledge structures by introducing a "prediction-verification" closed-loop mechanism.

[0035] First, in prediction step S101, potential future hot topics are predicted. This step, based on data analysis, proactively identifies events or topics that may emerge, enabling the construction of knowledge structures to be brought forward, shifting from "post-event response" to "pre-event planning," thereby solving the "cold start" delay problem in existing technologies. By predicting in advance, the system can prepare a structured knowledge framework in advance for the upcoming information deluge.

[0036] Next, in the isolation construction step S102, in response to the predicted hot topics, a temporary subtree 60 with a pending verification state is created, isolated from the main retrieval tree 50. By confining high-risk exploration behavior within the temporary subtree 60, the negative impact of prediction failures is prevented, such as erroneous knowledge directly "polluting" the stable and reliable main retrieval tree 50, leading to a decline in overall retrieval performance. Encapsulating uncertain predictive knowledge solves the problem of blind prediction polluting the main knowledge base in existing technologies, achieving effective risk isolation.

[0037] Subsequently, in verification step S103, the system utilizes real-time data streams to continuously verify the reliability of the temporary subtree 60 and generates verification results. Compared to the passive frameworks in existing technologies that can only evaluate the current corpus, this application employs a parallel and continuous verification process, using real-world data to test the accuracy of predictions, thereby enabling dynamic and near-real-time evaluation of the value of predictive knowledge. By introducing a data-driven verification process, the system ensures that the evolution of the knowledge structure is based on facts rather than guesswork, achieving a transformation from "subjective prediction" to "objective confirmation."

[0038] Finally, in evolution step S104, based on the verification results, the system automatically merges the verified reliable temporary subtree 60 into the main retrieval tree 50 and automatically deletes the unverified temporary subtree 60. Through automated promotion and pruning, the system can not only integrate valuable new knowledge into the core knowledge base "on time," but also automatically clean up invalid predictions and reclaim computing resources, solving the problem of resource waste caused by prediction failures in existing technologies, and constructing an adaptive and self-healing knowledge evolution closed loop. Through this knowledge evolution closed loop, the system achieves continuous, efficient, and low-risk dynamic evolution of the knowledge structure.

[0039] Furthermore, in a preferred embodiment, the prediction step S101 includes more specific operations. For example... Figure 1 As shown, the process first involves analyzing multimodal data 110. This multimodal data can include structured data from within the system, such as the selection frequency of historical optimal subtrees, the activation time decay weight of each knowledge node, and high-frequency growth words in user query logs; it can also include unstructured data from external sources, such as the detection results of emerging words on social media (e.g., Weibo, Twitter), early warning announcements from authoritative institutions (e.g., government, research institutions), and news forecasts for specific industries. Analyzing these diverse data sources allows for the capture of early signals of hot events from a broader perspective, resulting in higher predictive sensitivity and accuracy compared to a single data source. Subsequently, based on the analysis results, hot topics 120 are predicted, and a confidence score is generated for each predicted hot topic 120. This confidence score is a quantitative assessment of the reliability of the prediction. Finally, the subsequent isolation construction steps are only executed when the confidence score exceeds a preset trigger threshold. By setting a trigger threshold, a large number of low-confidence predictions can be effectively filtered out, avoiding the initiation of the construction process for these low-value predictions, thus achieving the first screening of prediction quality.

[0040] In another preferred embodiment, the specific implementation of the isolation construction step S102 may include at least one of the following two methods. The first method is to make the temporary subtree 60 independent of the main retrieval tree 50 in logical and / or physical storage space, blocking direct write operations from the temporary subtree 60 to the main retrieval tree 50. For example, logically, an independent namespace or database table can be allocated to the temporary subtree 60 to ensure that its data records do not logically intersect with the records of the main retrieval tree 50; physically, it can be stored in different memory areas, disk partitions, or even independent servers. This isolation mechanism can block direct write operations from the temporary subtree 60 to the main retrieval tree 50, ensuring that even if the construction process of the temporary subtree is abnormal or its content contains errors, the main retrieval tree 50 will not be accidentally modified or damaged, thereby guaranteeing the stability and integrity of the core knowledge base. The second method is to limit the resource consumption of the temporary subtree 60's construction. The limiting methods include limiting the maximum depth and / or limiting the number of external application interface calls. For example, a smaller exploration depth can be set (e.g., D_provisional <= 3), or a maximum limit on the total number of application programming interface (API) calls can be imposed (e.g., N_api_calls <= 20). The principle behind this resource constraint is to keep the cost of prediction failures within an acceptable range. By setting a clear resource "ceiling" for this exploratory building task, unlimited investment of computing resources due to uncertain predictions is avoided, achieving a balance between forward-looking exploration and economic cost control.

