Metadata management method and device based on semantic feedback, equipment and medium

Abstract

The invention relates to the technical field of big data, provides a metadata management method and device based on semantic feedback, equipment and a medium, is applied to financial and medical health care service scenes, and can reinforce a learning mechanism to initialize a state space, an action space and a reward function based on semantic feedback. The limitation that traditional reinforcement learning only depends on technical indexes is broken through, and alignment of technical optimization and business value is achieved; a metadata autonomous scheduling engine is utilized to generate a current state, a current action is selected from an action space by utilizing a current strategy, and a strategy executor is called to execute the current action, so that a global optimal scheme can be generated in combination with a running state and business semantics, and a local optimization trap is avoided; and in a preset time period, calling a metadata autonomous scheduling engine to perform strategy optimization according to the experience playback buffer area, and executing the next round of metadata management according to a new strategy, so that continuous optimization of the strategy is realized, and long-term change of a service mode can be self-adapted.

Description

Technical Field

[0001] This invention relates to the field of big data technology, and in particular to a method, apparatus, device and medium for metadata management based on semantic feedback. Background Technology

[0002] In modern cloud-native and big data architectures, data volumes are massive and access patterns are complex and varied. Traditional metadata management and query scheduling systems have the following limitations: (1) Static configuration, slow response: The system's caching strategy, index recommendation, data prefetching, query routing and other key parameters mostly rely on static settings made by the administrator based on historical experience. When the business query mode changes (such as from financial statement query to real-time analysis, or the emergence of new financial hot data), the static configuration cannot be adjusted adaptively, resulting in performance degradation, requiring manual readjustment, and high operation and maintenance costs.

[0003] (2) Optimization dimension fragmentation: Traditional optimizers usually only focus on the structure of the query itself, while resource schedulers only focus on physical indicators such as CPU (Central Processing Unit) and memory. The two lack coordination and fail to deeply integrate the semantic information of the data (such as "the types of diseases that are frequently accessed at the same time" or "the high-priority queries of a certain medical department") with the resource allocation strategy, resulting in local optima rather than global optima in optimization decisions.

[0004] (3) Lack of continuous learning and predictive capabilities: The existing system can only react passively to performance problems that have already occurred, and does not have the ability to proactively learn from historical interactions and predict future access patterns. It cannot achieve proactive optimization to "prevent problems before they occur". For example, it cannot preheat relevant data before business peaks.

[0005] Therefore, there is an urgent need for an autonomous metadata management method that can self-awareness, self-decision-making, and continuous self-optimization. Summary of the Invention

[0006] In view of the above, it is necessary to provide a metadata management method, apparatus, device and medium based on semantic feedback, which aims to solve the problem of the inability to automatically and continuously optimize the management of metadata.

[0007] A metadata management method based on semantic feedback, the method comprising: Initialize the state space, action space, and reward function based on a semantic feedback reinforcement learning mechanism; In response to metadata management instructions, the metadata sensing and collecting device is invoked to collect metadata in real time based on the data dimensions of the state space; The current state is generated using the metadata autonomous scheduling engine based on the metadata perception and real-time collection of the collector. Load the current dynamic policy logic of the metadata autonomous scheduling engine as the current policy; Using the current strategy, select the current action from the action space based on the current state; The policy executor is invoked to execute the current action, and after a preset time, the metadata perception and collection device is invoked to collect the new state, and the current reward is calculated according to the reward function; Construct a current experience tuple based on the current state, the current action, the current reward, and the new state, and store the current experience tuple in the experience replay buffer; Within a preset time period, the metadata autonomous scheduling engine is invoked to optimize the strategy based on the experience replay buffer to obtain a new strategy, and the next round of metadata management is executed according to the new strategy.

[0008] A metadata management device based on semantic feedback, the metadata management device based on semantic feedback comprising: An initialization unit is used to initialize the state space, action space, and reward function based on a semantic feedback reinforcement learning mechanism. The acquisition unit is used to respond to metadata management instructions and call the metadata perception and acquisition device to acquire metadata in real time based on the data dimensions of the state space; The generation unit is used to generate the current state based on the metadata perception and the metadata collected in real time by the collector using the metadata autonomous scheduling engine. The loading unit is used to load the current dynamic policy logic of the metadata autonomous scheduling engine as the current policy. The selection unit is used to select a current action from the action space based on the current state using the current strategy; The execution unit is used to call the strategy executor to execute the current action, and after a preset time, call the metadata perception and collector to collect the new state, and calculate the current reward according to the reward function; The construction unit is used to construct a current experience tuple based on the current state, the current action, the current reward, and the new state, and to store the current experience tuple in the experience replay buffer. The optimization unit is used to call the metadata autonomous scheduling engine within a preset time period to optimize the strategy based on the experience replay buffer to obtain a new strategy, and to execute the next round of metadata management based on the new strategy.

[0009] A computer device, the computer device comprising: A memory for storing at least one instruction; and a processor for executing the instructions stored in the memory to implement the semantic feedback-based metadata management method.

[0010] A computer-readable storage medium storing at least one instruction, which is executed by a processor in a computer device to implement the semantic feedback-based metadata management method.

