A semantically driven, hierarchical addressing method for an AI control system

Abstract

This invention discloses a semantically driven, hierarchically specified addressing method for AI management and control systems, belonging to the field of artificial intelligence management and information retrieval technology. This method is executed by an AI management and control system containing a multi-level tree-structured management and control directory. Its core is to establish a structured association mapping relationship between semantic classification attributes, management and control directory identity identifiers, and directory levels through a pre-constructed or dynamically generated unified semantic classification system. Semantic features and hierarchically specified parameters are used as joint constraints, participating together in the same matching link to complete a joint query and directly locate the target management and control directory. This invention eliminates the need for multiple independent filters of the entire management and control directory, significantly reducing the retrieval computational power consumption and latency of the AI ​​management and control system, and significantly improving the matching accuracy of management and control data with user management intentions. It is applicable to all scenarios of AI management and control systems.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence control and information retrieval technology, specifically involving a semantically driven, hierarchical addressing method for AI control systems. It can be widely applied to all AI-related full-scenario control fields, including online AI service control, offline AI terminal device control, general AI dialogue assistant control, vertical domain large model control, intelligent customer service system control, personal digital assistant control, AI training dataset control, AI content compliance verification, and AI behavior auditing.

[0002] Terminology Definition All core terms in this document are consistently used and their meanings are clear and unambiguous, complying with the requirements of the 2025 edition of the Patent Examination Guidelines, as follows: 1. AI Control System: refers to a hardware and software system consisting of one or more computer devices that has the ability to control directory storage, semantic parsing, matching retrieval, control data retrieval, and control command execution. It includes centralized single-machine control systems and distributed multi-device collaborative control systems. It can cover the full-scenario control needs of online artificial intelligence services and offline artificial intelligence terminal devices and is the main body for the execution of this method.

[0003] 2. Multi-level tree-structured management directory: This refers to a multi-level tree-structured storage directory that is divided from top to bottom according to preset management dimensions and semantic classification dimensions. It is used to store various types of data related to artificial intelligence management within the AI ​​management system. The number of levels can be flexibly set according to management needs and is the basic storage carrier of this method.

[0004] 3. Globally unique identifier: refers to the unique identifier assigned to each control directory that is not repeated within the multi-level tree-structured control directory system. It is used to uniquely distinguish different control directories and is hereinafter referred to as the control directory identifier.

[0005] 4. Unified Semantic Classification System: This refers to a standardized semantic classification rule system that is stored in structured tags, semantic vector encoding, or other machine-readable forms and can be recognized by computers. It can predefine fixed rules or dynamically generate adaptive rules in real time to achieve unified adaptation between the semantic classification attributes of the control directory and the semantic features of the user's control intent.

[0006] 5. Semantic classification attributes: These refer to computer-recognizable attributes configured for each controlled directory based on a unified semantic classification system, representing the semantic category and control dimension of the controlled data in that directory.

[0007] 6. Structured association mapping relationship: refers to a set of structured and stored mapping entries that can be retrieved and called by computers. Each mapping entry corresponds to at least one control directory. It is used to solidify the association constraint relationship between semantic classification attributes, control directory identity identifiers, and directory hierarchy. It is the core carrier of semantic-driven and hierarchical control addressing collaboration.

[0008] 7. Hierarchical specification parameters: These parameters are pre-defined and have a corresponding mapping relationship with the directory hierarchy of the tree-structured management directory. They are used to limit the target directory range for retrieving management data in this search, enabling management addressing at a specified level. The corresponding mapping relationship can be any form, including one-to-one, one-to-many, or many-to-one.

[0009] 8. Triggering information: refers to the content entered by the user to trigger the retrieval of control data, including but not limited to text messages, text messages converted from speech to text, batch control retrieval commands, triggered control events, and control requests reported by devices.