[0041] Furthermore, to enable more refined management of the temporary subtree 60, this embodiment of the application also introduces a state management step. Please refer to... Figure 3 This step maintains a lifecycle state machine for each temporary subtree 60. This lifecycle state machine includes at least four core states: creation state E1, pending verification state E2, promotion state E3, and pruning state E4. When the system decides to create a temporary subtree, it first enters the creation state E1, completes initialization, and immediately transitions to the pending verification state E2. In the pending verification state E2, it undergoes verification by real-time data streams. Based on the verification results, it may be "promoted" to the promotion state E3 or "removed" to the pruning state E4. The promotion state E3 and pruning state E4 are the endpoints of the lifecycle of this temporary subtree. The lifecycle state machine assigns a clear and traceable management state to an abstract "knowledge unit," enabling the system to clearly understand the current stage of each predictive knowledge. In this way, this solution solves the problem of chaotic management of a large number of uncertain knowledge units, achieving automated and process-oriented management of the entire process from knowledge creation and verification to final disposal, and enhancing the system's self-organization and self-maintenance capabilities.

[0042] In a preferred embodiment, based on the above-described state management, the verification step S103 is further refined. For example... Figure 1 As shown, this step specifically includes calculating the relevance score 140 between the temporary subtree and the real-time data stream. The system uses the real-time data stream to calculate the relevance score between each temporary subtree 60 in the pending verification state E2 and the real-time data stream. This relevance score can be calculated using mature metrics from the information retrieval field. For example, the structure of the temporary subtree can be used as a query or filter to measure the ranking quality (metric type such as NDCG) or quantity (metric type such as Recall) of relevant documents retrieved from the real-time news corpus. After calculating the score, the system compares the relevance score with preset promotion thresholds and preset pruning thresholds, and determines whether preset time window conditions are met. For example, it's not just a matter of the score exceeding the promotion threshold once, but rather requiring the score to remain consistently above the promotion threshold within the preset time window conditions (e.g., for two consecutive hours). This increases the robustness of the decision-making process, avoiding erroneous evolutionary decisions due to noise or short-term fluctuations in a single data point. Through this comprehensive judgment mechanism based on "score-threshold-time window," the objectivity and reliability of the verification results are ensured, solving the unreliability problem of judging knowledge value solely based on instantaneous performance.

[0043] Further, in a more preferred embodiment, the verification step S103 includes: when the relevance score is consistently higher than the promotion threshold and a preset time window condition is met, merging the temporary subtree into the main search tree and changing its state to promotion; when the relevance score is consistently lower than the pruning threshold or the lifespan of the temporary subtree exceeds a preset maximum lifespan, deleting the temporary subtree and changing its state to pruning. Please refer to [link to relevant documentation]. Figure 3 When the relevance score of temporary subtree 60 remains above the promotion threshold within a preset time window, the system determines the prediction is successful, executes the operation of merging it into the main search tree 150, and changes its state from pending verification state E2 to promotion state E3. This corresponds to... Figure 1 The decision box 106 in the code represents the "yes" path. Conversely, when the relevance score of temporary subtree 60 remains below the pruning threshold, or its lifespan since creation exceeds the preset maximum lifespan (e.g., 72 hours), the system determines that the prediction has failed or is outdated, executes the operation of deleting temporary subtree 160, and changes its status from pending verification state E2 to pruned state E4. This corresponds to... Figure 1 The decision box 108 in the system represents the "yes" path. If the score neither meets the promotion criteria nor triggers the pruning criteria, it will remain in a pending verification state and continue to undergo the next round of verification. Thus, the system constructs a fully automated, rule-based decision-making loop that can autonomously perform "survival of the fittest" in knowledge selection, integrating verified high-quality knowledge into the system while decisively removing invalid or outdated knowledge, achieving self-purification and self-evolution of the knowledge base.

