Perceptual Meaning Space Coordinate System Based on Triple-Layer Fusion and Method for Knowledge Retrieval and Settlement Using the Same

Abstract

The present invention relates to a triple-layer fusion-based perceptual semantic space coordinate system, comprising: a user terminal including a display unit and a user input unit; and a server connected to the user terminal via a wired or wireless network.The system includes a communication unit, a storage unit, and a control unit. The control unit receives text or a query input from a user terminal and generates a semantic block through the operation of an LLM-based block generation unit. It generates metadata for the semantic block, including blocks, relationships, timestamps, and probabilities. It generates a NAQL query through the operation of an NAQL query generation unit to define semantic, structural, and temporal associations between the semantic blocks. It defines, connects, updates, or expands relationships between the semantic blocks according to the NAQL query. It performs the operation of a multidimensional error verification unit on the semantic blocks, relationships, timestamps, probabilities, and NAQL execution results to identify semantic inconsistencies, structural errors, chronological errors, or probability distribution anomalies. Based on the verification results, it generates and executes a transaction including the creation, modification, connection, or disposal of semantic blocks. The transaction is processed to maintain atomicity, consistency, and sequentiality. It collects the frequency of errors occurring during the transaction processing, the NAQL execution failure rate, and the transaction rollback rate to perform the operation of a system monitoring unit. The system monitoring unit analyzes the collected information based on anomaly detection to ensure system stability. It is characterized by generating an indicator, dynamically adjusting semantic block generation conditions, NAQL execution conditions, multidimensional error verification criteria, or transaction processing order according to the above stability indicator, and controlling the accumulation of semantic blocks, relationships, timestamps, probabilities, and transaction change history in a storage unit to ensure quality management and traceability, thereby ensuring that semantic block generation, NAQL-based linking, multidimensional error verification, transaction processing, and quality management are consistently performed within a single perceptual semantic space.

Description

Technology Field

[0001] The present invention relates to the field of information processing technology combining artificial intelligence, natural language processing, knowledge representation, and knowledge management technology, and in particular to technology for structurally representing and utilizing the meaning of knowledge in an artificial intelligence system including a large-scale language model and a knowledge graph.

[0002] The present invention relates to a technology for representing and searching the semantic location of knowledge units in a multidimensional continuous space in an artificial intelligence-based knowledge management system. More specifically, the invention relates to a system and method for generating a PMST Vector by weightedly fusing Text Embedding, Graph Embedding, and Manifold Geometry, and managing this by mapping it 1:1 with an AI-readable Meaning Address URI, thereby performing semantic-based search, tracking semantic changes by version, and knowledge contribution-based settlement when generating AI responses.

[0003] Furthermore, the present invention belongs to a technical field that enables semantic-centered knowledge search, knowledge version management, and tracking of knowledge utilization during the AI ​​response generation process—which are difficult to achieve with simple keyword matching or single-vector-based representation methods—by quantitatively expressing and managing semantic relationships, structural relationships, and patterns of temporal change between knowledge units. Additionally, the present invention can be applied to a wide range of AI application technologies that enable the systematic management of semantic changes and contributions occurring during the processes of knowledge creation, modification, linking, and utilization. Background Technology

[0004] With the recent advancement of artificial intelligence technology centered on Large-Scale Language Models (LLM), the methods of generating, managing, and utilizing text-based knowledge are changing rapidly. In particular, as the amount of knowledge accumulated in unstructured or semi-structured forms—such as documents, policy data, technical literature, and conversation records—increases, there is a growing need for technologies that can structure this knowledge into a form understandable by AI and enable its search and utilization based on semantics.

[0005] In this regard, existing knowledge management methods have primarily utilized text embeddings to represent documents or sentences in vector form and perform searches or classifications based on similarity between vectors. However, this approach has limitations in that it is difficult to adequately reflect structural relationships between knowledge units, semantic changes over time, or contexts where multiple pieces of knowledge are combined.

[0006] Furthermore, while technologies utilizing knowledge graphs enable structural representation by explicitly defining entities and relationships, limitations exist in scalability and real-time updates in large-scale knowledge environments because the graph structure is often fixed or relies on manual definitions. Moreover, knowledge graphs alone have limitations in quantitatively representing subtle semantic similarities or differences in semantic intensity between knowledge units, or patterns of semantic change resulting from the modification or expansion of knowledge.

[0007] Meanwhile, in the process of generating AI responses using large-scale language models, multiple knowledge units are referenced in combination, and factors such as the relationships between knowledge units, the timing of generation, and reliability directly affect the response quality. However, conventional technology has faced the problem of being difficult to systematically verify and track which knowledge units were used during the response generation process, the extent to which each knowledge unit contributed to the response result, or whether the connection structure between knowledge units is appropriate.

[0008] Furthermore, in environments where the creation, modification, and linking of knowledge occur repeatedly, problems such as semantic inconsistencies, structural errors, distortion of chronological order, or the accumulation of unreliable knowledge may arise; however, systematic verification and management measures to pre-verify or automatically correct these issues have not been sufficiently provided. In particular, if such errors accumulate, the reliability of the entire knowledge management system deteriorates and can negatively impact the consistency and accuracy of AI responses.

[0009] Accordingly, there is an increasing demand for technologies that structure knowledge units into semantic blocks, manage the relationships, creation times, and reliability between each semantic block, and flexibly define and search the connection structures between knowledge based on queries. Furthermore, there is a need for integrated knowledge management technologies capable of multidimensionally verifying semantic, structural, and temporal relationships between knowledge units, managing the knowledge creation and linkage processes on a transactional basis based on verification results, and continuously monitoring and improving the stability and quality of the entire system. The problem to be solved

[0010] The present invention is proposed to solve the problems of the prior art as described above, and its main objective is to resolve the limitations in semantic representation, the lack of representation of structural relationships, and the difficulty in tracking semantic changes over time that arise from treating knowledge units as simple text or single vectors in an AI-based knowledge management environment.

[0011] Specifically, the present invention aims to solve the problem where conventional methods relying on text embeddings or fixed knowledge graph structures make it difficult to simultaneously consider semantic similarity, structural association, and semantic change based on the time of creation or modification between knowledge units, which results in semantically close knowledge being separated or, conversely, knowledge with different meanings being mixed within the same domain.

[0012] Furthermore, the present invention aims to solve the problem that, despite multiple knowledge units being referenced in combination during the process of generating AI responses using large-scale language models, it is difficult in conventional technology to clearly track or manage which knowledge unit each response was generated based on and to what extent each knowledge unit contributed to the response result. Through this, the invention aims to resolve issues such as reduced consistency in response quality, difficulty in analyzing the causes of errors, and reduced reliability regarding the utilization of knowledge.

[0013] Furthermore, the present invention aims to overcome the limitations of conventional technology, which cannot verify and manage in advance problems such as semantic inconsistencies, structural errors, distortion of chronological order, or the accumulation of unreliable knowledge that may occur during the repetitive process of knowledge creation, modification, and linking. In particular, a key objective is to address the issue where the reliability and stability of the entire knowledge management system deteriorate when such errors accumulate.

[0014] Furthermore, the present invention aims to resolve the management complexity and error propagation issues arising from the separation of relationship definition, verification, and updating between knowledge units into individual steps, and to provide an integrated structure capable of managing the entire process—from the creation of knowledge units to relationship definition, verification, transaction processing, and quality control—by organically combining them.

[0015] In other words, the objective of the present invention is to fundamentally improve the reliability, consistency, and usability of an AI-based knowledge management system by structuring knowledge units into semantic blocks, managing the relationships, creation times, and reliability of each semantic block together, and simultaneously performing semantic-based querying and searching, multidimensional error verification, transaction-based management, and system stability monitoring in an integrated manner. means of solving the problem

[0016] A triple-layer fusion-based perceptual semantic space coordinate system according to one embodiment of the present invention comprises: a user terminal including a display unit and a user input unit; and a server connected to the user terminal via a wired or wireless network.The system includes a communication unit, a storage unit, and a control unit. The control unit receives text or a query input from a user terminal and generates a semantic block through the operation of an LLM-based block generation unit. It generates metadata for the semantic block, including blocks, relationships, timestamps, and probabilities. It generates a NAQL query through the operation of an NAQL query generation unit to define semantic, structural, and temporal associations between the semantic blocks. It defines, connects, updates, or expands relationships between the semantic blocks according to the NAQL query. It performs the operation of a multidimensional error verification unit on the semantic blocks, relationships, timestamps, probabilities, and NAQL execution results to identify semantic inconsistencies, structural errors, chronological errors, or probability distribution anomalies. Based on the verification results, it generates and executes a transaction including the creation, modification, connection, or disposal of semantic blocks. The transaction is processed to maintain atomicity, consistency, and sequentiality. It collects the frequency of errors occurring during the transaction processing, the NAQL execution failure rate, and the transaction rollback rate to perform the operation of a system monitoring unit. The system monitoring unit analyzes the collected information based on anomaly detection to ensure system stability. It is characterized by generating an indicator, dynamically adjusting semantic block generation conditions, NAQL execution conditions, multidimensional error verification criteria, or transaction processing order according to the above stability indicator, and controlling the accumulation of semantic blocks, relationships, timestamps, probabilities, and transaction change history in a storage unit to ensure quality management and traceability, thereby ensuring that semantic block generation, NAQL-based linking, multidimensional error verification, transaction processing, and quality management are consistently performed within a single perceptual semantic space.

[0017] The above control unit is characterized by generating semantic blocks through an LLM-based block generation unit by decomposing input text into context units, forming semantic blocks based on semantic cohesion between decomposed contexts, assigning a probability to each semantic block, and setting the probability by reflecting the contextual match and reference possibility at the time of semantic block generation.

[0018] The above control unit is characterized by applying constraints including relationship type between semantic blocks, relationship direction, timestamp condition, and probability condition to a NAQL query generated through an NAQL query generation unit, and by configuring the NAQL query to restrict the search target to only paths between semantic blocks that satisfy the above constraints.

[0019] The above-described multidimensional error verification unit is characterized by setting the semantic block structure, NAQL query structure, timestamp consistency, and probability distribution as independent verification dimensions, and determining whether an error exists by synthesizing the verification results for each verification dimension.

[0020] The above control unit is characterized by controlling to roll back or replace with a modification transaction a transaction containing the corresponding semantic block or NAQL query when it is identified as an error as a result of multidimensional error verification.

[0021] The above control unit is characterized by connecting the semantic block generation, NAQL execution, multidimensional error verification, and transaction execution steps into a single transaction flow, and recording the execution results of each step based on timestamps.

[0022] The above-described system monitoring unit is characterized by generating system stability indicators by analyzing the error occurrence frequency, NAQL execution failure rate, and transaction rollback rate based on anomaly detection, and adjusting the semantic block generation conditions or multidimensional error verification criteria of the control unit according to the stability indicators.

[0023] The above control unit is characterized by performing a self-improving function that operates to reduce the possibility of errors occurring during subsequent semantic block generation or NAQL execution by using semantic blocks, relationships, timestamps, probabilities, and the change history of NAQL execution results as training data.

[0024] The above control unit is characterized by identifying a valid semantic block based on a probability comparison result when multiple versions with different timestamps exist for the same semantic block, and disabling or managing the remaining semantic blocks as reference-only.