[0011] As can be seen from the above technical solutions, this invention can initialize the state space, action space, and reward function based on the semantic feedback reinforcement learning mechanism, breaking the limitation of traditional reinforcement learning that relies solely on technical indicators, and aligning technical optimization with business value; it calls the metadata perception and collector to collect metadata in real time based on the data dimension of the state space, breaking the information barrier of only collecting technical indicators and lacking business semantics; it uses the metadata autonomous scheduling engine to generate the current state, selects the current action from the action space using the current policy, and calls the policy executor to execute the current action, which can combine the running state and business semantics to generate a globally optimal solution and avoid local optimization traps; within a preset time period, it calls the metadata autonomous scheduling engine to optimize the policy based on the experience replay buffer, and executes the next round of metadata management according to the new policy, realizing continuous policy optimization and adapting to long-term changes in business models. Attached Figure Description

[0012] Figure 1 This is a flowchart of a preferred embodiment of the metadata management method based on semantic feedback of the present invention.

[0013] Figure 2 This is a functional block diagram of a preferred embodiment of the metadata management device based on semantic feedback of the present invention.

[0014] Figure 3 This is a schematic diagram of the structure of a computer device that implements a preferred embodiment of the metadata management method based on semantic feedback according to the present invention. Detailed Implementation

[0015] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0016] like Figure 1 The diagram shown is a flowchart of a preferred embodiment of the semantic feedback-based metadata management method of the present invention. The order of the steps in this flowchart can be changed, and some steps can be omitted, depending on different requirements.

[0017] The semantic feedback-based metadata management method is applied to one or more computer devices. The computer device is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions. Its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0018] The computer device can be any electronic product that can interact with the user, such as a personal computer, tablet computer, smartphone, personal digital assistant (PDA), game console, interactive network television (IPTV), smart wearable device, etc.

[0019] The computer equipment may also include network equipment and / or user equipment. The network equipment includes, but is not limited to, a single network server, a server group consisting of multiple network servers, or a cloud based on cloud computing consisting of a large number of hosts or network servers.

[0020] The server can be a standalone server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.

[0021] Artificial intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0022] Foundational technologies for artificial intelligence generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies mainly encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0023] The network in which the computer device is located includes, but is not limited to, the Internet, wide area network, metropolitan area network, local area network, and virtual private network (VPN).

[0024] S10 initializes the state space, action space, and reward function based on the semantic feedback reinforcement learning mechanism.

[0025] In this embodiment, the semantic feedback reinforcement learning mechanism can deeply integrate the semantic information of metadata with resource allocation strategies.

[0026] In this embodiment, the initialization of the state space, action space, and reward function based on the semantic feedback reinforcement learning mechanism includes: The state space is constructed using basic metadata, performance metadata, and semantic access metadata as data dimensions; wherein, the semantic access metadata includes business semantic context transmitted through upper-layer applications and / or parsed from query tasks; The action space is constructed based on the optimized instructions executable by the Meta-data Autonomous Scheduling Engine (MOE); Define multiple semantic dimensions and corresponding reward / punishment rules for each semantic dimension, and construct the reward function based on the multiple semantic dimensions and corresponding reward / punishment rules for each semantic dimension.

[0027] The basic metadata may include, but is not limited to, basic attributes such as the schema structure of the data table, index configuration, and partition information.

[0028] The performance metadata may include, but is not limited to, operational metrics such as query response time, data transfer throughput, cache hit rate, and server I / O (Input / Output) load.

[0029] The semantic access metadata may include, but is not limited to, business semantics such as tenant information, business project, accessed logical data entity, and query urgency tag extracted from the query statement obtained from the upper-layer application interface or directly parsed.

[0030] Accordingly, the state dimensions within the state space may include current system load, query type distribution, hot data block set, cache usage status, business time period, etc.

[0031] The action space is a set of optimization instructions that the metadata autonomous scheduling engine can execute. For example, the action space may include actions such as: pre-allocating memory for a specified data block, adjusting the priority of the query queue, suggesting the creation of data table indexes, and migrating data shards to high-performance storage.

[0032] The reward function is the core driver of learning. This reward function deeply integrates semantic feedback. For example: a high positive reward is given when a high-priority business-critical query is successfully executed; a high negative reward is given if an interactive query from the "CEO (Chief Executive Officer) Kanban" is delayed due to resource contention; a positive reward is given if learning discovers that "data entity A and data entity B are frequently joined in queries" and proactively caches data entities A and B on the same node, thereby improving the speed of subsequent join queries; and a positive reward is given if a minor background task consumes too many resources, impacting critical business operations, and the MOE imposes resource restrictions on it.

[0033] The above embodiments can break through the limitations of traditional reinforcement learning that relies solely on technical indicators, and achieve alignment between technical optimization and business value.

[0034] S11, in response to the metadata management instruction, invoke the metadata perception and acquisition device to collect metadata in real time based on the data dimension of the state space.

[0035] In this embodiment, the metadata management command can be automatically triggered when the corresponding metadata system starts.

[0036] For example, when the metadata system is a financial metadata system, the collected metadata may include: the table structure of the banking system (corresponding to basic metadata), the response time of customer credit inquiries (corresponding to performance metadata), and queries initiated by "approval departments" with the annotation "urgent review" (corresponding to semantic metadata). When the metadata system is a medical metadata system, the collected metadata may include: the partitioning rules of the electronic medical record data table (such as partitioning by visit time, corresponding to basic metadata), the I / O load of medical record retrieval (corresponding to performance metadata), and queries initiated by "emergency departments" associated with "critical patient treatment" (corresponding to semantic metadata).

[0037] In this embodiment, the metadata sensing and collecting device can adopt a streaming collection architecture to ensure the real-time acquisition of data. At the same time, a data verification mechanism can be set to filter invalid or abnormal data to ensure the accuracy of the metadata snapshot.

[0038] Through the above embodiments, a panoramic data view of system operation status and business access logic can be constructed, providing complete and accurate data input for subsequent scheduling decisions, breaking the information barrier of traditional systems that only collect technical indicators and lack business semantics.