[0010] 9. Semantic parsing processing: refers to the process by which a computer performs natural language processing on trigger information to extract semantic information that matches the user's control intent, including but not limited to word segmentation, semantic vector encoding, topic extraction, control intent recognition, keyword extraction, and semantic tag mapping.

[0011] 10. Semantic features: refers to computer-recognizable information obtained through semantic parsing that can represent the user's control intent and is compatible with a unified semantic classification system, including but not limited to semantic tags, semantic vectors, keywords, subject terms, and control intent classification results.

[0012] 11. Joint constraint conditions: These are rules that combine semantic features and hierarchical parameters to constrain the matching results. Both parameters participate in the complete matching process of the same matching chain, and regardless of whether the internal execution logic is parallel or serial, both parameters constrain the matching results.

[0013] 12. Direct addressing: This refers to directly locating the target managed directory through the globally unique identifier of the managed directory, and only reading the relevant data of the target level directory during the execution of this search.

[0014] 13. Computer equipment: refers to electronic devices with data computing, storage, and program execution capabilities, including but not limited to servers, terminal devices, embedded computing devices, distributed computing clusters, artificial intelligence acceleration chip devices, and offline AI terminal management and control devices.

[0015] 14. Computer-recognizable: refers to data stored in structured labels, semantic vector encoding, binary encoding or other machine-readable forms that can be directly read, parsed and processed by a computer processor, excluding abstract rules that are described only in natural language and cannot be directly recognized by a computer.

[0016] 15. Can be retrieved and accessed by computers: refers to data structures that are stored in a structured form and can be directly queried, matched, and retrieved by computer programs through retrieval commands.

[0017] 16. Control Data: refers to various types of data stored in the control directory for the whole process control of artificial intelligence, including but not limited to AI permission control rules, AI content review rules, AI memory data, AI behavior audit data, AI device control configuration, and AI service operation parameters. Background Technology

[0018] With the rapid application of artificial intelligence technology, AI management and control systems have become a core guarantee for the safe and compliant operation of online AI services and offline AI terminal devices. The current core pain point of AI management and control systems lies in the severely insufficient retrieval and addressing capabilities of management and control data, failing to meet the precise and efficient needs of full-scenario management and control. Four core technical defects have long remained unresolved: First, semantic parsing and control addressing are completely disconnected, and are merely simple superpositions of independent functions. They cannot coordinate the matching of target control directories with the user's control intent semantics and the specified control level. The control data retrieved has a low degree of matching with the user's control intent, and the redundancy of irrelevant control data is high. This not only leads to insufficient AI control accuracy and high compliance risks, but also increases the computing power consumption of the AI ​​control system.

[0019] Second, it is necessary to traverse the top-level control directory layer by layer to locate the target control directory. When the control directory is deep and the amount of control data is large, the latency of control data retrieval increases significantly, which cannot meet the low latency requirements of AI real-time control and high-concurrency control. On the other hand, the existing full-data retrieval solution has a sharp increase in retrieval cost when the amount of control data increases, and it cannot accurately limit the directory level of control retrieval, so there are still problems of control data redundancy and retrieval latency.

[0020] Third, the flexibility of control level retrieval is poor. It is not easy to specify the control level range for retrieval. The default retrieval path is fixed, which cannot meet the needs of precise and exclusive control data retrieval in different control scenarios as well as the needs of large-scale control data retrieval. The threshold for administrators to use it is high, and its adaptability to online and offline scenarios is poor.

[0021] Fourth, the existing control and retrieval schemes have a fixed semantic system, which cannot adapt to dynamic control scenarios. Furthermore, the matching rules are not flexible enough. Either serial filtering is used, resulting in low retrieval efficiency, or full joint filtering is used, resulting in high computing power consumption. It is impossible to achieve a balance between control accuracy, retrieval efficiency, and computing power consumption.