[0025] The above control unit is characterized by tracking referenced semantic blocks, NAQL queries, and transaction information regarding response results provided to a user terminal, storing them in a quality management log, and managing semantic reliability and response consistency based on the quality management log.

[0026] In a triple-layer fusion-based perceptual semantic space coordinate system, a user terminal comprising a display unit and a user input unit; and a server connected to the user terminal via a wired or wireless network; wherein the server includes a communication unit, a storage unit, and a control unit; the control unit receives text or a query input from the user terminal and generates a semantic block through the operation of an LLM-based block generation unit, generates metadata including blocks, relationships, timestamps, and probabilities for the semantic block, and generates a NAQL query through the operation of an NAQL query generation unit to define semantic, structural, and temporal associations between the semantic blocks, wherein the NAQL query is configured to include complex constraints including relationship types, relationship directions, timestamp conditions, and probability conditions between the semantic blocks, and wherein the complex constraints restrict traversable paths on the semantic block graph according to structural, temporal, and probabilistic conditions; the control unit defines, connects, updates, or expands relationships between semantic blocks using the NAQL query, records semantic block relationship information including NAQL execution results in the storage unit, and controls the semantic block, relationships, timestamps, probabilities, and NAQL execution results so as to be utilized as reference data for subsequent transaction processing, verification, or quality control, It is characterized by ensuring that NAQL constraint-based semantic block search and relationship management are performed consistently within a single perceptual semantic space.

[0027] In a triple-layer fusion-based perceptual semantic space coordinate system, a user terminal comprising a display unit and a user input unit; The system comprises a server connected to the user terminal via a wired or wireless network, wherein the server includes a communication unit, a storage unit, and a control unit, wherein the control unit receives text or a query input from the user terminal and generates a semantic block through the operation of an LLM-based block generation unit, generates metadata including a block, relationship, timestamp, and probability for the semantic block, maps the semantic block, relationship, and NAQL execution result to coordinates in a single perceptual semantic space, and the coordinates in the perceptual semantic space are configured to be movable according to the creation, modification, linking, or disposal of the semantic block, wherein the coordinate movement occurs in conjunction with changes in relationships, timestamps, or probability changes between semantic blocks, and the control unit identifies the degree of semantic change by comparing changes in distance or position between coordinates at different points in time corresponding to the same semantic block, wherein the degree of semantic change is used as a criterion for determining whether to maintain, update, branch, or dispose of the semantic block, and controls the creation and execution of transactions for the semantic block or relationship according to the determination result, thereby ensuring that the consistency and quality of the semantic structure are managed based on changes in the perceptual semantic space coordinates. Effects of the invention

[0028] According to the present invention, by structuring knowledge units into semantic block units and managing each semantic block together with metadata including blocks, relationships, timestamps, and probabilities, the problems of incomplete semantic representation and semantic mixing that occurred in conventional text embeddings or single-vector-based knowledge management methods can be effectively improved. Accordingly, semantic-based knowledge management that considers not only semantic similarity between knowledge units but also structural associations and temporal changes becomes possible.

[0029] Furthermore, the present invention provides the effect of flexibly establishing and expanding relationships between knowledge without relying on a knowledge graph with a fixed structure by defining and exploring connection relationships between semantic blocks through NAQL queries. In particular, through a constraint-based search structure including relationship type, relationship direction, timestamp condition, and probability condition, only semantically valid and reliable knowledge paths can be selectively utilized, thereby improving the accuracy and consistency of knowledge search and query response.

[0030] Furthermore, by performing multidimensional error verification targeting semantic blocks, relationships, timestamps, and probabilities, the present invention can identify and manage semantic inconsistencies, structural errors, or chronological distortions that may occur during the knowledge generation and linking process in advance. As a result, it can effectively prevent the conventional problem where the reliability of the entire knowledge management system is degraded due to the accumulation of errors.

[0031] Furthermore, the present invention manages the processes of creating, modifying, linking, and discarding knowledge on a transaction basis, and ensures consistency and stability throughout the entire knowledge management process by performing rollback or modification transactions based on verification results. This transaction-based management structure minimizes error propagation during the knowledge modification process and enables the rapid recovery from abnormal states that occur during system operation.

[0032] Furthermore, the present invention collects error occurrence frequency, NAQL execution failure rate, and transaction rollback rate, and generates system stability indicators through anomaly detection-based analysis, thereby enabling the quantitative assessment of the operational status of a knowledge management system. Accordingly, semantic block generation conditions, NAQL execution conditions, or verification criteria can be dynamically adjusted, allowing the stability and quality of the system to be continuously maintained even in long-term operating environments.

[0033] Furthermore, the present invention provides the effect of gradually reducing the possibility of errors occurring during the knowledge management and response generation processes through a self-learning function that utilizes semantic blocks, relationships, timestamps, probabilities, and the change history of NAQL execution results as training data. As a result, as the system is operated repeatedly, it is expected that the quality of knowledge management and the reliability of AI responses will improve.

[0034] Meanwhile, according to the present invention, since the semantic blocks referenced and their contributions can be tracked during the process of generating artificial intelligence responses using a large-scale language model, the basis for generating the response result can be clearly identified. Accordingly, response quality management, error cause analysis, and knowledge utilization history management become easier, and there is an effect of being able to implement a management or settlement structure based on contributions per knowledge unit.

[0035] In summary, the present invention provides the effect of fundamentally improving the reliability, consistency, and usability of an AI-based knowledge management system by organically combining semantic block-based structuring, NAQL-based relationship exploration, multidimensional error verification, transaction management, and system stability monitoring. Brief explanation of the drawing

[0036] FIG. 1 is a configuration diagram showing the overall architecture of a triple-layer fusion-based perceptual semantic space coordinate system according to the present invention, namely a PMST (Perceptual Meaning Space Theory) system, illustrating a five-layer structure of an input layer, an embedding layer, a fusion layer, a bridge layer, and a service layer. FIG. 2 is a flowchart illustrating a triple-layer fusion process that generates an integrated PMST vector by fusing three layers of text embeddings, graph embeddings, and manifold characteristics in a PMST system according to the present invention. FIG. 3 is a flowchart illustrating the detailed process of an optimization loop for generating PMST vectors according to the present invention, showing an iterative optimization structure including global optimization, application of local constraints, stability evaluation, and convergence determination. FIG. 4 is a diagram showing the structure and index configuration of a bridge table according to the present invention, illustrating a structure in which a triple mapping relationship between a semantic address (URI), a physical address (OID), and a PMST vector is managed in a single table. FIG. 5 is a configuration diagram showing the 1:1 mapping relationship between semantic address (MAS URI), physical address (OID), and semantic coordinate (PMST Vector) and the integrated management structure through a bridge table in a PMST system according to the present invention. Figure 6 is a flowchart illustrating the flow of query PMST vector generation and Top-k search being performed during the semantic search process according to the present invention. FIG. 7 is a flowchart illustrating a semantic change tracking process by version, which calculates the amount of semantic change between versions of the same knowledge unit in a PMST system according to the present invention and classifies the processing method according to the amount of change. FIG. 8 is a flowchart illustrating an AI response contribution settlement process that calculates the contribution of knowledge units that contributed to the generation of AI responses in the PMST system according to the present invention, and calculates the settlement amount per author based thereon. Figure 9 is a comparative configuration diagram showing the structural differences between the PMST triple fusion method according to the present invention and the existing single vector method, as well as the limitations and advantages of each method. FIG. 10 is a conceptual diagram showing the semantic cluster structure formed by knowledge units in the PMST semantic space according to the present invention, semantic path tracing between clusters, and various search modes. FIG. 11 is a configuration diagram showing a multi-layer cache architecture for high-performance response in a PMST system according to the present invention, as well as the hit rate and latency for each layer. Figure 12 shows the performance benchmark results of a PMST system according to the present invention, and is a performance analysis diagram that comprehensively illustrates latency, throughput, cache efficiency, and improvement rate compared to the existing system. Specific details for implementing the invention

[0037] Embodiments of the present invention will be described in detail below with reference to the accompanying drawings. In the drawings, identical reference numbers or symbols refer to components that perform substantially the same function, and the size of each component in the drawings may be exaggerated for clarity and convenience of explanation. However, the technical concept of the present invention and its core components and operations are not limited only to the components or operations described in the following embodiments. In describing the present invention, if it is determined that a detailed description of known technologies or components related to the present invention may unnecessarily obscure the essence of the invention, such detailed description will be omitted.

[0038] In embodiments of the present invention, terms including ordinal numbers, such as first, second, etc., are used solely for the purpose of distinguishing one component from another, and singular expressions include plural expressions unless the context clearly indicates otherwise. Furthermore, in embodiments of the present invention, terms such as 'composed of,' 'include,' 'have,' etc., should be understood as not precluding the existence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof. Additionally, in embodiments of the present invention, 'module' or 'part' performs at least one function or operation and may be implemented in hardware or software, or as a combination of hardware and software, or may be integrated into at least one module and implemented as at least one processor. Furthermore, in embodiments of the present invention, 'at least one' among a plurality of elements refers not only to all of the plurality of elements but also to each individual element excluding the remainder or all combinations thereof. Additionally, "configured to" may be used interchangeably with, depending on the context, for example, "suitable for," "having the capacity to," "designed to," "adapted to," "made to," or "capable of." "Configured to" does not necessarily mean that it is "specifically designed to" in hardware. Instead, in some situations, the expression "device configured to" may mean that the device is "capable of" doing so in conjunction with other devices or components.For example, the phrase “a processor configured (or set) to perform A, B, and C” may mean a dedicated processor for performing said operations (e.g., an embedded processor), or a generic-purpose processor (e.g., a CPU or an application processor) capable of performing said operations by executing one or more software programs stored in a memory device.

[0039] Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the drawings. This description is intended to be detailed enough for a person skilled in the art to easily practice the invention, and it should be noted that the technical scope and concept of the present invention are not limited thereby.

[0040] FIG. 1 is a configuration diagram showing the overall architecture of a triple-layer fusion-based perceptual semantic space coordinate system, i.e., a PMST (Perceptual Meaning Space Theory) system, according to the present invention, illustrating a five-layer structure consisting of an input layer, an embedding layer, a fusion layer, a bridge layer, and a service layer; FIG. 2 is a flowchart showing the triple-layer fusion process of generating an integrated PMST vector by fusing three layers of text embeddings, graph embeddings, and manifold characteristics in the PMST system according to the present invention; FIG. 3 is a flowchart showing the detailed process of an optimization loop for generating a PMST vector according to the present invention, illustrating an iterative optimization structure including global optimization, application of local constraints, stability evaluation, and convergence determination; FIG. 4 is a diagram showing the structure and index configuration of a bridge table according to the present invention, illustrating a structure in which the triple mapping relationship between semantic addresses (URI), physical addresses (OID), and PMST vectors is managed in a single table; FIG. 5 is a 1:1 between semantic addresses (MAS URI), physical addresses (OID), and semantic coordinates (PMST Vector) in the PMST system according to the present invention FIG. 6 is a configuration diagram showing an integrated management structure through mapping relationships and bridge tables; FIG. 6 is a flowchart showing the flow in which query PMST vector generation and Top-k search are performed during the semantic search process according to the present invention; FIG. 7 is a flowchart showing a version-specific semantic change tracking process in which the amount of semantic change between versions of the same knowledge unit is calculated in the PMST system according to the present invention, and the processing method is classified according to the amount of change; FIG. 8 is a flowchart showing an AI response contribution settlement process in which the contribution of knowledge units that contributed to the generation of AI responses is calculated in the PMST system according to the present invention, and the settlement amount per author is calculated based thereon; FIG. 9 is a comparative configuration diagram showing the structural differences between the PMST triple fusion method according to the present invention and the existing single vector method, as well as the limitations and advantages of each method.FIG. 10 is a conceptual diagram showing the semantic cluster structure formed by knowledge units in the PMST semantic space according to the present invention, semantic path tracing between clusters, and various search modes; FIG. 11 is a configuration diagram showing a multi-layer cache architecture for high-performance response in the PMST system according to the present invention, as well as hit rates and latency for each layer; FIG. 12 is a performance analysis diagram showing the performance benchmark results of the PMST system according to the present invention, comprehensively illustrating latency, throughput, cache efficiency, and improvement rate compared to the existing system.