[0039] S12, the current state is generated using the metadata autonomous scheduling engine based on the metadata perception and the metadata collected in real time by the collector.

[0040] In this embodiment, the collected metadata needs to undergo a series of processing steps to obtain the current state.

[0041] Specifically, the step of generating the current state using the metadata autonomous scheduling engine based on the metadata collected in real time by the metadata perception and collector includes: The metadata is converted into a metadata snapshot using the metadata autonomous scheduling engine to obtain the current state.

[0042] S13, Load the current dynamic policy logic of the metadata autonomous scheduling engine as the current policy.

[0043] In this embodiment, the policy logic corresponding to the metadata autonomous scheduling engine is dynamically changing, and the policy is updated through continuous optimization and iteration.

[0044] In this embodiment, the dynamic strategy logic of the metadata autonomous scheduling engine is used to establish a precise mapping between system state and optimization actions, ensuring that the system can select the optimal scheduling action that fits the technical performance and business semantic value under different business scenarios.

[0045] S14, using the current strategy, select the current action from the action space based on the current state.

[0046] In this embodiment, the action space stores multiple candidate actions, and the most suitable action needs to be selected and executed.

[0047] Specifically, selecting the current action from the action space based on the current state using the current strategy includes: The metadata autonomous scheduling engine is used to convert the current state into an input format recognizable by the current policy to obtain the state to be processed; Using the metadata autonomous scheduling engine, the pending state is forward-computed based on the current policy in the action space to obtain the action value corresponding to each action; Select the action with the highest action value from the action space as the current action.

[0048] For example, when using a Deep Q-Network (DQN) as the update strategy in a reinforcement learning algorithm, the DQN outputs the Q-value for each action. If the Q-value for action 1 is 18, the Q-value for action 2 is 15, and the Q-value for action 3 is 12, then action 1 is selected as the current action. When using an Actor-Critic (AC algorithm) as the update strategy in a reinforcement learning algorithm, the Actor network outputs a probability distribution. If the output probability for action 1 is 0.3, the output probability for action 2 is 0.5, and the output probability for action 3 is 0.2, then action 2 is selected as the current action.

[0049] Through the above embodiments, a shift from "passive response" to "proactive decision-making" can be achieved. The scheduling strategy no longer relies on manual rules, but instead generates a globally optimal solution by combining the system's operating status and business semantics, thus avoiding the pitfalls of local optimization.

[0050] S15, the strategy executor is invoked to execute the current action, and after a preset time, the metadata perception and collection device is invoked to collect the new state, and the current reward is calculated according to the reward function.

[0051] In this embodiment, the invocation of the policy executor to execute the current action includes: The strategy executor converts the current action into a control command that can be executed by the underlying data platform, and sends the control command to the underlying data platform.

[0052] The underlying data platform may include a caching system, a query executor, a resource manager, etc.

[0053] Furthermore, a transaction mechanism can be configured to ensure the atomicity and safety of instruction execution and avoid system anomalies caused by execution interruption.

[0054] Among these measures, the reward function can be used as the core criterion to prioritize query needs with high business value.

[0055] For example, in the financial sector, after performing the clearing data pre-caching action, the average latency of clearing queries decreased by 30%, which, due to matching high-priority business needs, can be considered as receiving a positive reward of +15; in the medical sector, after performing the "emergency medical record storage migration" action, the emergency medical record retrieval response time was shortened by 40%, successfully ensuring the efficiency of critical patient treatment, which can be considered as receiving a positive reward of +20.

[0056] The above embodiments can ensure the accurate implementation of scheduling instructions and form a closed-loop feedback of "action-state-reward", providing real experience data support for subsequent strategy iterations.

[0057] S16, construct a current experience tuple based on the current state, the current action, the current reward, and the new state, and store the current experience tuple in the experience replay buffer.

[0058] For example, the current experience tuple can be constructed using the following structure: (current state, current action, current reward, new state).

[0059] The experience replay buffer can be configured with data deduplication and anomaly filtering mechanisms to ensure the validity of the experience data.

[0060] S17. Within a preset time period, the metadata autonomous scheduling engine is invoked to optimize the strategy based on the experience replay buffer to obtain a new strategy, and the next round of metadata management is executed according to the new strategy.

[0061] In this embodiment, the preset time period can be an idle time period.

[0062] Of course, updates can also be performed at a specified period, which can be configured according to the business rules of the metadata.

[0063] In this embodiment, the step of calling the metadata autonomous scheduling engine to optimize the strategy based on the experience replay buffer to obtain a new strategy includes: The metadata autonomous scheduling engine is invoked to randomly sample experience tuples from the experience replay buffer as experience tuples to be processed. The new policy is obtained by iteratively updating the current policy based on the experience tuples to be processed using a reinforcement learning algorithm.

[0064] For example, during system idle periods, empirical data can be randomly sampled from the buffer, and reinforcement learning algorithms such as deep Q-networks or Actor-Critic can be used to iteratively update the decision-making strategy to continuously optimize the action selection logic. For instance, when the system is idle, MOE can use successful experiences to further strengthen the strategy of "proactively reserving resources before business peaks," so that when a similar situation occurs again, the system will more decisively take the same action.

[0065] For example, in the financial sector, if learning from historical experience reveals that "pre-caching data one hour before settlement at the end of each quarter can improve query efficiency by 50%", then this strategy will be automatically triggered at the end of subsequent quarters, and the cache range will be continuously fine-tuned based on new business data. In the medical sector, if learning reveals that "during flu season, setting the cache validity period for emergency medical records to 3 days is better than 1 day in balancing performance and storage resources", then the cache validity period strategy will be automatically updated.