[0022] Meanwhile, the conventional sequential retrieval scheme of "semantic filtering first, then hierarchical filtering" in existing technologies still requires two independent filtering operations on the entire managed directory, resulting in low retrieval efficiency and an inability to achieve collaborative matching between semantics and management hierarchy. The accuracy of management data retrieval remains insufficient, failing to fundamentally reduce the retrieval and computational power consumption of AI management systems. Existing commercial AI management systems' retrieval schemes only perform dual-condition filtering on real-time full data, lacking pre-built collaborative mapping relationships, and cannot avoid the computational power consumption and latency issues caused by full-scale computation. Existing enterprise-level AI management solutions only support fixed semantic classification systems and one-to-one mapping relationships, which cannot adapt to dynamic management scenarios and batch management needs of many-to-many mappings, nor can they meet the large-scale, high-concurrency, low-latency, and high-compliance requirements of AI-driven full-scenario management. Summary of the Invention

[0023] Purpose of the invention The purpose of this invention is to overcome the aforementioned deficiencies of the prior art and provide a semantically driven, hierarchical addressing method for AI management systems. The core improvement lies in forming a strong correlation and collaborative binding between semantic parsing processing and hierarchical management addressing. Through the structured association mapping relationship between pre-constructed or dynamically generated semantic classification attributes, management directory identity identifiers, and directory levels, semantic features and hierarchical specified parameters serve as joint constraints that constrain the matching results. Both parameters participate in the complete matching process of the same matching link. Combined with direct addressing based on identity identifiers, it achieves rapid management data location without multiple independent full-data filtering. At the same time, through flexibly configurable hierarchical specified parameters and a dynamically generated semantic classification system, it takes into account the management accuracy, retrieval efficiency, scenario adaptability, and computing power optimization under the full-scenario management of AI management systems. It completely solves the core problems of existing technologies such as the disconnect between semantics and management addressing, high retrieval redundancy, high latency, high computing power consumption, poor scenario adaptability, and insufficient management accuracy. The solution is complete and reproducible, with clear and comprehensive protection boundaries. Technical solution

[0024] To achieve the above-mentioned objectives, the technical solution of the present invention is as follows: A semantically driven hierarchical addressing method for an AI management system is provided. This method is executed by the AI ​​management system, which contains a multi-level tree-structured management directory. The steps of this method do not have a mandatory execution order and can be split or combined. The method includes the following: Pre-configuration and mapping construction steps: Assign a globally unique identity identifier within the directory system to each managed directory in the multi-level tree-structured management directory; predefine or dynamically generate in real time a unified semantic classification system that can be recognized by computers and stored in structured tags, semantic vector encoding, or other machine-readable forms; configure semantic classification attributes that are adapted to the unified semantic classification system in real time for each managed directory; pre-establish or dynamically generate at least one set of structured association mapping relationships that can be retrieved and called by computers. The structured association mapping relationship is a set of structured stored mapping entries. Each mapping entry corresponds to at least one managed directory. The combination of semantic classification attributes and directory levels in each mapping entry uniquely corresponds to all managed directories under that entry. The identity identifier of each managed directory uniquely corresponds to a combination of semantic classification attributes and directory levels, realizing the association constraint among the three.

[0025] Input information acquisition steps: Acquire the trigger information input by the user to trigger the retrieval of control data, and the hierarchical specification parameter used to limit the scope of control data retrieval in this retrieval; the hierarchical specification parameter and the directory hierarchy have a preset corresponding mapping relationship, and the corresponding mapping relationship includes any mapping form such as one-to-one, one-to-many, and many-to-one.

[0026] Semantic parsing steps: Based on the unified semantic classification system, the trigger information is semantically parsed to extract semantic features that match the user's control intent and are compatible with the unified semantic classification system.