[0044] Please see Figure 2 This application also provides an active time-sensitive knowledge structure evolution system for implementing the methods described in any of the above embodiments. The system includes a main retrieval tree 50 for storing confirmed knowledge structures. Furthermore, the system includes multiple functional modules to execute various steps of the method.

[0045] Specifically, the system includes a prediction module 10, which performs the aforementioned prediction steps. The prediction module 10 can receive external signals (such as social media data streams) and internal data (such as user query logs), analyze this multimodal data using a built-in time series analysis model or machine learning model to predict potential future trending topics, and output a list of topics with confidence scores. This prediction module 10 is the source of the system's overall forward-looking capability.

[0046] The system also includes an isolation building module 20, which performs the aforementioned isolation building steps. Upon receiving a high-confidence prediction topic from the prediction module 10, the isolation building module 20 is responsible for creating at least one temporary subtree 60 isolated from the main retrieval tree 50. This isolation building module 20 is a key component for implementing risk control and cost management.

[0047] The system further includes a verification module 30, which performs the aforementioned verification steps. The verification module 30 continuously receives real-time data streams and uses this data to evaluate the temporary subtrees 60 created by the isolation construction module 20 and placed in a state awaiting verification, calculating relevance scores for them. This verification module 30 acts as a bridge connecting prediction and reality, providing data support for subsequent evolutionary decisions.

[0048] Finally, the system includes an evolution module 40, which executes the aforementioned evolution steps. The evolution module 40 receives the verification results (i.e., relevance scores) from the verification module 30 and, according to preset promotion and pruning rules, makes a final decision on the temporary subtree 60. It either commands the successfully verified temporary subtree 60 to be incorporated into the main retrieval tree 50, or commands the unverified temporary subtree 60 to be deleted. This evolution module 40 is the executor that enables the automatic evolution and self-healing capabilities of the knowledge structure.

[0049] In a preferred embodiment, the isolation construction module 20 may further include an isolation implementation unit 21 and a resource limiting unit 22. The isolation implementation unit 21 is specifically responsible for technical isolation operations, such as creating an independent database sandbox, using containerization technology, or allocating independent memory address space to ensure that the temporary subtree 60 is logically and / or physically completely separated from the main retrieval tree 50. The resource limiting unit 22 incorporates monitoring tools such as a depth counter and / or an API (Application Programming Interface) call counter. The depth counter tracks the depth of knowledge exploration in real time during the construction of the temporary subtree, and forcibly terminates the continued construction of that branch once a preset threshold (e.g., depth 3) is reached. The API call counter limits the maximum number of calls to external application interfaces (e.g., calling large language models for content generation) during the construction process, and blocks subsequent calls once a preset threshold (e.g., 20 times) is reached. The collaborative work of the isolation implementation unit 21 and the resource limiting unit 22 strengthens the risk control effect of the isolation construction step from two dimensions: "security boundary" and "resource budget."

[0050] In another preferred embodiment, the system further includes a state management module. This state management module is responsible for performing the aforementioned state management steps, maintaining and updating the lifecycle state machines (e.g., for all temporary subtrees 60 in the system) for all temporary subtrees 60. Figure 3 (As shown). It records the current state of each temporary subtree (creation state E1, pending verification state E2, promotion state E3, or pruning state E4), creation time, and state change history. When the evolution module 40 makes a promotion or pruning decision, it notifies the state management module to update the state of the corresponding temporary subtree. The existence of this state management module provides a clear and centralized view for the entire system to monitor the dynamic changes of all predictive knowledge units, which is the foundation for achieving refined management.

[0051] This application also provides an electronic device, such as a server, personal computer, or mobile terminal. The electronic device includes a processor, a memory, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it can implement the proactive time-sensitive knowledge structure evolution method described in any of the foregoing embodiments. For example, the memory stores data of the main retrieval tree 50, as well as code instructions that implement the functions of the prediction module 10, the isolation construction module 20, the verification module 30, and the evolution module 40. The processor loads and executes these instructions, thereby completing the entire process from predicting hotspots and isolating and constructing temporary subtrees 60 to verifying and finally evolving the main retrieval tree 50.