[0066] Through the above embodiments, the strategy can be continuously evolved, forming a fully autonomous closed loop of continuous perception, continuous decision-making, continuous learning, and continuous optimization, so as to have lifelong learning capabilities and be able to adapt to long-term changes in business models without the need for manual intervention in optimization.

[0067] In this embodiment, after executing the next round of metadata management according to the new strategy, the method further includes: When a new service is added, the metadata sensing and collecting device is invoked to collect the metadata of the new service in real time based on the data dimension of the state space; The metadata autonomous scheduling engine is invoked to optimize the strategy based on the metadata of the new service until the dynamic strategy logic of the metadata autonomous scheduling engine is adapted to the metadata of the new service.

[0068] For example, suppose a company launches a new business, resulting in a completely new data access pattern. Traditional static configuration systems require administrators to manually analyze logs and adjust parameters. However, in this embodiment, although MOE may initially experience performance fluctuations due to unfamiliarity with the new pattern, through continuous semantic feedback reinforcement learning loops, MOE can automatically discover the data correlations and resource requirements of the new business within days to weeks, and autonomously evolve new optimization strategies without any manual intervention.

[0069] This embodiment eliminates the need for manual operation and maintenance, replacing tedious and error-prone manual tuning through self-learning, greatly reducing operation and maintenance costs and management complexity. By integrating business semantics (such as tenants and project priorities) into the reward function, it can also make decisions that are more in line with business logic, achieving a leap from "technical indicator optimization" to "business value optimization". It has built a complete technology stack from data semantic fusion and query compilation to resource scheduling, focusing on the most basic resource autonomous management, providing the core data platform with "autonomous driving" capabilities, thereby forming a powerful collaborative defense system.

[0070] As can be seen from the above technical solutions, this invention can initialize the state space, action space, and reward function based on the semantic feedback reinforcement learning mechanism, breaking the limitation of traditional reinforcement learning that relies solely on technical indicators, and aligning technical optimization with business value; it calls the metadata perception and collector to collect metadata in real time based on the data dimension of the state space, breaking the information barrier of only collecting technical indicators and lacking business semantics; it uses the metadata autonomous scheduling engine to generate the current state, selects the current action from the action space using the current policy, and calls the policy executor to execute the current action, which can combine the running state and business semantics to generate a globally optimal solution and avoid local optimization traps; within a preset time period, it calls the metadata autonomous scheduling engine to optimize the policy based on the experience replay buffer, and executes the next round of metadata management according to the new policy, realizing continuous policy optimization and adapting to long-term changes in business models.

[0071] like Figure 2 The diagram shown is a functional block diagram of a preferred embodiment of the metadata management device based on semantic feedback of the present invention. The metadata management device 11 based on semantic feedback includes an initialization unit 110, a collection unit 111, a generation unit 112, a loading unit 113, a selection unit 114, an execution unit 115, a construction unit 116, and an optimization unit 117. The module / unit referred to in this invention refers to a series of computer program segments that can be executed by a processor and perform a fixed function, and are stored in memory. In this embodiment, the functions of each module / unit will be described in detail in subsequent embodiments.

[0072] The initialization unit 110 is used to initialize the state space, action space and reward function based on the semantic feedback reinforcement learning mechanism.

[0073] In this embodiment, the semantic feedback reinforcement learning mechanism can deeply integrate the semantic information of metadata with resource allocation strategies.

[0074] In this embodiment, the initialization unit 110 initializes the state space, action space, and reward function based on the semantic feedback reinforcement learning mechanism, including: The state space is constructed using basic metadata, performance metadata, and semantic access metadata as data dimensions; wherein, the semantic access metadata includes business semantic context transmitted through upper-layer applications and / or parsed from query tasks; The action space is constructed based on the optimized instructions executable by the Meta-data Autonomous Scheduling Engine (MOE); Define multiple semantic dimensions and corresponding reward and punishment rules for each semantic dimension, and construct the reward function based on the multiple semantic dimensions and corresponding reward and punishment rules for each semantic dimension.

[0075] The basic metadata may include, but is not limited to, basic attributes such as the schema structure of the data table, index configuration, and partition information.

[0076] The performance metadata may include, but is not limited to, operational metrics such as query response time, data transfer throughput, cache hit rate, and server I / O (Input / Output) load.

[0077] The semantic access metadata may include, but is not limited to, business semantics such as tenant information, business project, accessed logical data entity, and query urgency tag extracted from the query statement obtained from the upper-layer application interface or directly parsed.

[0078] Accordingly, the state dimensions within the state space may include current system load, query type distribution, hot data block set, cache usage status, business time period, etc.

[0079] The action space is a set of optimization instructions that the metadata autonomous scheduling engine can execute. For example, the action space may include actions such as: pre-allocating memory for a specified data block, adjusting the priority of the query queue, suggesting the creation of data table indexes, and migrating data shards to high-performance storage.

[0080] The reward function is the core driver of learning. This reward function deeply integrates semantic feedback. For example: a high positive reward is given when a high-priority business-critical query is successfully executed; a high negative reward is given if an interactive query from the "CEO (Chief Executive Officer) Kanban" is delayed due to resource contention; a positive reward is given if learning discovers that "data entity A and data entity B are frequently joined in queries" and proactively caches data entities A and B on the same node, thereby improving the speed of subsequent join queries; and a positive reward is given if a minor background task consumes too many resources, impacting critical business operations, and the MOE imposes resource restrictions on it.

[0081] The above embodiments can break through the limitations of traditional reinforcement learning that relies solely on technical indicators, and achieve alignment between technical optimization and business value.