[0041] Referring to FIGS. 1 to 12, a triple-layer fusion-based perceptual semantic space coordinate system and a knowledge search and settlement method using the same according to an embodiment of the present invention comprises: a user terminal including a display unit and a user input unit; and a server connected to the user terminal via a wired or wireless network, wherein the server includes a communication unit, a storage unit, and a control unit. This configuration is based on a general system configuration that allows the input, processing, search, and settlement of knowledge to be performed in a distributed manner in a user environment and a server environment.

[0042] The above-described user terminal includes a display unit and a user input unit. The display unit is configured to visually provide information generated or processed within a server or user terminal to a user, and may display text, shapes, graphs, result lists, or other visual elements. The display unit may include a screen provided on a smartphone, tablet, laptop, desktop computer, wearable device, kiosk, or other electronic display means, and is not limited by its implementation method, resolution, output format, etc.

[0043] A user input unit is configured to receive user commands, queries, or data inputs and may include a keyboard, touchscreen, mouse, stylus, voice input device, gesture recognition means, or a similar input interface. Information input through the user input unit may be a knowledge search request, a query sentence, a control command, a setting value, or other input reflecting user intent, and the format or content of the input data is not limited in a specific manner.

[0044] The above user terminal is connected to a server via a wired or wireless network. The wired or wireless network may include the Internet, a mobile communication network, a local area wireless communication network, a dedicated network, or a combination thereof, and is not limited by a communication method or protocol. Through this, data transmission and reception, request forwarding, and return of processing results can be performed between the user terminal and the server.

[0045] The above server includes a communication unit, a storage unit, and a control unit. The communication unit is a component responsible for transmitting and receiving data with a user terminal or an external system, and performs the function of transmitting or receiving data packets over a wired or wireless network. The communication unit may be implemented as a network interface card, a communication module, a software communication stack, or a combination thereof, and is not limited to a specific communication protocol or method.

[0046] A storage unit is a component for storing data, processing results, configuration information, or other operational information necessary for system operation. The storage unit may store data input by users, data generated by servers, intermediate processing data, or data requiring long-term retention. The storage unit may include volatile or non-volatile memory, databases, file systems, distributed storage structures, or combinations thereof, and is not limited by storage structure or physical implementation method.

[0047] The control unit is a component that controls the overall operations performed within the server, controls the processing flow of data received through the communication unit, controls data storage and retrieval for the storage unit, and executes computation and control logic to enable the system to perform intended functions. The control unit may be implemented by one or more processors, arithmetic units, or logic control means, and may be configured in the form of software, firmware, or a combination of hardware. The control operations performed by the control unit are not limited to specific algorithms or processing methods and may include various control flows depending on system requirements.

[0048] Accordingly, a system according to one embodiment of the present invention is configured based on a role-sharing structure between a user terminal and a server so that user input and output functions and communication, storage, and control functions on the server side are organically linked. This configuration is intended to describe a general system environment for implementing the technical concept of the present invention, and various modifications are possible regarding the specific form, quantity, arrangement, or implementation method of individual components.

[0049] A control unit according to an embodiment of the present invention receives text or a query input from a user terminal and generates a semantic block through the operation of an LLM-based block generation unit. The control unit performs the function of receiving text or a query input from a user terminal. The text or query may include sentences entered by the user in natural language form, sets of keywords, parts of documents, query sentences, input in the form of commands, or similar character-based inputs, and is not limited by input format, length, or type of language. The control unit may receive input data received from the user terminal through a communication unit or acquire it through direct data linkage with the user terminal.

[0050] The control unit processes the received text or query as is, or, if necessary, normalizes, splits, or reconstructs it into a form suitable for internal processing and transmits it to the LLM-based block generation unit. At this time, the control unit can identify whether the input data is a single continuous text, a query composed of multiple sentences, or a query that has continuity with the previous input, and the result of this identification can be utilized in the subsequent processing flow.

[0051] The LLM-based block generator interprets input text or queries using a large-scale language model and generates semantic blocks by extracting or constructing semantic units contained within the input. Here, a semantic block may be a unit that represents the entire input text or query as is, or it may be the result of breaking down the input into contextual sub-units, conceptual units, argument units, or information units. The generation criteria, decomposition levels, or combination methods of semantic blocks are not limited to a specific method and can be configured in various ways depending on the system's purpose or operating environment.

[0052] The control unit controls or triggers the operation of the LLM-based block generation unit to generate one or more semantic blocks corresponding to the input text or query. In this process, the control unit may control the generation of multiple semantic blocks for the same input or the generation of a single semantic block depending on the nature of the input, and the setting of boundaries between semantic blocks or the generation order may also be included within the control range of the control unit.

[0053] The generated semantic block is utilized as a unit to structurally represent the meaning inherent in the input text or query, and can be the subject of relationship definition, storage, validation, retrieval, or other subsequent processing in subsequent stages. The control unit can control the management of these semantic blocks as an internal data structure or their recording in a storage unit, and can transmit the fact that a semantic block has been generated to subsequent processing stages.

[0054] Accordingly, a control unit according to one embodiment of the present invention performs a basic step for structuring user input into semantic units by generating semantic blocks through the operation of an LLM-based block generation unit based on text or a query input from a user terminal. This operation is intended to explain the basic processing flow for implementing the technical concept of the present invention, and various modifications are possible for the input processing method, the method of defining semantic blocks, or the generation procedure.

[0055] A control unit according to one embodiment of the present invention generates metadata including a block, a relationship, a timestamp, and a probability for the semantic block.

[0056] According to one embodiment of the present invention, the metadata is information for identifying or managing the semantic block itself, and is configured to include a block, a relationship, a timestamp, and a probability. The control unit may assign the metadata after the semantic block is created or simultaneously with its creation, and the timing or order of metadata creation is not limited to a specific method.

[0057] Here, a block is unit information for identifying semantic blocks and may include identifiers, classification information, or structural units to distinguish individual semantic blocks from other semantic blocks. Block information may be defined in various ways depending on the creation order, group affiliation, logical distinction, or management purpose of the semantic blocks, and its method of representation or format is not restricted.

[0058] Relationships are metadata representing the association between one semantic block and another, and can express semantic associations, structural connections, reference relationships, or dependency relationships. The control unit can define relationships between semantic blocks as unidirectional or bidirectional, and multiple relationships may be established simultaneously for a single semantic block. This relationship information can be utilized in subsequent processes of connecting, searching, or managing semantic blocks.

[0059] A timestamp is information indicating the point in time when a semantic block or its corresponding metadata is created or updated. The control unit may record at least one of the time of the initial creation, modification, or relationship establishment of the semantic block as a timestamp, and the representation unit or precision of the timestamp may be set differently depending on the system operating environment. Timestamp information can be utilized for managing the temporal order of semantic blocks, tracking change history, or controlling subsequent processing flows.

[0060] Probability is information indicating the reliability, relevance, or likelihood that a semantic block represents the input text or query. The control unit may set the probability based on the output result of the LLM-based block generator, the characteristics of the input data, or internal judgment criteria, and the range of the probability value or the calculation method is not limited to specific numerical values ​​or formulas. Probability can be used as a weight for determining the priority of semantic blocks, selective utilization, or subsequent processing.

[0061] The control unit can control the management of metadata, including the blocks, relationships, timestamps, and probabilities generated as described above, in conjunction with semantic blocks or record it in the storage unit. This metadata is utilized as information to expand semantic blocks from simple text units into management units that include structural, temporal, and reliability information.

[0062] Accordingly, a control unit according to one embodiment of the present invention generates metadata including blocks, relationships, timestamps, and probabilities for semantic blocks, thereby providing foundational information that enables systematic management of semantic blocks and subsequent processing, exploration, or utilization. Various modifications are possible for the configuration or utilization method of such metadata.

[0063] A control unit according to one embodiment of the present invention generates a NAQL query through the operation of a NAQL query generation unit to define semantic, structural, and temporal associations between the semantic blocks. The control unit generates the NAQL query as a means to integrally define associations of different natures. The control unit may operate the NAQL query generation unit after semantic blocks are generated and metadata is assigned, or at a time when association definition is required.

[0064] Semantic association is a relationship indicating the degree to which different semantic blocks share the same or similar concepts, topics, or contexts, and the control unit may define such semantic associations based on the content, creation background, or metadata of the semantic blocks. Semantic associations may include relationships where one semantic block complements or extends another semantic block, relationships involving similar meanings, or superordinate / subordinate conceptual relationships, and the criteria for their definition are not limited to a specific method.

[0065] Structural association is a relationship representing the logical connection or management structure between semantic blocks; the control unit can define reference relationships, inclusion relationships, dependency relationships, or sequential connection relationships between semantic blocks as structural associations. These structural associations are intended to indicate the order or hierarchy in which semantic blocks are connected and can be utilized in subsequent search or management processes.

[0066] Temporal association is a relationship defined based on time information related to the creation, modification, or relationship establishment time of semantic blocks. The control unit can define temporal association by considering the chronological relationship between semantic blocks, whether they were created simultaneously, or time intervals, and such temporal association can be utilized for managing the change history of semantic blocks or for time-based processing flow control.

[0067] The NAQL query generation unit generates an NAQL query to express the semantic, structural, and temporal associations under the control of the control unit. The NAQL query serves as query or definition information for describing relationships between semantic blocks and is configured to explicitly express associations targeting multiple semantic blocks. The syntax, expression method, or internal structure of the NAQL query is not limited to a specific format and may include various forms capable of describing associations between semantic blocks.

[0068] The control unit can control the definition or updating of associations between semantic blocks using the generated NAQL query, and the generated NAQL query can be recorded in the storage unit or referenced in subsequent processing steps. Through this, associations between semantic blocks can be managed as an explicitly defined query-based structure rather than as simple implicit relationships.

[0069] Accordingly, a control unit according to one embodiment of the present invention generates a NAQL query through the operation of a NAQL query generation unit, thereby providing a basis for systematically defining and managing semantic, structural, and temporal associations between semantic blocks. Various variations of this association definition method are possible.

[0070] A control unit according to one embodiment of the present invention defines, connects, updates, or expands relationships between semantic blocks according to the NAQL query, performs the operation of a multidimensional error verification unit on semantic blocks, relationships, timestamps, probabilities, and NAQL execution results to identify whether there is a semantic mismatch, structural error, time order error, or probability distribution anomaly, and creates and executes a transaction including the creation, modification, connection, or disposal of semantic blocks according to the verification result.