[0052] This application also provides a computer-readable storage medium, such as a non-volatile storage medium like a hard disk, solid-state drive (SSD), USB flash drive, or optical disk. The storage medium stores a computer program. When the computer program is executed by the processor of an electronic device, it can implement the proactive time-sensitive knowledge structure evolution method described in any of the foregoing embodiments. This allows the technical solution of this application to be distributed, installed, and deployed as a software product.

[0053] To more clearly demonstrate the effect of the synergistic operation of all the technical features of this application, the workflow of this application is described below based on a specific embodiment.

[0054] In one specific embodiment, the proactive time-sensitive knowledge structure evolution system of this application is deployed to process dynamic information. First, the prediction module 10 starts working, continuously analyzing real-time data streams from social media and user query growth rate data within the system using a built-in LSTM (Long Short-Term Memory) model and a sudden word detection algorithm. At a certain moment, it predicts a new hot topic, "New Breakthrough in Quantum Computing," and generates a confidence score of 0.9 for it. Since this confidence score is higher than the system's preset trigger threshold of 0.7, the prediction module 10 passes the hot topic and confidence score to the isolated construction module 20.

[0055] Upon receiving the instruction, the isolation construction module 20 starts immediately. The isolation implementation unit 21 of the isolation construction module 20 creates a new table in an independent database schema to store the temporary subtree 60 to be built, thus achieving logical isolation from the main search tree 50. Simultaneously, the resource limiting unit 22 sets constraints for this construction task, including a maximum depth of 4 and a maximum number of external API calls of 15. The isolation construction module 20 generates a temporary subtree 60 around the theme of "New Breakthroughs in Quantum Computing," containing initial branches such as "Core Principles," "Technical Challenges," "Application Prospects," and "Related Companies." After construction is complete, the state management module sets the state of the temporary subtree 60 to "Pending Verification State E2" and records its creation timestamp.

[0056] Next, the verification module 30 begins continuous monitoring of this new temporary subtree 60. It takes the real-time feed of tech news as input and uses a lightweight evaluation model to calculate the NDCG@10 relevance score for the temporary subtree 60 every 10 minutes. In the initial few hours, due to the limited amount of relevant news, the score hovers between 0.1 and 0.3, below the pruning threshold of 0.2. However, because the "consistently below" time window condition is not met, pruning is not triggered.

[0057] Six hours after the creation of temporary subtree 60, a technology company officially released its research findings, which triggered media coverage. Verification module 30 detected a surge in the number of high-quality news articles related to this topic, and the relevance score of temporary subtree 60 rose to 0.95 within half an hour. For the next two consecutive hours, its score remained stable above 0.9, consistently exceeding the system's preset promotion threshold of 0.85, and meeting the "two-hour duration" time window condition.

[0058] The verification module 30 sends a successful verification signal to the evolution module 40. The evolution module 40 then performs a promotion operation: it seamlessly merges the entire structure and content of the temporary subtree 60 stored in the isolated database table into a new branch of the main retrieval tree 50. After the merge is complete, the evolution module 40 notifies the state management module to update the state of the temporary subtree 60 to "promotion state E3". This completes the prediction-triggered knowledge structure evolution cycle.

[0059] Meanwhile, the system may have also predicted another topic with a confidence level of 0.75, "room-temperature superconducting materials," and created another temporary subtree for it. However, during its maximum lifespan of 72 hours, no substantial news reports emerged to support this prediction. Its correlation score remained below 0.1. When its lifespan reached 72 hours, the evolution module 40 triggered a pruning operation according to preset rules, completely deleting this invalid temporary subtree 60 from the system and updating its state to "pruned state E4," thereby reclaiming the small amount of computing and storage resources it occupied.

[0060] Through the complete collaborative workflow described above, this embodiment demonstrates how various technical features work closely together to form an efficient, adaptive, and risk-controlled knowledge evolution ecosystem. It achieves accurate prediction through "multimodal data analysis + confidence threshold"; secure construction through "logical isolation + resource constraints"; refined management through "lifecycle state machine"; objective verification through "relevance score + dual threshold + time window"; and finally, self-healing and self-evolution of the knowledge base through "automatic promotion / pruning." This comprehensive effect helps to achieve both "prevention is better than cure" and avoid "blind investment" when dealing with emergencies, striking a balance between foresight and economy.