[0082] The acquisition unit 111 is used to respond to metadata management instructions by calling the metadata perception and acquisition device to collect metadata in real time based on the data dimensions of the state space.

[0083] In this embodiment, the metadata management command can be automatically triggered when the corresponding metadata system starts.

[0084] For example, when the metadata system is a financial metadata system, the collected metadata may include: the table structure of the banking system (corresponding to basic metadata), the response time of customer credit inquiries (corresponding to performance metadata), and queries initiated by "approval departments" with the annotation "urgent review" (corresponding to semantic metadata). When the metadata system is a medical metadata system, the collected metadata may include: the partitioning rules of the electronic medical record data table (such as partitioning by visit time, corresponding to basic metadata), the I / O load of medical record retrieval (corresponding to performance metadata), and queries initiated by "emergency departments" associated with "critical patient treatment" (corresponding to semantic metadata).

[0085] In this embodiment, the metadata sensing and collecting device can adopt a streaming collection architecture to ensure the real-time acquisition of data. At the same time, a data verification mechanism can be set to filter invalid or abnormal data to ensure the accuracy of the metadata snapshot.

[0086] Through the above embodiments, a panoramic data view of system operation status and business access logic can be constructed, providing complete and accurate data input for subsequent scheduling decisions, breaking the information barrier of traditional systems that only collect technical indicators and lack business semantics.

[0087] The generation unit 112 is used to generate the current state using the metadata autonomous scheduling engine based on the metadata collected in real time by the metadata perception and collector.

[0088] In this embodiment, the collected metadata needs to undergo a series of processing steps to obtain the current state.

[0089] Specifically, the generation unit 112 uses the metadata autonomous scheduling engine to generate the current state based on the metadata collected in real time by the metadata perception and collector, including: The metadata is converted into a metadata snapshot using the metadata autonomous scheduling engine to obtain the current state.

[0090] The loading unit 113 is used to load the current dynamic policy logic of the metadata autonomous scheduling engine as the current policy.

[0091] In this embodiment, the policy logic corresponding to the metadata autonomous scheduling engine is dynamically changing, and the policy is updated through continuous optimization and iteration.

[0092] In this embodiment, the dynamic strategy logic of the metadata autonomous scheduling engine is used to establish a precise mapping between system state and optimization actions, ensuring that the system can select the optimal scheduling action that fits the technical performance and business semantic value under different business scenarios.

[0093] The selection unit 114 is used to select a current action from the action space based on the current state using the current strategy.

[0094] In this embodiment, the action space stores multiple candidate actions, and the most suitable action needs to be selected and executed.

[0095] Specifically, the selection unit 114 selects the current action from the action space based on the current state using the current strategy, including: The metadata autonomous scheduling engine is used to convert the current state into an input format recognizable by the current policy to obtain the state to be processed; Using the metadata autonomous scheduling engine, the pending state is forward-computed based on the current policy in the action space to obtain the action value corresponding to each action; Select the action with the highest action value from the action space as the current action.

[0096] For example, when using a Deep Q-Network (DQN) as the update strategy in a reinforcement learning algorithm, the DQN outputs the Q-value for each action. If the Q-value for action 1 is 18, the Q-value for action 2 is 15, and the Q-value for action 3 is 12, then action 1 is selected as the current action. When using an Actor-Critic (AC algorithm) as the update strategy in a reinforcement learning algorithm, the Actor network outputs a probability distribution. If the output probability for action 1 is 0.3, the output probability for action 2 is 0.5, and the output probability for action 3 is 0.2, then action 2 is selected as the current action.

[0097] Through the above embodiments, a shift from "passive response" to "proactive decision-making" can be achieved. The scheduling strategy no longer relies on manual rules, but instead generates a globally optimal solution by combining the system's operating status and business semantics, thus avoiding the pitfalls of local optimization.

[0098] The execution unit 115 is used to call the strategy executor to execute the current action, and after a preset time, call the metadata perception and collector to collect the new state, and calculate the current reward according to the reward function.

[0099] In this embodiment, the execution unit 115 invoking the policy executor to execute the current action includes: The strategy executor converts the current action into a control command that can be executed by the underlying data platform, and sends the control command to the underlying data platform.

[0100] The underlying data platform may include a caching system, a query executor, a resource manager, etc.

[0101] Furthermore, a transaction mechanism can be configured to ensure the atomicity and safety of instruction execution and avoid system anomalies caused by execution interruption.

[0102] Among these measures, the reward function can be used as the core criterion to prioritize query needs with high business value.

[0103] For example, in the financial sector, after performing the clearing data pre-caching action, the average latency of clearing queries decreased by 30%, which, due to matching high-priority business needs, can be considered as receiving a positive reward of +15; in the medical sector, after performing the "emergency medical record storage migration" action, the emergency medical record retrieval response time was shortened by 40%, successfully ensuring the efficiency of critical patient treatment, which can be considered as receiving a positive reward of +20.

[0104] The above embodiments can ensure the accurate implementation of scheduling instructions and form a closed-loop feedback of "action-state-reward", providing real experience data support for subsequent strategy iterations.

[0105] The construction unit 116 is used to construct a current experience tuple based on the current state, the current action, the current reward, and the new state, and to store the current experience tuple in the experience replay buffer.

[0106] For example, the current experience tuple can be constructed using the following structure: (current state, current action, current reward, new state).

[0107] The experience replay buffer can be configured with data deduplication and anomaly filtering mechanisms to ensure the validity of the experience data.

[0108] The optimization unit 117 is used to call the metadata autonomous scheduling engine to perform strategy optimization based on the experience replay buffer within a preset time period to obtain a new strategy, and to execute the next round of metadata management based on the new strategy.