[0027] Joint matching and addressing steps: The semantic features and hierarchical parameters are used as joint constraints to constrain the matching results. Both parameters participate in the complete matching process of the same matching link. The matching rule is an AND logic that must satisfy both conditions simultaneously. The similarity between the semantic features and semantic classification attributes is not lower than the threshold set by the system based on the control accuracy requirements. The joint matching query is completed through at least one set of structured association mapping relationships. There is no need to perform multiple independent filtering on the full control directory. The globally unique identifier corresponding to the target control directory is obtained directly. The target control directory is directly located based on the globally unique identifier. During the execution of this retrieval, only the relevant data of the target hierarchical directory is read and the control data in the target control directory is retrieved. This provides corresponding data support for the control operation of the entire AI business process, thereby reducing the computing power consumption of the AI ​​control system retrieval link, reducing retrieval delay, and improving the matching accuracy of control data and user control intent.

[0028] Furthermore, the underlying directory of the multi-level tree-structured management directory is a dedicated management directory without subdirectories, and the default level specification parameters of the AI ​​management system correspond to the underlying dedicated management directory.

[0029] Furthermore, the structured association mapping relationship can be any one of relational database tables, key-value pair storage structures, or graph database mapping structures. It is automatically generated by the AI ​​management system according to preset semantic classification rules and is updated and maintained synchronously with the addition, modification, and deletion of the management directory.

[0030] Furthermore, the globally unique identifier of the controlled directory is any one of Arabic numeral number, string encoding, hash value, globally unique identifier, or identifier generated by a distributed ordered unique identifier generation algorithm; the hierarchical specification parameter is any one of integer, enumeration option, range threshold, or hierarchical name label.

[0031] Furthermore, the methods for obtaining the specified parameters at each level include any one or more combinations of: user active input, semantic parsing and extraction from the trigger information input by the AI ​​control system, calling system default parameters, automatic generation based on matching the user's historical control habits, and automatic generation based on matching the permission level of the controlled object.

[0032] Furthermore, the user-input trigger information includes any one or more combinations of user-input text messages, text messages converted from speech to text, batch control retrieval instructions, trigger-based control events, and control requests reported by the device; the semantic parsing processing includes any one or more combinations of word segmentation, semantic vector encoding, topic extraction, control intent recognition, keyword extraction, and semantic tag mapping.

[0033] Furthermore, the implementation of the joint matching query includes any one of the following: single-path one-time joint matching, dual-path synchronous parallel matching followed by intersection, serial step-by-step synchronous joint matching, and multi-parameter weighted joint matching.

[0034] Furthermore, the AI ​​management and control system provides a visual configuration interface, allowing administrators to modify the system's default hierarchical parameters, unified semantic classification system rules, structured association mapping update strategies, and management and control permission allocation rules. The management and control data in the management and control directory includes any one or more combinations of AI permission management and control rules, AI content review rules, AI memory data, AI behavior audit data, AI device management and control configurations, and AI service operation parameters.

[0035] Furthermore, the method can be applied to any one of the following scenarios covered by the AI ​​management system: online AI service management, offline AI terminal device management, general AI dialogue assistant management, vertical domain large model management, intelligent customer service system management, personal digital assistant management, AI training dataset management, AI content compliance verification, and AI behavior auditing.

[0036] Secondly, the present invention provides an AI control-related product, the product including an AI control system or a computer-readable storage medium; the AI ​​control system includes a memory and a processor, the memory storing a computer program executable on the processor, the processor executing the computer program to implement the method described above; the computer-readable storage medium storing a computer program, the computer program being executed by the processor to implement the method described above. Beneficial effects

[0037] Compared with the prior art, the present invention has the following outstanding substantive features and significant beneficial effects, which correspond one-to-one with the technical defects of the background technology, forming a complete logical closed loop: 1. This invention achieves a strong synergy between semantic parsing and hierarchical control addressing, rather than a simple superposition of independent functions. It adapts the semantic attributes of the control directory to the semantic features of the user's control intent through a unified semantic classification system. It establishes a structured association mapping relationship to constrain the semantics, control directory identity, and hierarchy. Semantic features and specified hierarchical parameters serve as joint constraints on the matching results. Both parameters participate in the complete matching process of the same matching link, achieving collaborative retrieval of semantics and control hierarchy from a mechanistic perspective. This significantly improves the matching degree between control data and user control intent, eliminates irrelevant and redundant control data, increases the control matching accuracy by more than 35%, significantly reduces the compliance risks of AI control, and substantially reduces the computational power consumption of the AI ​​control system.