[0061] The technical solution of this application can be applied to various scenarios that require rapid response to dynamic information. The following two specific application scenario examples will further illustrate the practical application value of this application.

[0062] The system described in this application can be applied to epidemic monitoring and information retrieval scenarios in the public health field. In this scenario, the prediction module 10 continuously monitors medical preprint servers (such as medRxiv), professional medical forums, and mainstream social media worldwide. On a certain day, the prediction module 10, through its built-in LSTM model, detected an abnormal 300% increase in the frequency of discussions about "unexplained respiratory diseases in Southeast Asia" within 6 hours, and simultaneously extracted emerging keywords such as "high fever," "dry cough," and "contact history" from unstructured text. Based on this, the prediction module 10 generates a prediction topic: "novel respiratory infectious disease," and gives a confidence score of 0.78, which exceeds the preset trigger threshold of 0.7.

[0063] The system responds immediately, with the isolation building module 20 creating a temporary subtree 60 for the topic of "novel respiratory infectious diseases". Considering the limited initial information, the resource limiting unit 22 restricts the maximum depth of this temporary subtree 60 to 3 and the number of API calls to no more than 15. Its automatically generated initial branches include: {virus origin, clinical symptoms, transmission routes, preventative measures, potential treatment options}. This temporary subtree 60 is assigned a "pending verification state E2" and is completely isolated from the main retrieval tree 50, which stores mature disease information.

[0064] Eighteen hours after the creation of temporary subtree 60, the World Health Organization (WHO) issued its first official notification about the disease, followed by reports from multiple media outlets. When processing these real-time news feeds, verification module 30 found that the NDCG@10 relevance score of temporary subtree 60 rose from 0.2 to 0.92 within one hour and remained stable above 0.9, exceeding the system's set promotion threshold of 0.85. Evolution module 40 automatically triggered a promotion operation, seamlessly merging temporary subtree 60 into the main search tree 50 and updating its status to "Promotion Status E3". At this point, when users search for "latest virus" or related symptoms, the system no longer returns scattered news links but directly presents a structured result organized by the temporary subtree, containing officially defined symptoms, known transmission methods, and authoritative prevention recommendations. This helps improve the quality and efficiency of information access in the early stages of a public health emergency and shortens response time.

[0065] Another application scenario is business competitive intelligence analysis in the technology sector. In this scenario, the prediction module 10 analyzes recent patent applications, recruitment information of supply chain partners, and relevant statements by executives of a well-known technology company (referred to as "Company A") on social media to predict that the company is highly likely to release an "augmented reality (AR) glasses" within the next month, and gives a prediction confidence level of 0.82.

[0066] The system triggered the creation of a temporary subtree 60 for "Company A AR Glasses". Considering the complexity of technological products, the exploration depth of this temporary subtree was set to 4, with initial branches including: {Hardware specifications (processor, display), software ecosystem (operating system, core applications), price and release date, major competitors (product comparison)}. Simultaneously, the prediction module 10 also predicted the "Company A Smart Car" project based on some weaker signals, with a low confidence level of 0.68 (but still above the trigger threshold of 0.65), and the system also created a temporary subtree for it.

[0067] For the next month, the temporary subtree of the "Company A Smart Car" project maintained a relevance score close to 0 due to the lack of any related news or official information. 72 hours after its creation, its lifecycle expired, and Evolution Module 40 automatically triggered a pruning operation, removing it from the system. Its state changed to "Pruned State E4," thus releasing the small amount of computing and storage resources it occupied. This demonstrates the system's self-healing capability.

[0068] Twenty-five days after the creation of the temporary subtree, Company A officially held a new product launch event, releasing its AR glasses. During and after the event, a large number of reviews, analyses, and news reports emerged. The relevance score of the temporary subtree regarding "Company A AR glasses" quickly exceeded the promotion threshold and was automatically merged into the main search tree 50 by evolution module 40. This enabled queries about the product to obtain comprehensive information, including specifications, price, and competitor analysis, immediately after the product launch, providing industry analysts, investors, and consumers with highly timely and in-depth intelligence. This scenario illustrates the beneficial effects of this application in improving the efficiency of computing resource utilization and achieving "Just-in-Time" knowledge management.