[0109] In this embodiment, the preset time period can be an idle time period.

[0110] Of course, updates can also be performed at a specified period, which can be configured according to the business rules of the metadata.

[0111] In this embodiment, the optimization unit 117 calls the metadata autonomous scheduling engine to perform policy optimization based on the experience replay buffer to obtain a new policy, including: The metadata autonomous scheduling engine is invoked to randomly sample experience tuples from the experience replay buffer as experience tuples to be processed. The new policy is obtained by iteratively updating the current policy based on the experience tuples to be processed using a reinforcement learning algorithm.

[0112] For example, during system idle periods, empirical data can be randomly sampled from the buffer, and reinforcement learning algorithms such as deep Q-networks or Actor-Critic can be used to iteratively update the decision-making strategy to continuously optimize the action selection logic. For instance, when the system is idle, MOE can use successful experiences to further strengthen the strategy of "proactively reserving resources before business peaks," so that when a similar situation occurs again, the system will more decisively take the same action.

[0113] For example, in the financial sector, if learning from historical experience reveals that "pre-caching data one hour before settlement at the end of each quarter can improve query efficiency by 50%", then this strategy will be automatically triggered at the end of subsequent quarters, and the cache range will be continuously fine-tuned based on new business data. In the medical sector, if learning reveals that "during flu season, setting the cache validity period for emergency medical records to 3 days is better than 1 day in balancing performance and storage resources", then the cache validity period strategy will be automatically updated.

[0114] Through the above embodiments, the strategy can be continuously evolved, forming a fully autonomous closed loop of continuous perception, continuous decision-making, continuous learning, and continuous optimization, so as to have lifelong learning capabilities and be able to adapt to long-term changes in business models without the need for manual intervention in optimization.

[0115] In this embodiment, after the next round of metadata management is executed according to the new strategy, when there is a new service, the metadata sensing and collecting device is invoked to collect the metadata of the new service in real time based on the data dimension of the state space. The metadata autonomous scheduling engine is invoked to optimize the strategy based on the metadata of the new service until the dynamic strategy logic of the metadata autonomous scheduling engine is adapted to the metadata of the new service.

[0116] For example, suppose a company launches a new business, resulting in a completely new data access pattern. Traditional static configuration systems require administrators to manually analyze logs and adjust parameters. However, in this embodiment, although MOE may initially experience performance fluctuations due to unfamiliarity with the new pattern, through continuous semantic feedback reinforcement learning loops, MOE can automatically discover the data correlations and resource requirements of the new business within days to weeks, and autonomously evolve new optimization strategies without any manual intervention.

[0117] This embodiment eliminates the need for manual operation and maintenance, replacing tedious and error-prone manual tuning through self-learning, greatly reducing operation and maintenance costs and management complexity. By integrating business semantics (such as tenants and project priorities) into the reward function, it can also make decisions that are more in line with business logic, achieving a leap from "technical indicator optimization" to "business value optimization". It has built a complete technology stack from data semantic fusion and query compilation to resource scheduling, focusing on the most basic resource autonomous management, providing the core data platform with "autonomous driving" capabilities, thereby forming a powerful collaborative defense system.

[0118] As can be seen from the above technical solutions, this invention can initialize the state space, action space, and reward function based on the semantic feedback reinforcement learning mechanism, breaking the limitation of traditional reinforcement learning that relies solely on technical indicators, and aligning technical optimization with business value; it calls the metadata perception and collector to collect metadata in real time based on the data dimension of the state space, breaking the information barrier of only collecting technical indicators and lacking business semantics; it uses the metadata autonomous scheduling engine to generate the current state, selects the current action from the action space using the current policy, and calls the policy executor to execute the current action, which can combine the running state and business semantics to generate a globally optimal solution and avoid local optimization traps; within a preset time period, it calls the metadata autonomous scheduling engine to optimize the policy based on the experience replay buffer, and executes the next round of metadata management according to the new policy, realizing continuous policy optimization and adapting to long-term changes in business models.

[0119] like Figure 3 The diagram shown is a structural schematic of a computer device that implements a preferred embodiment of the semantic feedback-based metadata management method of the present invention.

[0120] The computer device 1 may include a memory 12, a processor 13, and a bus (the arrow in the figure represents the bus), and may also include a computer program stored in the memory 12 and executable on the processor 13, such as a metadata management program based on semantic feedback.

[0121] Those skilled in the art will understand that the schematic diagram is merely an example of computer device 1 and does not constitute a limitation on computer device 1. Computer device 1 can be either a bus topology or a star topology. Computer device 1 may also include more or fewer other hardware or software than shown in the diagram, or different component arrangements. For example, computer device 1 may also include input / output devices, network access devices, etc.

[0122] It should be noted that the computer device 1 described is merely an example. Other existing or future electronic products that are adaptable to this invention should also be included within the scope of protection of this invention and are incorporated herein by reference.

[0123] The memory 12 includes at least one type of readable storage medium, such as flash memory, portable hard drive, multimedia card, card-type memory (e.g., SD or DX memory), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 12 can be an internal storage unit of the computer device 1, such as a portable hard drive of the computer device 1. In other embodiments, the memory 12 can be an external storage device of the computer device 1, such as a plug-in portable hard drive, smart media card (SMC), secure digital card (SD), flash card, etc., equipped on the computer device 1. Furthermore, the memory 12 can include both internal and external storage units of the computer device 1. The memory 12 can be used not only to store application software and various types of data installed on the computer device 1, such as the code of a metadata management program based on semantic feedback, but also to temporarily store data that has been output or will be output.