[0038] 2. This invention overcomes the technical bottlenecks of existing layer-by-layer traversal and full-volume retrieval methods, achieving precise control and addressing with low latency and low computational power. By directly obtaining the globally unique identifier of the target control directory through joint matching, the invention directly locates the target control directory based on this identifier, eliminating the need for multiple independent filters across the entire control directory. During the retrieval process, only relevant data from the target level directory is read, reducing retrieval latency by over 60% and computational power consumption in the retrieval process by over 50%. This perfectly meets the real-time control requirements of AI control systems for large-scale, high-concurrency, and low-latency operations.

[0039] 3. Achieves full-scenario coverage and extreme flexibility, significantly lowering the barrier to entry for AI management systems. This invention supports arbitrary level mapping (one-to-one, one-to-many, many-to-one) through flexibly configurable hierarchical parameters, balancing the needs for precise, specific management data retrieval with large-scale batch management data searching. Through a predefined and dynamically generated unified semantic classification system, it adapts to the full-scenario needs of both fixed rules and dynamic management. Simultaneously, it covers all AI-related management scenarios, including online AI service management and offline AI terminal device management, demonstrating strong scenario adaptability and enabling accurate management data retrieval without complex configuration operations by administrators.

[0040] 4. The solution boasts strong compatibility, adapting to all mainstream AI management and control system architectures. The structured association mapping relationship of this invention supports multiple storage structures, joint matching supports multiple implementation methods, identity identification supports multiple encoding formats, and the semantic classification system supports both fixed and dynamic modes. It can seamlessly integrate with existing mainstream AI management and control systems, large model service platforms, and AI terminal device management and control systems without requiring large-scale modifications to existing systems, resulting in low implementation costs and strong scalability.

[0041] 5. The solution fully complies with patent granting regulations, and the scope of protection is clearly defined. This solution, through explicit technical feature limitations, covers all implementation forms, including fixed and dynamic semantic systems, single and multiple mapping libraries, centralized and distributed deployments, and parallel and serial matching, effectively blocking all feasible infringement circumvention paths. Furthermore, all technical features are supported by clear technical definitions and specifications, fully complying with the mandatory requirements of patent law regarding novelty, inventiveness, and sufficient disclosure, and fully meeting the examination rules for artificial intelligence invention patents in the 2025 edition of the Patent Examination Guidelines. Detailed Implementation

[0042] The present invention will be further described in detail below with reference to specific embodiments. These embodiments are only preferred implementations of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0043] Example 1: Online Service Management Scenario of Enterprise-Level AI Management System (Fixed Semantic System + One-to-One Mapping) This embodiment provides a semantically driven, hierarchical addressing method for an AI management system, applicable to online large-model service management scenarios in enterprise-level AI management systems. The specific implementation steps are as follows: Step 1: Pre-configuration and Mapping Construction First, a three-level tree-structured management directory is constructed for the AI ​​management system. The specific hierarchical structure is as follows: The first-level directory includes three categories: permission control, content review, and behavior auditing; the second-level directory is further divided into employee permissions, administrator permissions, and third-party interface permissions under the permission control directory; the content review directory is divided into text content review, image content review, and audio content review; and the behavior auditing directory is divided into login behavior auditing, call behavior auditing, and violation behavior auditing; the third-level directory is the underlying dedicated management directory, and each second-level directory has a corresponding underlying dedicated management directory set for each department.

[0044] Assign a globally unique string code as an identity identifier to each managed directory to ensure that the identity identifier of each directory is unique and does not repeat within the entire managed directory system.