[0071] A control unit according to one embodiment of the present invention defines and manages relationships between semantic blocks based on a previously generated NAQL query. The control unit may define new relationships between semantic blocks using relationship definition information included in the NAQL query, and may perform operations to connect already defined relationships to each other, or to update or extend existing relationships. In this case, the definition, connection, update, or extension of relationships may be performed to reflect semantic association, structural connectivity, or temporal connectivity between semantic blocks, and the processing order or scope is not limited in a specific manner.

[0072] The control unit controls the operation of the multidimensional error verification unit for semantic blocks and relationships defined or modified as described above. The multidimensional error verification unit is configured to identify errors in multiple dimensions, using semantic blocks, relationships, timestamps, probabilities, and NAQL execution results as verification targets. Here, the verification target is not limited to a single data item but may be a set including the entire semantic block, its associated relationships, and metadata.

[0073] The multidimensional error verification unit can identify semantic inconsistencies based on the content or interrelationships of semantic blocks. Semantic inconsistencies may include cases where logical conflicts exist between interconnected semantic blocks, or where semantic blocks defined by the same relationship contain mutually contradictory meanings. The criteria or methods for determining such semantic inconsistencies are not limited to specific rules or formulas.

[0074] In addition, the multidimensional error verification unit can identify whether there are structural errors based on the connection structure between semantic blocks. Structural errors may include the omission of relationship definitions, redundant connections, the unintentional occurrence of circular structures, or the abnormal formation of relationship hierarchies, and the control unit can obtain these structural abnormalities as the identification result.

[0075] The multidimensional error verification unit can identify whether there are chronological errors between semantic blocks or relationships based on timestamps. Chronological errors may include cases where the timing of creation, modification, or connection of semantic blocks does not align with the logical flow, or where the temporal precedence is reversed. The criteria for interpreting timestamps or the method for determining errors are not limited to a specific method.

[0076] In addition, the multidimensional error verification unit can identify whether there are anomalies in the probability distribution based on probability. Anomalies in the probability distribution may include cases where the probability value assigned to a specific semantic block or relationship is abnormally skewed from the overall distribution, or where the probability distribution between semantic blocks generated under the same conditions is unbalanced. The determination of such probability distribution anomalies can also be performed according to various criteria.

[0077] The control unit creates and executes transactions targeting semantic blocks and relationships based on the verification results of the multidimensional error verification unit. The transactions may include the creation of semantic blocks, modification of semantic blocks, linking between semantic blocks, or disposal of semantic blocks, and one or more of these operations may be performed in combination. The control unit may control the scope, order, or execution of the transactions based on the type or severity of the verification results.

[0078] Accordingly, a control unit according to one embodiment of the present invention controls the consistent maintenance of semantic blocks and their relationships through the definition and management of relationships based on NAQL queries, the identification of errors through a multidimensional error verification unit, and the creation and execution of transactions based on the verification results. Various modifications to this processing flow are possible.

[0079] According to one embodiment of the present invention, the transaction is processed to maintain atomicity, consistency, and sequentiality, and the control unit collects the frequency of errors occurring during the transaction processing, the NAQL execution failure rate, and the transaction rollback rate, and controls the operation of the system monitoring unit.

[0080] A transaction according to one embodiment of the present invention is configured such that a plurality of processing operations, such as creating, modifying, linking, or discarding semantic blocks, are managed as a single processing unit, and such a transaction is processed to maintain atomicity, consistency, and sequentiality. Here, atomicity means that the operations included in the transaction are processed in such a way that they are either all performed or none are performed, and is a concept intended to prevent a state in which only some operations are selectively reflected. The control unit can control the transaction to maintain atomicity by confirming or reverting the processing results on a transaction-by-transaction basis.

[0081] Consistency means that semantic blocks, relationships, and metadata are maintained to satisfy defined rules or state conditions before and after transaction processing. The control unit can control the processing flow to ensure that the combination of relationships between semantic blocks, timestamps, and probabilities maintains a logically consistent state even after the transaction is completed. The criteria for this consistency or the method of maintaining it are not limited to a specific set of rules.

[0082] Sequentiality refers to the characteristic that, when multiple transactions are executed, each transaction is managed to maintain a specific processing order. The control unit may be configured to control the transaction processing order or maintain a sequential execution state to prevent the execution order between transactions from being reversed or overlapping, thereby preventing the semantic block state from changing unpredictably.

[0083] The control unit collects various operational indicators generated during the transaction processing process described above. The operational indicators collected here include error occurrence frequency, NAQL execution failure rate, and transaction rollback rate. Error occurrence frequency is information indicating the number or frequency of errors occurring during the transaction processing process, and the control unit can generate this information by aggregating error events detected during transaction execution.

[0084] The NAQL execution failure rate is information representing the proportion of NAQL queries executed during the transaction processing that failed to execute successfully. The control unit can calculate or manage the NAQL execution failure rate based on the number of execution attempts and failures of NAQL queries, and the criteria for defining failure or the aggregation method are not restricted to a specific manner.

[0085] The transaction rollback rate is information representing the proportion of cases where a transaction fails to complete normally and is reverted to a previous state. The control unit can identify instances where a transaction rollback occurs to maintain atomicity or consistency, and manage the transaction rollback rate by accumulating information on such rollback occurrences.

[0086] The control unit collects the aforementioned error occurrence frequency, NAQL execution failure rate, and transaction rollback rate, and then controls the system monitoring unit to perform its operations. The system monitoring unit can be utilized as a configuration to monitor the status during the transaction processing or to determine the operational status of the system based on the collected operational indicators. The control unit can control the system monitoring unit to transmit the collected indicators or to register them as monitoring target data.

[0087] In this way, a control unit according to one embodiment of the present invention provides a basis for continuously managing the processing status of a system by controlling the operation of a system monitoring unit by collecting the frequency of errors occurring during the transaction processing, the NAQL execution failure rate, and the transaction rollback rate, while ensuring that transactions are processed in a manner that maintains atomicity, consistency, and sequentiality.

[0088] According to one embodiment of the present invention, the system monitoring unit analyzes the collected information based on anomaly detection to generate a system stability index, dynamically adjusts semantic block generation conditions, NAQL execution conditions, multidimensional error verification criteria, or transaction processing order according to the stability index, and controls the semantic block, relationship, timestamp, probability, and transaction change history to be accumulated and stored in a storage unit so that quality management and traceability are ensured, thereby ensuring that semantic block generation, NAQL-based linking, multidimensional error verification, transaction processing, and quality management are consistently performed in a single perceptual semantic space.

[0089] A system monitoring unit according to one embodiment of the present invention performs the function of analyzing collected information, including error occurrence frequency, NAQL execution failure rate, and transaction rollback rate, collected by a control unit. The system monitoring unit is configured to analyze the collected information based on anomaly detection, thereby enabling the identification of abnormal patterns or abnormal states by comparing them with a normal system operation state. The application method, analysis cycle, or judgment criteria for anomaly detection are not limited to a specific method and can be configured in various ways depending on the system operating environment.

[0090] The system monitoring unit generates system stability indicators based on the analysis results derived from the anomaly detection described above. System stability indicators serve as indicator information representing the overall operational stability of the system and can comprehensively reflect transaction processing status, NAQL execution status, or error occurrence trends. The expression format, numerical range, or update cycle of the stability indicators are not restricted in any specific manner and can be utilized as reference information for determining the state of the system.

[0091] The control unit is controlled to dynamically adjust system operation conditions according to stability indicators generated by the system monitoring unit. The system operation conditions dynamically adjusted here include semantic block generation conditions, NAQL execution conditions, multidimensional error verification criteria, or transaction processing order. Semantic block generation conditions may be conditions related to the frequency, range, or processing method of semantic blocks, and NAQL execution conditions may be conditions related to whether NAQL queries are executed, the execution order, or the timing of execution.

[0092] The multidimensional error verification criteria may include judgment criteria for identifying semantic inconsistencies, structural errors, temporal order errors, or probability distribution anomalies, and the transaction processing order refers to a sequential flow in which multiple transactions are processed. The control unit can control the system to maintain a stable operating state by adjusting at least one of the above conditions according to changes in the stability indicator.

[0093] Additionally, the control unit controls the accumulation of semantic blocks, relationships, timestamps, probabilities, and transaction change history in the storage unit. The change history may include the processes of creating, modifying, linking, or discarding semantic blocks, changes to relationship definitions, metadata updates, and transaction processing results. The change history accumulated in the storage unit is not limited to a specific format or storage structure and can be continuously accumulated over a long period.

[0094] Quality control and traceability are ensured through such cumulative storage. Quality control includes management activities to maintain the consistency and reliability of semantic block generation results, NAQL-based linkage results, or transaction processing results, while traceability refers to the characteristic of being able to chronologically verify how a specific semantic block or relationship reached its current state through a specific processing step.

[0095] In this way, through the cooperative operation of the system monitoring unit and the control unit according to one embodiment of the present invention, semantic block generation, NAQL-based linking, multidimensional error verification, transaction processing, and quality management are controlled to be performed consistently within a single perceptual semantic space. This consistent execution structure is intended to ensure that semantic blocks, their relationships, and history management are not dispersed but maintained within a single integrated semantic space, and various modifications are possible.

[0096] A control unit according to one embodiment of the present invention is characterized in that, when generating semantic blocks through an LLM-based block generation unit, it decomposes an input text into context units, forms semantic blocks based on semantic cohesion between decomposed contexts, assigns a probability to each semantic block, and the probability is set by reflecting the contextual match and reference possibility at the time of semantic block generation.

[0097] A control unit according to one embodiment of the present invention controls the decomposition of text input from a user terminal into context units during the process of generating semantic blocks through an LLM-based block generation unit. Here, the input text may include natural language text in the form of a single sentence, multiple sentences, or paragraphs, and the control unit may decompose the input text into multiple context units depending on the structure or length of the input text. A context unit may include a sentence, a clause, a semantic paragraph, or a text part where semantic flow is maintained, and the decomposition criteria are not limited to specific rules.

[0098] The control unit controls the operation of the LLM-based block generator to form semantic blocks based on semantic cohesion, taking into account the relationships between multiple decomposed contextual units. Semantic cohesion is a concept representing the degree to which multiple contextual units share the same or similar semantic flow, and the control unit can determine semantic cohesion by considering semantic continuity, topic consistency, or reference relationships between contexts. A semantic block can be formed as a set of contextual units in which such semantic cohesion is determined to be above a certain level, and a single semantic block may include a single contextual unit or multiple contextual units.

[0099] The control unit assigns a probability to each semantic block formed as described above. Probability is information indicating the likelihood that the corresponding semantic block represents the meaning of the input text or will be referenced in subsequent processing steps, and is set individually for each semantic block. The control unit may assign different probability values ​​to each semantic block, and the expression method or range of the probability values ​​is not limited to a specific numerical system.

[0100] The above probability is set by reflecting the contextual consistency and referenceability at the time of semantic block generation. Contextual consistency is a concept indicating how consistently the contextual units constituting the semantic block match the overall context of the input text; the control unit may determine the contextual consistency by considering the semantic consistency between contextual units or the correspondence relationship with the input text. Referenceability is a concept indicating the likelihood that the semantic block will be referenced later during NAQL-based linking, error verification, or transaction processing; the control unit may reflect referenceability by considering the information density, linkability, or usability of the semantic block.