[0124] In some embodiments, the processor 13 may be composed of integrated circuits, such as a single packaged integrated circuit or multiple integrated circuits packaged with the same or different functions, including combinations of one or more central processing units (CPUs), microprocessors, digital processing chips, graphics processors, and various control chips. The processor 13 is the control unit of the computer device 1, connecting various components of the computer device 1 via various interfaces and lines. It performs various functions of the computer device 1 and processes data by running or executing programs or modules stored in the memory 12 (e.g., executing a metadata management program based on semantic feedback), and by accessing data stored in the memory 12.

[0125] The processor 13 executes the operating system of the computer device 1 and various installed applications. The processor 13 executes the applications to implement the steps in the various embodiments of the semantic feedback-based metadata management method described above, for example... Figure 1 The steps are shown.

[0126] For example, the computer program may be divided into one or more modules / units, which are stored in the memory 12 and executed by the processor 13 to complete the present invention. The one or more modules / units may be a series of computer-readable instruction segments capable of performing specific functions, which describe the execution process of the computer program in the computer device 1. For example, the computer program may be divided into an initialization unit 110, a data acquisition unit 111, a generation unit 112, a loading unit 113, a selection unit 114, an execution unit 115, a construction unit 116, and an optimization unit 117.

[0127] The integrated unit implemented as a software functional module described above can be stored in a computer-readable storage medium. This software functional module, stored in a storage medium, includes several instructions to cause a computer device (which may be a personal computer, computer equipment, or network device, etc.) or processor to execute portions of the semantic feedback-based metadata management method described in the various embodiments of this invention.

[0128] If the modules / units integrated in the computer device 1 are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware devices. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above.

[0129] The computer program includes computer program code, which may be in the form of source code, object code, executable file, or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory, etc.

[0130] Furthermore, the computer-readable storage medium may primarily include a stored program area and a stored data area, wherein the stored program area may store the operating system, an application program required for at least one function, etc.; and the stored data area may store data created based on the use of blockchain nodes, etc.

[0131] The blockchain referred to in this invention is a novel application model of computer technologies such as distributed data storage, peer-to-peer transmission, consensus mechanisms, and encryption algorithms. Essentially, a blockchain is a decentralized database, a chain of data blocks linked together using cryptographic methods. Each data block contains information about a batch of network transactions, used to verify the validity of the information (anti-counterfeiting) and generate the next block. A blockchain can include an underlying blockchain platform, a platform product service layer, and an application service layer.

[0132] The bus can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This bus can be divided into address bus, data bus, control bus, etc. For ease of representation, in... Figure 3 The bus is represented by only one straight line, but this does not mean that there is only one bus or one type of bus. The bus is configured to enable communication between the memory 12 and at least one processor 13, etc.

[0133] Although not shown, the computer device 1 may also include a power supply (such as a battery) to power various components. Preferably, the power supply can be logically connected to the at least one processor 13 through a power management device, thereby enabling functions such as charging management, discharging management, and power consumption management. The power supply may also include one or more DC or AC power supplies, recharging devices, power fault detection circuits, power converters or inverters, power status indicators, and other arbitrary components. The computer device 1 may also include various sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be described in detail here.

[0134] Furthermore, the computer device 1 may also include a network interface. Optionally, the network interface may include a wired interface and / or a wireless interface (such as a Wi-Fi interface, a Bluetooth interface, etc.), which is typically used to establish communication connections between the computer device 1 and other computer devices.

[0135] Optionally, the computer device 1 may further include a user interface, which may be a display, an input unit (such as a keyboard), and optionally, a standard wired interface or a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen, etc. The display may also be appropriately referred to as a screen or display unit, used to display information processed in the computer device 1 and to display a visual user interface.

[0136] It should be understood that the embodiments described are for illustrative purposes only and are not limited to this structure in the scope of the patent application.

[0137] It will be understood by those skilled in the art that Figure 3 The structure shown does not constitute a limitation on the computer device 1, and may include fewer or more components than shown, or combine certain components, or have different component arrangements.

[0138] Combination Figure 1 The memory 12 in the computer device 1 stores multiple instructions to implement a metadata management method based on semantic feedback, and the processor 13 can execute the multiple instructions to achieve: Initialize the state space, action space, and reward function based on a semantic feedback reinforcement learning mechanism; In response to metadata management instructions, the metadata sensing and collecting device is invoked to collect metadata in real time based on the data dimensions of the state space; The current state is generated using the metadata autonomous scheduling engine based on the metadata perception and real-time collection of the collector. Load the current dynamic policy logic of the metadata autonomous scheduling engine as the current policy; Using the current strategy, select the current action from the action space based on the current state; The policy executor is invoked to execute the current action, and after a preset time, the metadata perception and collection device is invoked to collect the new state, and the current reward is calculated according to the reward function; Construct a current experience tuple based on the current state, the current action, the current reward, and the new state, and store the current experience tuple in the experience replay buffer; Within a preset time period, the metadata autonomous scheduling engine is invoked to optimize the strategy based on the experience replay buffer to obtain a new strategy, and the next round of metadata management is executed according to the new strategy.

[0139] Specifically, the processor 13's implementation method for the above instructions can be found in [reference needed]. Figure 1 The descriptions of the relevant steps in the corresponding embodiments are not repeated here.

[0140] It should be noted that all data involved in this case was legally obtained. Software tools or components not belonging to this company that appear in the embodiments of this application are merely illustrative examples and do not represent actual use.

[0141] In the several embodiments provided by this invention, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of modules is only a logical functional division, and other division methods may be used in actual implementation.