[0069] The above description is merely a preferred embodiment of this application and is not intended to limit this application. For those skilled in the art, without departing from the core ideas of this application, various improvements, modifications, combinations, or equivalent substitutions can be made to the specific implementation methods of this application based on the content disclosed in the specification. For example, the model type of the trend prediction agent, the resource limitation parameters for constructing temporary subtrees, the calculation method of the correlation score, or the specific values ​​of the promotion threshold and lifecycle can be adjusted according to the actual application scenario. Any modifications, equivalent substitutions, or partial optimizations made based on the spirit of this application should be considered to fall within the protection scope of this application. Any embodiment or set of features disclosed in this specification can be combined with other embodiments or features without contradicting them, and the resulting new technical solutions should also be considered as disclosed in this specification.

Claims

1. A proactive time-sensitive knowledge structure evolution method, characterized in that, include: Forecasting steps: Predict potential future hot topics; Isolation construction steps: In response to predicted hot topics, create temporary subtrees with pending verification states that are isolated from the main search tree; Verification steps: Verify the reliability of the temporary subtree using real-time data stream and generate verification results; Evolutionary steps: Based on the verification results, the temporary subtrees with reliable verification are automatically merged into the main search tree, and the temporary subtrees with unreliable verification are automatically deleted.

2. The method according to claim 1, characterized in that, The isolation construction step also includes at least one of the following: Make the temporary subtree independent of the main search tree in logical storage space and / or physical storage space, and block the temporary subtree from direct write operations to the main search tree; Limit the resource consumption for building the temporary subtree, including limiting the maximum depth and / or limiting the number of external application interface calls.

3. The method according to claim 1, characterized in that, It also includes a state management step: maintaining a lifecycle state machine for each temporary subtree, the lifecycle state machine including at least a creation state, a pending verification state, a promotion state, and a pruning state.

4. The method according to claim 3, characterized in that, The verification steps include: Calculate the correlation score between the temporary subtree and the real-time data stream; The correlation score is compared with a preset promotion threshold and a preset pruning threshold, and it is determined whether the preset time window condition is met.

5. The method according to claim 4, characterized in that, The verification steps include: When the relevance score is consistently higher than the promotion threshold and meets the preset time window conditions, the temporary subtree is merged into the main search tree and its status is changed to promotion status. When the correlation score remains below the pruning threshold or the lifespan of the temporary subtree exceeds the preset maximum lifespan, the temporary subtree is deleted and its state is changed to pruning state.

6. The method according to any one of claims 1 to 5, characterized in that, The prediction steps include: Analyze multimodal data; Generate confidence scores for predicted trending topics; The isolation construction step is executed only when the confidence score exceeds a preset trigger threshold.

7. A proactive time-sensitive knowledge structure evolution system, characterized in that, To implement the method of any one of claims 1-6, comprising: The main search tree is used to store the confirmed knowledge structure; A prediction module is used to perform the prediction steps; An isolation build module is used to perform the isolation build steps; The verification module is used to perform the verification steps; An evolution module is used to execute the evolution steps.

8. The system according to claim 7, characterized in that, Includes at least one of the following: (1) The isolation construction module further includes an isolation implementation unit and / or a resource limiting unit; the isolation implementation unit is used to make the temporary subtree independent of the main search tree in the logical storage space and / or physical storage space, and to block the direct write operation of the temporary subtree to the main search tree; the resource limiting unit includes a depth counter and / or an API call counter, the depth counter is used to limit the maximum depth of the temporary subtree, and the construction is forcibly terminated when the depth of the temporary subtree reaches a preset threshold, and the API call counter is used to limit the maximum number of times the external application interface is called during the construction process, and the construction is forcibly terminated when the number of calls reaches a preset threshold; (2) Status management module, used to execute status management steps.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the active time-sensitive knowledge structure evolution method as described in any one of claims 1 to 6.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the proactive time-sensitive knowledge structure evolution method as described in any one of claims 1 to 6.

Images