[0101] Accordingly, a control unit according to one embodiment of the present invention decomposes an input text into context units, forms semantic blocks based on semantic cohesion between decomposed contexts, and assigns a probability reflecting contextual consistency and reference possibility to each semantic block, thereby controlling the semantic block generation process to faithfully reflect the semantic structure of the input text. Various modifications are possible for this semantic block generation method.

[0102] A control unit according to one embodiment of the present invention is characterized by applying constraints including a relationship type between semantic blocks, a relationship direction, a timestamp condition, and a probability condition to a NAQL query generated through a query generation unit, and the NAQL query is configured to restrict the search target to only paths between semantic blocks that satisfy the constraints.

[0103] A control unit according to one embodiment of the present invention is configured to apply constraints to a NAQL query generated through a query generation unit. In the process of generating an NAQL query for exploring or defining relationships between semantic blocks, the control unit may control the inclusion of multiple constraints to prevent the search range from expanding indiscriminately. These constraints are utilized as reference information to restrict the paths between semantic blocks that are the targets of execution for the NAQL query.

[0104] The above constraints include relationship types between semantic blocks. Relationship types are information for distinguishing the nature of relationships defined between semantic blocks and may include criteria for identifying the type of relationship, such as semantic relationships, structural relationships, or temporal relationships. The control unit may impose relationship type constraints so that the NAQL query targets only connections between semantic blocks corresponding to a specific relationship type, and may also be configured to allow multiple relationship types simultaneously or selectively allow only specific relationship types.

[0105] Additionally, the constraints include relationship direction. Relationship direction is information indicating whether the relationship between semantic blocks is unidirectional, bidirectional, or valid only in a specific direction. The control unit can set constraints on relationship direction to ensure that the NAQL query performs a search considering the directionality between the starting and ending semantic blocks of the relationship. This allows for control to ensure that the search path between semantic blocks maintains its intended directionality.

[0106] Constraints also include timestamp conditions. A timestamp condition is a condition used to restrict the search target based on time information assigned to semantic blocks or relationships. The control unit can apply timestamp conditions to ensure that the NAQL query targets only semantic blocks and relationships created or updated within a specific time range, or it can be configured to restrict the search range by reflecting temporal precedence. The expression method or application criteria for timestamp conditions are not limited to specific time units or ranges.

[0107] In addition, the constraints include probability conditions. Probability conditions are used to restrict search targets based on probability values ​​assigned to semantic blocks; the control unit can configure the NAQL query to include only semantic blocks with a probability of a certain level or higher as search targets. The threshold criteria or comparison method of the probability conditions are not limited to specific numerical values ​​or formulas and can be configured in various ways depending on the system's operational purpose.

[0108] The control unit is configured to restrict the NAQL query to search targets only for paths between semantic blocks that satisfy the aforementioned constraints by applying constraints to the NAQL query that include the relationship type, relationship direction, timestamp condition, and probability condition. Accordingly, the execution result of the NAQL query is limited to paths between semantic blocks that satisfy the constraints, thereby preventing unnecessary or low-reliability paths from being included in the search targets.

[0109] Accordingly, a control unit according to one embodiment of the present invention controls the search for relationships between semantic blocks by applying constraints to NAQL queries to limit the search targets, thereby ensuring that the search is performed by reflecting structural, temporal, and probabilistic criteria. Various variations of this method of applying constraints are possible.

[0110] The multidimensional error verification unit according to one embodiment of the present invention is characterized by setting a semantic block structure, a NAQL query structure, timestamp consistency, and a probability distribution as independent verification dimensions, and determining whether an error exists by synthesizing the verification results for each verification dimension.

[0111] A multidimensional error verification unit according to one embodiment of the present invention is configured to independently set a plurality of verification dimensions to determine whether there is an error regarding a semantic block and its processing result. Here, each verification dimension functions as a reference axis for which verification is performed individually without mutual dependency, and has a structure to prevent deviations that may occur due to error judgment relying on a single criterion.

[0112] The multidimensional error verification unit establishes the semantic block structure as an independent verification dimension. Verification of the semantic block structure can be performed based on the unit configuration defined by the semantic block, the connection type between blocks, or the arrangement state of internal components. This allows for verification of whether the semantic block is structurally missing or abnormally connected, and the detailed criteria for structural verification are not limited to a specific format.

[0113] Furthermore, the multidimensional error validation unit establishes the NAQL query structure as an independent validation dimension. Validation of the NAQL query structure can be performed based on whether the NAQL query properly includes a structure for defining relationships between semantic blocks, and whether relationship types, relationship directions, or constraints are logically organized. This validation can be utilized as a dimension to determine the executableness or structural consistency of the NAQL query.

[0114] The multidimensional error verification unit establishes timestamp consistency as another independent verification dimension. Verification of timestamp consistency can be performed based on whether the time information assigned to semantic blocks, relationships, or transactions corresponds to the logical order or chronological flow. This enables the identification of chronological order errors, reversals of precedence relationships, or inconsistencies between time information, and the criteria for judging timestamp consistency are not limited to specific time units or ranges.

[0115] Furthermore, the multidimensional error verification unit establishes the probability distribution as an independent verification dimension. Verification of the probability distribution can be performed based on whether the probability values ​​assigned to semantic blocks are abnormally biased from the perspective of the overall distribution, or whether the probability distributions between semantic blocks generated under the same conditions are unbalanced. The method for determining the normality of the probability distribution is not limited to a specific mathematical distribution or threshold value.

[0116] The multidimensional error verification unit determines the presence of an error by synthesizing the verification results for each verification dimension regarding the semantic block structure, NAQL query structure, timestamp consistency, and probability distribution configured as described above. Here, the synthesizing judgment may be performed by simply listing the results of each verification dimension or by combining multiple verification results to determine the overall error status, and the method of synthesizing is not limited to a specific rule.

[0117] In this way, the multidimensional error verification unit according to one embodiment of the present invention sets a plurality of independent verification dimensions and determines whether an error exists by synthesizing the results of each verification dimension, thereby ensuring that error verification regarding semantic blocks and their relationships is performed from multiple perspectives without being biased toward a single criterion.

[0118] A control unit according to one embodiment of the present invention is characterized by controlling to roll back or replace a transaction containing a semantic block or NAQL query with a modification transaction when the result of multidimensional error verification is identified as an error.

[0119] A control unit according to one embodiment of the present invention performs a subsequent control operation for a transaction associated with an error when an error is identified based on the verification result of a multidimensional error verification unit. Here, the case identified as an error includes a case where an abnormal state is determined in at least one verification dimension among a semantic block structure, a NAQL query structure, timestamp consistency, or a probability distribution.

[0120] The control unit may control the rollback of transactions containing targets identified as errors. Transaction rollback refers to the operation of restoring the system to a state prior to the execution of the transaction, and includes processing to ensure that the results of semantic block creation, modification, linking, or NAQL query execution are not reflected in the system state. To maintain atomicity and consistency, the control unit may set the entire transaction containing the error as a rollback target, and the scope or timing of the rollback is not limited in a specific manner.

[0121] Additionally, the control unit may control the replacement of transactions identified as errors with correction transactions. A correction transaction is a transaction intended to supplement or modify a semantic block or NAQL query where an error occurred in an existing transaction, and may be configured to eliminate or mitigate the cause of the error. The control unit may create and execute correction transactions that include the regeneration of semantic blocks, modification of metadata, changes to relation definitions, or adjustment of NAQL query conditions.

[0122] The control unit can selectively apply either a rollback or a correction transaction, or control the execution of correction transactions sequentially following a rollback, depending on the type of error, location of occurrence, or verification results. This control method is not limited to specific error handling scenarios and can be applied in various ways depending on system operation policies or states.

[0123] Accordingly, a control unit according to one embodiment of the present invention is configured to maintain consistency and stability of the system state by controlling the rollback of a transaction containing a semantic block or NAQL query or its replacement with a modification transaction when the transaction is identified as an error as a result of multidimensional error verification.

[0124] A control unit according to one embodiment of the present invention is characterized by connecting the semantic block generation, NAQL execution, multidimensional error verification, and transaction execution steps into a single transaction flow, and recording the execution result of each step based on a timestamp.

[0125] A control unit according to one embodiment of the present invention controls the semantic block generation step, the NAQL execution step, the multidimensional error verification step, and the transaction execution step to be connected into a single continuous transaction flow, rather than performing them as separate individual processes. Here, a single transaction flow means that the entire processing flow is configured to be managed as a single transaction unit, rather than a structure in which the completion status of each step is determined independently.

[0126] The control unit controls the linkage of results generated in the semantic block generation stage to the input of the NAQL execution stage, and processes the results of the NAQL execution stage to be continuously passed to the multidimensional error verification stage. Additionally, the results of the multidimensional error verification stage are connected to the transaction execution stage, thereby controlling the execution of semantic block generation, relationship definition or modification, and rollback decisions within a single flow. This linkage between stages may be a method in which each stage is executed sequentially, or it may involve conditional execution or branching structures, and is not limited to a specific execution structure.

[0127] The control unit records the execution results of each step constituting the single transaction flow based on timestamps. Here, recording based on timestamps means that for each of the semantic block generation results, NAQL execution results, multidimensional error verification results, and transaction execution results, the execution time or completion time is stored together as time information. The control unit may assign individual timestamps to each step and manage the records to clearly identify the temporal order within the same transaction flow.

[0128] The above timestamp-based records can be utilized as reference information to verify the order and timing of each step within a single transaction flow. The control unit can use these records to perform subsequent operations such as tracking processing history, identifying the point of error, or reconstructing transactions, and the recording method or precision of the timestamps is not limited to specific time units.

[0129] Accordingly, a control unit according to one embodiment of the present invention connects the semantic block generation, NAQL execution, multidimensional error verification, and transaction execution steps into a single transaction flow, and by recording the execution results of each step based on timestamps, controls the entire processing flow to be maintained in a temporally continuous and traceable structure.

[0130] According to one embodiment of the present invention, the system monitoring unit analyzes the error occurrence frequency, NAQL execution failure rate, and transaction rollback rate based on anomaly detection to generate a system stability indicator, and adjusts the semantic block generation conditions or multidimensional error verification criteria of the control unit according to the stability indicator.

[0131] A system monitoring unit according to one embodiment of the present invention performs the function of evaluating the operational status of a system by analyzing operational information collected during a transaction processing process. The operational information includes error occurrence frequency, NAQL execution failure rate, and transaction rollback rate, and the system monitoring unit is configured to analyze this information based on anomaly detection. Analysis based on anomaly detection can be utilized as an analysis method to identify abnormal fluctuations or signs of anomalies by comparing them with normal operation patterns, and the method of application or judgment criteria are not limited to a specific form.

[0132] The system monitoring unit generates system stability indicators by synthesizing the anomaly detection-based analysis results regarding the aforementioned error occurrence frequency, NAQL execution failure rate, and transaction rollback rate. System stability indicators are indicator information representing the overall processing stability or reliability of the system, and can be generated in the form of a single indicator or multiple detailed indicators. The representation method, update cycle, or management criteria of the stability indicators can be set in various ways depending on the system operating environment.