[0142] This invention can be used in a wide variety of general-purpose or special-purpose computer system environments or configurations. Examples include: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, and distributed computing environments including any of the above systems or devices. This invention can be described in the general context of computer-executable instructions, such as program modules, that are executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform specific tasks or implement specific abstract data types. This invention can also be practiced in distributed computing environments where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0143] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0144] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in the form of hardware plus software functional modules.

[0145] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0146] Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be embraced within the invention. No appended diagram markings in the claims should be construed as limiting the scope of the claims.

[0147] Furthermore, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices described in this invention can also be implemented by a single unit or device through software or hardware. Terms such as "first," "second," etc., are used to indicate names and do not indicate any specific order.

[0148] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A metadata management method based on semantic feedback, characterized in that, The metadata management method based on semantic feedback includes: Initialize the state space, action space, and reward function based on a semantic feedback reinforcement learning mechanism; In response to metadata management instructions, the metadata sensing and collecting device is invoked to collect metadata in real time based on the data dimensions of the state space; The current state is generated using the metadata autonomous scheduling engine based on the metadata perception and real-time collection of the collector. Load the current dynamic policy logic of the metadata autonomous scheduling engine as the current policy; Using the current strategy, select the current action from the action space based on the current state; The policy executor is invoked to execute the current action, and after a preset time, the metadata perception and collection device is invoked to collect the new state, and the current reward is calculated according to the reward function; Construct a current experience tuple based on the current state, the current action, the current reward, and the new state, and store the current experience tuple in the experience replay buffer; Within a preset time period, the metadata autonomous scheduling engine is invoked to optimize the strategy based on the experience replay buffer to obtain a new strategy, and the next round of metadata management is executed according to the new strategy.

2. The metadata management method based on semantic feedback as described in claim 1, characterized in that, The initialization of the state space, action space, and reward function based on the semantic feedback reinforcement learning mechanism includes: The state space is constructed using basic metadata, performance metadata, and semantic access metadata as data dimensions; wherein, the semantic access metadata includes business semantic context transmitted through upper-layer applications and / or parsed from query tasks; The action space is constructed based on the optimized instructions executable by the metadata autonomous scheduling engine; Define multiple semantic dimensions and corresponding reward / punishment rules for each semantic dimension, and construct the reward function based on the multiple semantic dimensions and corresponding reward / punishment rules for each semantic dimension.

3. The metadata management method based on semantic feedback as described in claim 1, characterized in that, The process of generating the current state using the metadata autonomous scheduling engine based on the metadata collected in real time by the metadata sensing and collecting device includes: The metadata is converted into a metadata snapshot using the metadata autonomous scheduling engine to obtain the current state.

4. The metadata management method based on semantic feedback as described in claim 1, characterized in that, The step of selecting a current action from the action space based on the current state using the current strategy includes: The metadata autonomous scheduling engine is used to convert the current state into an input format recognizable by the current policy to obtain the state to be processed; Using the metadata autonomous scheduling engine, the pending state is forward-computed based on the current policy in the action space to obtain the action value corresponding to each action; Select the action with the highest action value from the action space as the current action.

5. The metadata management method based on semantic feedback as described in claim 1, characterized in that, The invocation of the policy executor to perform the current action includes: The strategy executor converts the current action into a control command that can be executed by the underlying data platform, and sends the control command to the underlying data platform.

6. The metadata management method based on semantic feedback as described in claim 1, characterized in that, The step of calling the metadata autonomous scheduling engine to optimize the strategy based on the experience replay buffer to obtain a new strategy includes: The metadata autonomous scheduling engine is invoked to randomly sample experience tuples from the experience replay buffer as experience tuples to be processed. The new policy is obtained by iteratively updating the current policy based on the experience tuples to be processed using a reinforcement learning algorithm.

7. The metadata management method based on semantic feedback as described in claim 1, characterized in that, After executing the next round of metadata management according to the new strategy, the method further includes: When a new service is added, the metadata sensing and collecting device is invoked to collect the metadata of the new service in real time based on the data dimension of the state space; The metadata autonomous scheduling engine is invoked to optimize the strategy based on the metadata of the new service until the dynamic strategy logic of the metadata autonomous scheduling engine is adapted to the metadata of the new service.

8. A metadata management device based on semantic feedback, characterized in that, The metadata management device based on semantic feedback includes: An initialization unit is used to initialize the state space, action space, and reward function based on a semantic feedback reinforcement learning mechanism. The acquisition unit is used to respond to metadata management instructions and call the metadata perception and acquisition device to acquire metadata in real time based on the data dimensions of the state space; The generation unit is used to generate the current state based on the metadata perception and the metadata collected in real time by the collector using the metadata autonomous scheduling engine. The loading unit is used to load the current dynamic policy logic of the metadata autonomous scheduling engine as the current policy. The selection unit is used to select a current action from the action space based on the current state using the current strategy; The execution unit is used to call the strategy executor to execute the current action, and after a preset time, call the metadata perception and collector to collect the new state, and calculate the current reward according to the reward function; The construction unit is used to construct a current experience tuple based on the current state, the current action, the current reward, and the new state, and to store the current experience tuple in the experience replay buffer. The optimization unit is used to call the metadata autonomous scheduling engine within a preset time period to optimize the strategy based on the experience replay buffer to obtain a new strategy, and to execute the next round of metadata management based on the new strategy.

9. A computer device, characterized in that, The computer device includes: A memory for storing at least one instruction; and a processor for executing the instructions stored in the memory to implement the semantic feedback-based metadata management method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores at least one instruction, which is executed by a processor in a computer device to implement the metadata management method based on semantic feedback as described in any one of claims 1 to 7.

Images