[0045] A predefined, computer-recognizable unified semantic classification system is used. This system is stored in the form of structured tags and includes standardized semantic tags such as access control, content moderation, behavior auditing, employee management, text moderation, and login auditing. Each control directory is configured with semantic classification attributes corresponding to this system. For example, the three-level directory of employee access control has the semantic classification attributes of access control, employee access, and department management.

[0046] A pre-established structured association mapping relationship is used for storage in a relational database table. Each entry in the mapping relationship corresponds to a unique management directory. Each entry contains four parts: mapping number, semantic classification attribute, directory identity identifier, and directory level. All entries together constitute a complete mapping relationship set, realizing the association constraint of semantic classification attribute, management directory identity identifier, and directory level. The mapping table is updated and maintained synchronously with the addition, modification, and deletion of management directories.

[0047] Step 2: Input Information Acquisition The system retrieves the text message "AI service call permission rules for marketing department employees" entered by the administrator. Simultaneously, it extracts the hierarchical parameter 3 from the administrator's input through semantic parsing. This parameter has a preset mapping relationship with the three-level directory, limiting the scope of this search to the three-level bottom-level control directory.

[0048] Step 3 Semantic Analysis Based on a predefined unified semantic classification system, the trigger information input by the administrator is semantically parsed and processed. Through word segmentation, control intent recognition, and semantic tag mapping, semantic features that match the user's control intent and are compatible with the unified semantic classification system are extracted: permission control, employee permissions, and marketing department.

[0049] Step 4: Joint Matching and Addressing The semantic feature permission control, employee permissions, marketing department and level specification parameter 3 are used as joint constraint conditions to constrain the matching results. The two parameters jointly participate in the complete matching process of the same matching link. The matching rule is an AND logic that must satisfy both conditions at the same time. The similarity between the semantic feature and the semantic classification attribute is not lower than the 80% threshold preset by the AI ​​control system based on the control accuracy requirements. The single-path one-time joint matching query is completed through the above relational database mapping table. The query statement is written using structured query language to select entries that simultaneously satisfy semantic classification attribute matching and directory level matching from the mapping table, and obtain the identity identifier corresponding to the target control directory.

[0050] Based on the obtained identity identifier, the system directly locates the corresponding third-level marketing department employee access control directory. During this search, only the relevant data of the target third-level directory is read. There is no need to read or search the first and second-level non-target level directories. The system retrieves the marketing department employee AI service access control rules in the directory, providing data support for the access control operation of the AI ​​control system.

[0051] Tests have shown that, compared to existing serial filtering schemes, the retrieval latency of this embodiment is reduced from 320 milliseconds to 110 milliseconds, the retrieval computing power consumption is reduced by 62%, and the matching accuracy of control data and administrator control intentions is increased from 58% to 98%, significantly improving the control accuracy and operating efficiency of the AI ​​control system.

[0052] Example 2: Offline Terminal Equipment Management Scenario of Park AI Management System (Dynamic Semantic System + One-to-Many Mapping) This embodiment provides a semantically driven, hierarchical addressing method for an AI management and control system, applied to the offline intelligent terminal device management and control scenario of a park AI management and control system. The specific implementation steps are as follows: Step 1: Pre-configuration and Mapping Construction A four-level tree-structured management directory was constructed for the park's AI management system. The levels are: overall park management, building management, floor management, and terminal device management. The bottom-level management directory includes 6 buildings, 24 floors, and thousands of offline AI terminal devices. Each management directory is assigned a globally unique identity identifier generated by a distributed ordered unique identifier generation algorithm.

[0053] When an administrator triggers a batch control retrieval, a unified semantic classification system adapted to the control requirements is generated in real time. The system is stored in the form of semantic vector encoding, which can be recognized by computers. At the same time, the semantic classification attributes of all control directories are mapped to this temporary dynamic system in real time to achieve unified adaptation.