[0133] The control unit is controlled to adjust system operating conditions according to stability indicators generated by the system monitoring unit. The system operating conditions subject to adjustment include the control unit's semantic block generation conditions or multidimensional error verification criteria. The semantic block generation conditions may include conditions related to the criteria, processing range, or generation method for generating semantic blocks, and the multidimensional error verification criteria may include error judgment criteria regarding semantic block structure, NAQL query structure, timestamp consistency, or probability distribution.

[0134] The control unit can control the system to respond to situations where processing stability is degraded or to mitigate the possibility of errors by adjusting at least one of the semantic block generation conditions or multidimensional error verification criteria according to changes in stability indicators. Such adjustment of conditions may be performed automatically or applied in stages, and is not limited to a specific adjustment method.

[0135] Accordingly, a system monitoring unit according to one embodiment of the present invention analyzes the frequency of error occurrence, NAQL execution failure rate, and transaction rollback rate based on anomaly detection to generate a system stability indicator, and supports the system in maintaining a stable state by adjusting the semantic block generation conditions or multidimensional error verification criteria of the control unit according to the stability indicator.

[0136] A control unit according to one embodiment of the present invention is characterized by performing a self-improving function that operates to reduce the possibility of errors occurring during subsequent semantic block generation or NAQL execution by using a semantic block, a relationship, a timestamp, a probability, and a change history of the NAQL execution result as training data.

[0137] A control unit according to one embodiment of the present invention performs a self-learning function that utilizes various change histories occurring during the operation of a system as training data. Here, the change history includes histories related to the creation, modification, connection, or disposal of semantic blocks; histories of the definition or update of relationships between semantic blocks; histories of changes in timestamps; histories of changes in probabilities; and histories of changes in NAQL execution results. The control unit may manage such change histories as individual data items or as historical data in chronological order.

[0138] The control unit is configured to use the aforementioned change history as training data and reflect it in the operation during subsequent semantic block generation or NAQL execution. The change history used as training data may include cases where errors occurred, cases where no errors occurred, or cases where errors were corrected during past processing, and the control unit may operate by comprehensively referring to these histories to reduce the likelihood of errors occurring in subsequent processing.

[0139] The self-learning function can influence the method of context decomposition of input text, the judgment of semantic cohesion, or the method of assigning probabilities during the semantic block generation stage, and can influence the method of defining relationships, setting constraints, or the execution order during the NAQL execution stage. The control unit can refer to past results included in the change history to control the avoidance of processing patterns in which errors occurred repeatedly, or to prioritize the application of processing patterns that had a low frequency of errors.

[0140] The self-learning function performed by the control unit is not limited to a specific learning model, parameter update rules, or the number of iterations. The degree of accumulation of change history, the timing of application, or the scope of reflection can be flexibly set according to the system operating status, and the self-learning function may be performed continuously or only when certain conditions are met.

[0141] Accordingly, a control unit according to one embodiment of the present invention is configured to gradually improve the processing accuracy and stability of the system by using semantic blocks, relationships, timestamps, probabilities, and the change history of NAQL execution results as training data, and by performing a self-learning function that operates to reduce the possibility of errors occurring during subsequent semantic block generation or NAQL execution.

[0142] A control unit according to one embodiment of the present invention is characterized by identifying a valid semantic block based on a probability comparison result when multiple versions having different timestamps exist for the same semantic block, and disabling or managing the remaining semantic blocks as reference-only.

[0143] A control unit according to one embodiment of the present invention is configured to manage cases where multiple versions having different timestamps exist for the same semantic block. Here, multiple versions having different timestamps include a state in which multiple instances of a semantic block exist that were created based on the same semantic content but have different creation times, modification times, or update times. The control unit can identify such version information based on timestamps.

[0144] The control unit compares probabilities for the plurality of semantic block versions. The probability comparison is performed based on the probability value assigned to each semantic block and is utilized as information reflecting the likelihood, reliability, or usability of the semantic block representing the input text or system processing flow. The method or criteria for probability comparison are not limited to a specific numerical comparison method and may include relative comparison, rank-based comparison, or condition-based comparison.

[0145] The control unit identifies a valid semantic block based on the result of the probability comparison above. A valid semantic block refers to a semantic block selected from among multiple versions of the same semantic block to be actually utilized in system operation. The control unit can identify a single version of a semantic block as a valid semantic block, and in some cases, it is also possible to identify multiple semantic blocks simultaneously as valid targets.

[0146] The control unit controls the remaining semantic blocks not identified as valid semantic blocks to be disabled or managed as reference-only. Disablement refers to a state in which the semantic block is restricted from being automatically selected or used during subsequent semantic block generation, NAQL execution, or transaction processing. Reference-only management includes a state in which the semantic block is not a direct target for processing but is managed to be accessible only for history verification, comparative analysis, or tracing purposes.

[0147] The control unit may reflect these disabled or reference-only management states in the metadata or management information of the semantic block, and the timing of state transitions or the duration of retention are not restricted in a specific manner. This ensures that system operation remains consistent without confusion, even if multiple versions of the same semantic block exist.

[0148] Accordingly, a control unit according to one embodiment of the present invention is configured to systematically perform version management of semantic blocks by identifying a valid semantic block based on a probability comparison result when multiple versions having different timestamps exist for the same semantic block, and by deactivating or managing the remaining semantic blocks as reference-only.

[0149] A control unit according to one embodiment of the present invention is characterized by tracking referenced semantic blocks, NAQL queries, and transaction information regarding response results provided to a user terminal, storing them in a quality management log, and managing semantic reliability and response consistency based on the quality management log.

[0150] A control unit according to one embodiment of the present invention is configured to track how a response result provided to a user terminal was generated through a processing process. Here, the response result provided to the user terminal may include result information regarding a user query, semantic block-based processing results, or output information reflecting NAQL execution results, and the format or expression method of the response is not limited to a specific form.

[0151] The control unit identifies and tracks the semantic blocks referenced during the process of generating the response result. The referenced semantic blocks refer to the semantic blocks used to construct or derive the response result, and the control unit may include the identification information, version information, or related metadata of the semantic blocks as tracking targets. Through this, it is possible to verify which semantic blocks were based on when the response result was generated.

[0152] Additionally, the control unit tracks NAQL queries used in the response result generation process. NAQL queries contain query information executed to define relationships between semantic blocks, search for paths, or perform conditional linking, and the control unit can manage whether the corresponding NAQL query was executed, applied constraints, or execution results in conjunction with the response results.

[0153] The control unit also tracks transaction information related to the response result. Transaction information includes processing units related to the creation, modification, linking, or rollback of semantic blocks, and may indicate whether the response result was generated within a single transaction flow or affected by multiple transactions. The control unit tracks this transaction information in conjunction with the response result.

[0154] The control unit stores the semantic blocks, NAQL queries, and transaction information tracked as described above as a quality control log. The quality control log is log data intended to accumulate the generation history of response results and related information, and can be continuously stored in the storage unit. The storage structure, recording items, or retention period of the quality control log are not restricted in a specific manner.

[0155] The control unit manages semantic reliability and response consistency based on the quality control logs mentioned above. Semantic reliability is a concept representing the reliability level or stability of the semantic block referenced by the response result; the control unit may manage this by considering consistency when identical or similar response results are repeatedly generated, whether errors have occurred, or the history of corrections. Response consistency includes a concept indicating whether the response results provided for identical or similar queries maintain consistent semantics despite the passage of time or changes in processing conditions.

[0156] Accordingly, a control unit according to one embodiment of the present invention is configured to continuously maintain and manage the quality and reliability of response results by tracking referenced semantic blocks, NAQL queries, and transaction information regarding response results provided to a user terminal, storing them as a quality management log, and managing semantic reliability and response consistency based on the quality management log.

[0157] According to one embodiment of the present invention, the NAQL query is configured to include complex constraints including a relationship type between semantic blocks, a relationship direction, a timestamp condition, and a probability condition, and the complex constraints restrict a path that can be explored on a semantic block graph according to structural, temporal, and probabilistic conditions, and the control unit defines, connects, updates, or expands the relationship between semantic blocks using the NAQL query, records semantic block relationship information including the NAQL execution result in a storage unit, and controls the semantic blocks, relationships, timestamps, probability, and NAQL execution result so that they are used as reference data for subsequent transaction processing, verification, or quality control, thereby ensuring that NAQL constraint-based semantic block exploration and relationship management are consistently performed on a single perceptual semantic space.

[0158] A NAQL query according to one embodiment of the present invention is a query for exploring and defining relationships between semantic blocks, and is configured to include complex constraints including relationship type, relationship direction, timestamp condition, and probability condition between semantic blocks. The complex constraints are set in a combined form of multiple different conditions rather than a single criterion, so that the exploration of relationships between semantic blocks is performed according to multidimensional criteria.

[0159] The aforementioned composite constraints serve as criteria for restricting explorable paths on the semantic block graph. Structural conditions include conditions that restrict explorable paths based on the type or direction of relationships between semantic blocks, thereby ensuring that only relationship paths with a specific structure are included as explorable targets. Temporal conditions include conditions that restrict the scope of exploration by considering the creation or update time of semantic blocks or relationships based on timestamp conditions. Probabilistic conditions include conditions that restrict explorable paths by reflecting the reliability or availability of semantic blocks based on probability conditions.

[0160] The control unit controls the definition, connection, update, or extension of relationships between semantic blocks using an NAQL query that includes the complex constraints described above. Defining relationships includes the operation of establishing new relationships between semantic blocks, and connecting relationships includes the operation of linking existing relationships. Updating relationships includes the operation of changing the attributes or conditions of existing relationships, and extending relationships includes the operation of extending the existing relationship structure to include additional semantic blocks.

[0161] The control unit controls the recording of semantic block relationship information, including the execution results of an NAQL query, to the storage unit. The semantic block relationship information may include relationships between semantic blocks defined or modified by the NAQL query, constraints applied to the relationships, searched path information, or state changes resulting from the execution results. The format or storage structure of the information recorded in the storage unit is not restricted in a specific manner.

[0162] The control unit controls the semantic block, relationship, timestamp, probability, and NAQL execution results to be utilized as reference data for subsequent transaction processing, multidimensional error verification, or quality control. In the transaction processing stage, the data can be used as a criterion for determining transaction creation, execution, or rollback, and in the verification stage, it can be used as input data to determine the presence of structural, temporal, or probabilistic errors. Additionally, in the quality control stage, it can be utilized as reference data for semantic reliability, response consistency, or processing history tracking.

[0163] In this way, a control unit according to one embodiment of the present invention manages relationships between semantic blocks using an NAQL query including complex constraints, and controls the execution results and related data to be utilized consistently, thereby configuring NAQL constraint-based semantic block search and relationship management to be performed consistently in a single perceptual semantic space.

[0164] A control unit according to an embodiment of the present invention maps the semantic block, relationship, and NAQL execution result to coordinates in a single perceptual semantic space, and the coordinates in the perceptual semantic space are configured to be movable according to the creation, modification, connection, or disposal of the semantic block, and the coordinate movement occurs in conjunction with changes in relationships between semantic blocks, timestamp changes, or probability changes, and the control unit identifies the degree of semantic change by comparing changes in distance or position between coordinates at different points in time corresponding to the same semantic block, and the degree of semantic change is used as a criterion for determining whether to maintain, update, branch, or dispose of the semantic block, and controls the creation and execution of a transaction for the semantic block or relationship according to the determination result, thereby managing the consistency and quality of the semantic structure based on changes in the perceptual semantic space coordinates.

[0165] A control unit according to one embodiment of the present invention is configured to map semantic blocks, relationships between semantic blocks, and NAQL execution results to coordinates on a single perceptual semantic space. Here, the perceptual semantic space refers to an abstract space for integrally representing semantic blocks and their relationships, and the coordinates are utilized as location information representing the state of semantic blocks or relationships on the space. The control unit may assign coordinates on a semantic block unit, a relationship unit, or an NAQL execution result unit, and the number of dimensions or the method of representation of the coordinates is not limited to a specific form.

[0166] The coordinates in the above perceptual semantic space are configured to be movable depending on the creation, modification, connection, or disposal of semantic blocks. The control unit can set new coordinates or change existing coordinates when a semantic block is newly created, and can control that changes are reflected as coordinate shifts when a semantic block is modified or connected to another semantic block. Additionally, when a semantic block is disposed of, it can be managed by deactivating the corresponding coordinates or removing them from the semantic space.

[0167] The above coordinate shift occurs in conjunction with changes in relationships between semantic blocks, changes in timestamps, or changes in probability. Changes in relationships include cases where the connection structure between semantic blocks is altered, and changes in timestamps include cases reflecting changes in the creation time or update time. Changes in probability include cases where a change occurs in the reliability or referability of semantic blocks. The control unit may control the occurrence of a coordinate shift in the perceptual semantic space when at least one of these change factors occurs.

[0168] The control unit compares the distance or positional change between coordinates at different points in time corresponding to identical semantic blocks. Here, the distance or positional change between coordinates is utilized as information indicating how much the identical semantic block has moved within the semantic space due to the passage of time or processing results. Through these comparison results, the control unit can identify the degree of semantic change of the semantic block, and the distance comparison method or the criteria for interpreting positional changes are not limited to specific mathematical formulas.

[0169] The above degree of semantic change is used as a criterion for determining whether to maintain, update, branch, or discard a semantic block. A determination to maintain includes cases where the semantic change of the semantic block is minimal, and a determination to update includes cases where the semantic meaning of the semantic block has changed to a certain level but inherits the existing semantic meaning. A determination to branch includes cases where changes occur in multiple semantically different directions within the same semantic block, and a determination to discard includes cases where it is determined that the validity of the semantic block has been lost. The control unit may select at least one of the above determinations according to the degree of semantic change.

[0170] The control unit controls the creation and execution of a transaction regarding a semantic block or relationship based on the above judgment result. The transaction may include the creation, modification, linking, or disposal of a semantic block, or the definition, updating, or removal of a relationship, and may consist of a single transaction or multiple transactions. Through the execution of the transaction, the control unit can ensure that the coordinate state in the perceptual semantic space is reflected in accordance with the judgment result.

[0171] Accordingly, a control unit according to one embodiment of the present invention is configured to manage the state of a semantic structure based on changes in perceptual semantic space coordinates and to control transactions regarding semantic blocks and relationships, thereby ensuring that the consistency and quality of the semantic structure are continuously maintained and managed.

[0172] Meanwhile, as an example of a method for generating a Perceptual Meaning Space Topology (PMST) vector for generating perceptual meaning space coordinates in one embodiment of the present invention, the control unit can generate an integrated meaning vector (z) by weightedly fusing a text embedding vector (f_text), a graph embedding vector (f_graph), and manifold coordinates (M). The above integrated semantic vector (z) can be calculated according to the formula z = α·f_text + β·f_graph + γ·M, where f_text is a text embedding vector of d₁ dimensions (e.g., 768 dimensions) generated using a transformer-based language model (BERT, GPT, BGE, E5, etc.), f_graph is a graph embedding vector of d₂ dimensions (e.g., 256 dimensions) generated using a graph neural network (GCN, GraphSAGE, GAT, etc.), and M is a manifold coordinate of d₃ dimensions generated through local tangent space estimation, geodesic distance calculation, and curvature-aware position adjustment. Additionally, α, β, and γ can be set as weights to satisfy α + β + γ = 1 and the conditions α > 0, β > 0, and γ > 0. The above weights may be adjusted according to domain characteristics, such that γ is set relatively high in domains where structural consistency is important (medical, legal, financial), α is set relatively high in domains where textual meaning is important (creative, literary), and β is set relatively high in domains where relational information is important (science, technology). However, the above formula is merely one embodiment of the present invention, and other fusion methods for generating perceptual semantic space coordinates may also be included within the scope of the present invention.

[0173] In addition, a loss function may be defined for the optimization of the above-mentioned integrated semantic vector (z), and as an example, it may be constructed in the form of L(z) = λ₁·||z - P·f_text||² + λ₂·Σ_struct||z_i - z_j||² + λ₃·Σ_ref||z_i - z_j||² + λ₄·Σ_usage||z_i - z_j||² + λ5·Σ_version||z_i - z_j||². Here, λ1 is a text semantic preservation weight that reflects the minimization of distance from the original text semantics, λ2 is a structural adjacency constraint weight that reflects the distance constraint between adjacent nodes in the hierarchical structure, λ3 is a reference relationship constraint weight that reflects supports, refutes, and derived relationships, λ4 is a usage pattern constraint weight that reflects the distance constraint between concurrently referenced knowledge units, and λ5 is a version continuity constraint weight that reflects the continuity between versions of the same knowledge unit. In addition, P represents the projection matrix from the text embedding space to the PMST space, and Σ_struct, Σ_ref, Σ_usage, and Σ_version represent the sum of pairs of nodes connected according to the structure graph, reference graph, usage graph, and version graph, respectively. Optimization of the loss function can be performed iteratively using Gradient Descent according to the formula z ← z - η·∂L / ∂z, and can be repeated until the amount of coordinate change converges to below a predetermined threshold. The loss function and optimization method are merely one embodiment of the present invention, and other forms of loss functions or optimization algorithms may be applied.

[0174] In addition, the optimization loop of the above PMST vector may include a global relaxation step, a local constraint application step, a stability score calculation step, and a convergence check step. In the global relaxation step, all coordinates are updated simultaneously in the direction of the gradient of the total loss function, and the formula z^(t+1) = z^(t) - η × ∂L / ∂z may be applied, where η represents the learning rate. In the local constraint application step, adjacent node proximity constraints that adjust connected nodes in the graph to be located within a predetermined distance in the PMST space, version same patch constraints that adjust versions of the same knowledge unit to cluster within the same region, and semantic similarity clustering constraints that adjust semantically similar knowledge units to cluster may be applied. In the stability score calculation step, σ_i = ||z_i(v₁) - z_i(v₂)|| -1 A stability score can be calculated according to the formula, and the stability score takes a higher value when there is less semantic change between versions, and can be used as an indicator representing the semantic stability of the knowledge unit. In the convergence check stage, if the condition ||z^(t+1) - z^(t)|| < ε is satisfied, it can be confirmed as the final PMST vector (z*), where ε represents the threshold value for convergence determination. Parameters such as the number of iterations, learning rate, and threshold value of the optimization loop can be appropriately set according to system requirements and performance characteristics.

[0175] Furthermore, the structural relationship information used in the graph embedding section may be composed of multiple graph types and may include a structural graph representing hierarchical relationships such as Chapter, Section, and Parent-Child; a reference graph representing logical relationships such as supports, refutes, and derived; a procedure graph representing sequential relationships such as step-to-step; a usage graph representing co-reference relationships between knowledge units simultaneously referenced in a user or AI session; and a version graph representing relationships between versions based on temporal changes of the same knowledge unit. The multiple graphs may be integrated using methods such as weighted summation, concatenation, or attention mechanisms after graph embedding is performed individually. However, the graph types are exemplary, and other types of graphs may be added or some may be omitted depending on system requirements.

[0176] In addition, the above-described multilayer graph structure may further include a procedure graph, which may be configured to represent the stepwise execution order between knowledge units, the precedence relationship between steps, or the step-to-step sequential relationship. The procedure graph may be used to represent the semantic structure of knowledge units containing procedural knowledge, workflows, or algorithmic steps, and may be reflected in the PMST vector generation and optimization process either independently of or in combination with other graph types.

[0177] In addition, the learning rate η, convergence threshold ε, and number of iterations used in the optimization loop can be set according to system requirements and performance characteristics. As an example, the learning rate η is 10 -4 to 10 -1It can be set within a range, and 0.01 can be used as the initial value; it can also be configured to gradually decrease η if the decay rate of the loss function stagnates more than a predetermined number of times. Additionally, the convergence threshold ε is 10 -4 to 10 -2 It can be set within a range, and the number of iterations can be set with an upper limit of a maximum number of iterations N_max (e.g., 100 times), but can be configured to terminate early if the amount of coordinate change in multiple consecutive iterations remains less than ε.

[0178] Meanwhile, the symbols, subscripts, and formula notations used in this specification are used with the same meaning throughout the document; in particular, λ1 to λ5 each refer to the same weight, and Σ_struct, Σ_ref, Σ_usage, and Σ_version are used uniformly to denote the summation operation of connected pairs of nodes according to each graph type. Additionally, different symbols or subscripts are not mixed for the same formula element, so as to maintain clarity and consistency in the interpretation of the formula.

[0179] Meanwhile, the Meaning Address URI used in one embodiment of the present invention is an address system for identifying, referencing, and managing knowledge units in a perceptual semantic space, and may be configured in the format abk: / / {work_id} / {kind} / {path}@{version}. Here, abk: / / represents a schema identifier representing the semantic address system of the present invention, {work_id} represents a unique identifier of the workspace or domain to which the knowledge belongs, and {kind} is a value representing the type of knowledge unit and may include Theory, Fact, Example, Procedure, etc. Additionally, {path} represents the hierarchical location path of the knowledge unit, and {version} represents the version number of the knowledge unit.

[0180] As an example of the above semantic address, a theoretical knowledge unit in the physics domain can be expressed in the form of abk: / / physics / Theory / quantum.uncertainty@v1, factual information in the medical domain can be expressed in the form of abk: / / medical / Fact / diabetes.symptom.polyuria@v3, and a procedural knowledge unit in the legal domain can be expressed in the form of abk: / / legal / Procedure / contract.review.step1@v2. A semantic address constructed in this way is designed to enable AI to immediately identify the domain of the knowledge, the type of the knowledge unit, its hierarchical position, and version information solely from the address string, thereby providing semantic transparency.

[0181] Furthermore, the above semantic address can be associated with semantic blocks, relationships, timestamps, and probabilities to serve as a consistent identifier during the creation, referencing, updating, and tracking of knowledge units, and can be used as a condition element in NAQL queries or as a reference key in transaction processing and quality control logs. However, the above URI format is merely one embodiment of the present invention, and other semantic address systems may be applied depending on the knowledge representation method, system requirements, or application domain, and are included within the scope of the present invention.

[0182] In addition, the semantic address (Meaning Address URI) described above can be managed as a value stored in the mas_uri field of the bridge table, and the mas_uri is used as a unique identifier containing the entire semantic address string. Accordingly, the semantic address (URI) performs the role of a logical identifier and can be managed by directly corresponding to the oid and PMST vectors representing the physical storage location through the bridge table.