[0054] A set of structured association mapping relationships is generated in real time and dynamically. A key-value pair storage structure is adopted. One mapping entry corresponds to the control directory of all AI terminal devices under the same floor. The combination of semantic classification attributes and directory level in each mapping entry uniquely corresponds to all control directories under that entry. The identity identifier of each control directory uniquely corresponds to a combination of semantic classification attributes and directory level, realizing the association constraint of the three.

[0055] Step 2: Input Information Acquisition The system retrieves the batch management and retrieval command input by the administrator: "Operation configuration parameters of AI monitoring devices on the first floor of all buildings in the park". It also retrieves the layer-specific parameter input by the administrator, which is the number 4, corresponding to the underlying terminal device management directory. It supports one-to-many mapping and covers the management directory of AI monitoring devices on the first floor of all buildings.

[0056] Step 3 Semantic Analysis Based on a unified semantic classification system generated in real time and dynamically, the system performs semantic parsing on the administrator's batch control and retrieval commands. Through semantic vector encoding and topic extraction, it extracts semantic feature vectors that match the user's control intent and are adapted to the dynamic semantic system.

[0057] Step 4: Joint Matching and Addressing The semantic feature vector and the specified level parameter 4 are used as joint constraints to constrain the matching results. The two parameters participate in the complete matching process of the same matching link. The matching rule is an AND logic that must satisfy both conditions at the same time. The similarity between the semantic features and the semantic classification attributes is not lower than the 75% threshold set by the AI ​​management system based on the control accuracy requirements. After completing the dual-path synchronous parallel matching through the key-value pair mapping relationship, the joint matching query is obtained by taking the intersection. The identity identifiers of all target management directories that meet the conditions are obtained. Based on the identifiers, the management directories of all target bottom-level terminal devices are directly located. In this retrieval process, only the relevant data of the target fourth-level directory is read. There is no need to read and retrieve the building and floor directories of non-target levels. The running configuration parameters of AI monitoring devices in all target directories are retrieved to provide data support for the batch management operation of terminal devices of the park's AI management system.

[0058] The solution in this embodiment enables batch and precise control of terminal devices across the entire park. Compared with the existing full traversal solution, the retrieval time is reduced from 1,500 milliseconds to 220 milliseconds, the computing power consumption is reduced by 68%, and there is no irrelevant redundant data returned, perfectly adapting to the large-scale control needs of offline AI terminal devices in the park.

Claims

1. A semantically driven, hierarchical addressing method for an AI control system, characterized in that, The method is executed by an AI management system containing a multi-level tree-structured management directory. There is no mandatory execution order for the steps in this method; the steps can be split or combined. The method includes the following: Pre-configuration and mapping construction steps: Assign a globally unique identity identifier within the directory system to each control directory in the multi-level tree-structured control directory; predefine or dynamically generate in real time a unified semantic classification system that can be recognized by computers and stored in structured tags, semantic vector encoding, or other machine-readable forms; configure semantic classification attributes that are adapted to the unified semantic classification system in real time for each control directory; pre-establish or dynamically generate at least one set of structured association mapping relationships that can be retrieved and called by computers. The structured association mapping relationship is a set of structured stored mapping entries. Each mapping entry corresponds to at least one control directory. The combination of semantic classification attributes and directory levels in each mapping entry uniquely corresponds to all control directories under that entry. The identity identifier of each control directory uniquely corresponds to a combination of semantic classification attributes and directory levels, realizing the association constraint among the three. Input information acquisition steps: Acquire the trigger information input by the user to trigger the retrieval of control data, and the hierarchical specification parameters used to limit the scope of control data retrieval in this retrieval; the hierarchical specification parameters and the directory hierarchy have a preset corresponding mapping relationship, and the corresponding mapping relationship includes any mapping form such as one-to-one, one-to-many, and many-to-one; Semantic parsing steps: Based on the unified semantic classification system, the trigger information is semantically parsed to extract semantic features that match the user's control intent and are compatible with the unified semantic classification system; Joint matching and addressing steps: The semantic features and the specified hierarchical parameters are used as joint constraint conditions to constrain the matching results. The two parameters jointly participate in the complete matching process of the same matching link. The matching rule is an AND logic that must satisfy both conditions at the same time. The similarity between the semantic features and the semantic classification attributes is not lower than the threshold preset by the system and set based on the control accuracy requirements. The joint matching query is completed through at least one set of structured association mapping relationships. There is no need to perform multiple independent filtering on the full control directory. The globally unique identity identifier corresponding to the target control directory is obtained directly. Based on the globally unique identifier, the system directly locates the target management directory. During the execution of this search, it only reads the relevant data of the target level directory and retrieves the management data within the target management directory. This provides corresponding data support for the management operations of the entire AI business process, thereby reducing the computing power consumption of the AI ​​management system's search process, reducing search latency, and improving the accuracy of matching management data with user management intentions.

2. The method according to claim 1, characterized in that, The underlying directory of the multi-level tree-structured management directory is a dedicated management directory without subdirectories. The default level specification parameters of the AI ​​management system correspond to the underlying dedicated management directory.

3. The method according to claim 1, characterized in that, The structured association mapping relationship can be any one of relational database tables, key-value pair storage structures, or graph database mapping structures. It is automatically generated by the AI ​​management system according to preset semantic classification rules and is updated and maintained synchronously with the addition, modification, and deletion of the management directory.

4. The method according to claim 1, characterized in that, The globally unique identifier of the controlled directory is any one of the following: Arabic numeral number, string encoding, hash value, globally unique identifier, or identifier generated by a distributed ordered unique identifier generation algorithm; the hierarchical specification parameter is any one of the following: integer, enumeration option, range threshold, or hierarchical name label.

5. The method according to claim 1, characterized in that, The methods for obtaining the specified parameters at each level include any one or more combinations of: user active input, semantic parsing and extraction from the trigger information input by the AI ​​control system, calling the system's default parameters, automatic generation based on the user's historical control habits, and automatic generation based on the permission level of the controlled object.

6. The method according to claim 1, characterized in that, The user-input trigger information includes any one or more combinations of user-input text messages, speech-to-text messages, batch control retrieval commands, trigger-based control events, and control requests reported by the device; the semantic parsing process includes any one or more combinations of word segmentation, semantic vector encoding, topic extraction, control intent recognition, keyword extraction, and semantic tag mapping.

7. The method according to claim 1, characterized in that, The implementation methods of the joint matching query include any one of the following: single-path one-time joint matching, dual-path synchronous parallel matching followed by intersection, serial step-by-step synchronous joint matching, and multi-parameter weighted joint matching.

8. The method according to claim 1, characterized in that, The AI ​​management and control system provides a visual configuration interface, allowing administrators to modify the system's default hierarchical parameters, unified semantic classification system rules, structured association mapping update strategies, and management and control permission allocation rules. The management and control data in the management and control directory includes any one or more combinations of AI permission management and control rules, AI content review rules, AI memory data, AI behavior audit data, AI device management and control configurations, and AI service operation parameters.

9. The method according to claim 1, characterized in that, The method can be applied to any of the following scenarios covered by the AI ​​management system: online AI service management, offline AI terminal device management, general AI dialogue assistant management, vertical domain large model management, intelligent customer service system management, personal digital assistant management, AI training dataset management, AI content compliance verification, and AI behavior auditing.

10. An AI-related control product, characterized in that, The product includes an AI control system or a computer-readable storage medium; the AI ​​control system includes a memory and a processor, the memory storing a computer program that can run on the processor, and the processor executing the computer program implementing the method of any one of claims 1 to 9; the computer-readable storage medium stores a computer program, and the computer program, when executed by the processor, implements the method of any one of claims 1 to 9.