[0183] Meanwhile, in one embodiment of the present invention, a bridge table may be used to manage semantic addresses (URI), physical addresses (OID), and PMST vectors (z*) in an integrated manner, and the bridge table may be implemented with reference to FIG. 4 according to the present invention. The bridge table includes a mas_uri field that stores semantic addresses, and the mas_uri field may be configured as a VARCHAR (512) type and used as a primary key representing semantic addresses. In addition, an oid field for identifying physical storage locations may be configured as a BIGINT type, and a pmst_vector field for storing PMST vectors corresponding to knowledge units may be configured as a VECTOR (512) type. Along with this, a version field for storing the version number of a knowledge unit may be configured as an INTEGER type, and a model_id field for identifying the model used for embedding may be configured as a VARCHAR (64) type. Additionally, the created_at field indicating the creation time and the updated_at field indicating the last modified time can be configured as TIMESTAMP type, the stability_score field indicating semantic stability between versions can be configured as FLOAT type, and the content_hash field to verify the integrity of the content can be configured as CHAR(64) type.

[0184] In the pmst_vector field of the above bridge table, a Hierarchical Navigable Small World (HNSW), Inverted File (IVF), or other approximate nearest neighbor search index may be constructed, thereby improving the efficiency of semantic search. Additionally, the above bridge table may be configured to efficiently perform bidirectional URI↔OID conversion between semantic addresses and physical addresses by additionally constructing a hash index for the mas_uri field and a B-tree index for the oid field. However, the above table structure and index configuration are merely one embodiment of the present invention, and fields may be added, deleted, or modified according to system requirements.

[0185] In addition, in one embodiment of the present invention, the semantic drift of the same knowledge unit per version can be calculated according to the formula Δ = ||z_v2 - z_v1||, where z_v1 represents the PMST vector of the first version and z_v2 represents the PMST vector of the second version, and ||·|| represents the Euclidean Norm or other distance measurement methods. Depending on the semantic drift (Δ), the type of change of the knowledge unit can be classified according to multiple criteria. If Δ is smaller than θ_small, it is classified as an “Expression Change,” which corresponds to modifications with minimal semantic change, such as changes in sentence structure or synonym substitution, and the same URI is maintained while only the version number is increased. Additionally, if Δ is greater than or equal to θ_small but less than θ_large, it is classified as a “Content Change,” which involves the addition, deletion, or semantic reorganization of information, and the version number is increased while the same URI is maintained and the change history is recorded. On the other hand, if Δ is greater than or equal to θ_large, it is determined to be a new knowledge unit semantically different from the original and may be classified as a “Fork,” in which case a new URI is automatically generated and the derivation relationship (derived_from) with the original knowledge unit may be recorded. The threshold values ​​θ_small and θ_large can be appropriately set according to domain characteristics and system requirements.

[0186] In addition, in one embodiment of the present invention, the contribution of each knowledge unit referenced when the AI ​​generates a response can be calculated according to the formula contribution_i = similarity(z_response, z_i) / Σ_j similarity(z_response, z_j). Here, z_response refers to the PMST vector corresponding to the AI ​​response text, z_i refers to the PMST vector of the i-th referenced knowledge unit, and Σ_j refers to the sum of all referenced knowledge units. Additionally, similarity can be defined as cosine similarity or other similarity measurement functions, and as an example, the formula cosine_similarity(a, b) = (a · b) / (||a|| × ||b||) can be applied.

[0187] As an example of a settlement amount per author based on the above contribution, it may be calculated according to the formula Settlement_i = Usage × Contribution_i × Unit Price, where usage refers to the amount of resources consumed in generating AI responses, which may include the number of tokens or API calls; contribution_i refers to the normalized contribution rate of the i-th knowledge unit; and the unit price is the settlement unit price per unit of usage, which may be set according to the service policy. Furthermore, the list of URIs of the knowledge units referenced during the above settlement process, the contribution of each knowledge unit, and the settlement amount may be recorded in the settlement log to ensure traceability. However, the above contribution calculation method and settlement method are merely one embodiment of the present invention, and other forms of contribution calculation methods or settlement models may be applied.

Claims

Claim 1 A triple-layer fusion-based perceptual semantic space coordinate system comprises: a user terminal including a display unit and a user input unit; and a server connected to the user terminal via a wired or wireless network; wherein the server includes a communication unit, a storage unit, and a control unit; wherein the control unit receives text or a query input from the user terminal and generates a semantic block through the operation of an LLM-based block generation unit, generates metadata including blocks, relationships, timestamps, and probabilities for the semantic block, generates an NAQL query through the operation of an NAQL query generation unit to define semantic, structural, and temporal associations between the semantic blocks, defines, connects, updates, or expands relationships between the semantic blocks according to the NAQL query, maps the semantic block, relationships, and NAQL execution results to coordinates in a single perceptual semantic space, wherein the mapping to the coordinates generates an integrated semantic vector (z) by weighted fusion of a text embedding vector (f_text), a graph embedding vector (f_graph), and a manifold coordinate (M), wherein z = α·f_text + It is performed according to the formula β·f_graph + γ·M (where α + β + γ = 1, α > 0, β > 0, γ > 0), and the coordinates in the perceptual semantic space are configured to be movable according to the creation, modification, connection, or disposal of semantic blocks; said coordinate movement occurs in conjunction with changes in relationships between semantic blocks, changes in timestamps, or changes in probability; said control unit identifies the degree of semantic change by comparing changes in distance or position between coordinates at different points in time corresponding to the same semantic block; said degree of semantic change is used as a criterion for determining whether to maintain, update, branch, or dispose of the semantic block; and controls the creation and execution of transactions regarding semantic blocks or relationships according to the result of said determination, semantic block, relationship, timestamp,A triple-layer fusion-based perceptual semantic space coordinate system characterized by performing operations of a multidimensional error verification unit on probability and NAQL execution results to identify semantic inconsistencies, structural errors, temporal order errors, or probability distribution anomalies, creating and executing transactions including semantic block creation, modification, linking, or disposal according to the verification results, processing the transactions to maintain atomicity, consistency, and sequentiality, collecting error occurrence frequency, NAQL execution failure rate, and transaction rollback rate occurring during the transaction processing to perform operations of a system monitoring unit, the system monitoring unit analyzing the collected information based on anomaly detection to generate a system stability index, dynamically adjusting semantic block creation conditions, NAQL execution conditions, multidimensional error verification criteria, or transaction processing order according to the stability index, and controlling the accumulation of semantic blocks, relationships, timestamps, probabilities, and transaction change history in a storage unit to ensure quality control and traceability, thereby ensuring that semantic block creation, NAQL-based linking, multidimensional error verification, transaction processing, and quality control are consistently performed within a single perceptual semantic space. Claim 2 A triple-layer fusion-based perceptual semantic space coordinate system according to claim 1, wherein the control unit, in generating semantic blocks through an LLM-based block generation unit, decomposes input text into context units, forms semantic blocks based on semantic cohesion between decomposed contexts, assigns a probability to each semantic block, and the probability is set by reflecting the contextual match and reference possibility at the time of semantic block generation. Claim 3 A triple-layer fusion-based perceptual semantic space coordinate system according to claim 1, wherein the control unit imposes constraints including a relationship type between semantic blocks, a relationship direction, a timestamp condition, and a probability condition on a NAQL query generated through an NAQL query generation unit, and the NAQL query is configured to restrict the search target to only paths between semantic blocks that satisfy the constraints. Claim 4 A triple-layer fusion-based perceptual semantic space coordinate system according to claim 1, wherein the multidimensional error verification unit sets the semantic block structure, NAQL query structure, timestamp consistency, and probability distribution as independent verification dimensions, and determines whether there is an error by synthesizing the verification results for each verification dimension. Claim 5 A triple-layer fusion-based perceptual semantic space coordinate system according to claim 4, wherein the control unit controls to roll back or replace with a modification transaction a transaction containing the semantic block or NAQL query when identified as an error by a multidimensional error verification result. Claim 6 A triple-layer fusion-based perceptual semantic space coordinate system according to claim 1, wherein the control unit connects the semantic block generation, NAQL execution, multidimensional error verification, and transaction execution steps into a single transaction flow and records the execution results of each step based on timestamps. Claim 7 A triple-layer fusion-based perceptual semantic space coordinate system according to claim 1, wherein the system monitoring unit analyzes the frequency of error occurrence, NAQL execution failure rate, and transaction rollback rate based on anomaly detection to generate a system stability indicator, and adjusts the semantic block generation conditions or multidimensional error verification criteria of the control unit according to the stability indicator. Claim 8 A triple-layer fusion-based perceptual semantic space coordinate system according to claim 1, characterized in that the control unit performs a self-improving function that operates to reduce the possibility of errors occurring during subsequent semantic block generation or NAQL execution by using semantic blocks, relationships, timestamps, probabilities, and change history of NAQL execution results as training data. Claim 9 A triple-layer fusion-based perceptual semantic space coordinate system according to claim 1, wherein the control unit identifies a valid semantic block based on a probability comparison result when multiple versions having different timestamps exist for the same semantic block, and disables or manages the remaining semantic blocks for reference only. Claim 10 A triple-layer fusion-based perceptual semantic space coordinate system according to claim 1, wherein the control unit tracks referenced semantic blocks, NAQL queries, and transaction information regarding response results provided to a user terminal and stores them in a quality control log, and manages semantic reliability and response consistency based on the quality control log. Claim 11 A triple-layer fusion-based perceptual semantic space coordinate system comprises: a user terminal including a display unit and a user input unit; and a server connected to the user terminal via a wired or wireless network; wherein the server includes a communication unit, a storage unit, and a control unit; wherein the control unit receives text or a query input from the user terminal and generates a semantic block through the operation of an LLM-based block generation unit, generates metadata including blocks, relationships, timestamps, and probabilities for the semantic block, and generates a NAQL query through the operation of an NAQL query generation unit to define semantic, structural, and temporal associations between the semantic blocks, wherein the NAQL query is configured to include complex constraints including relationship types between semantic blocks, relationship directions, timestamp conditions, and probability conditions, and wherein the complex constraints restrict traversable paths on a semantic block graph according to structural, temporal, and probabilistic conditions; wherein the control unit defines, connects, updates, or expands relationships between semantic blocks using the NAQL query, maps the semantic blocks, relationships, and NAQL execution results to coordinates in a single perceptual semantic space, and Mapping to coordinates generates an integrated semantic vector (z) by weightedly fusing the text embedding vector (f_text), the graph embedding vector (f_graph), and the manifold coordinates (M), and is performed according to the formula z = α·f_text + β·f_graph + γ·M (where α + β + γ = 1, α > 0, β > 0, γ > 0); the coordinates in the perceptual semantic space are configured to be movable according to the creation, modification, connection, or disposal of semantic blocks; the coordinate movement occurs in conjunction with changes in relationships between semantic blocks, changes in timestamps, or changes in probability; and the control unit identifies the degree of semantic change by comparing changes in distance or position between coordinates at different points in time corresponding to the same semantic block.A triple-layer fusion-based perceptual semantic space coordinate system characterized by using the degree of semantic change as a criterion for determining whether to maintain, update, branch, or discard a semantic block, controlling the creation and execution of a transaction regarding the semantic block or relationship according to the determination result, recording semantic block relationship information including NAQL execution results in a storage unit, and controlling the semantic block, relationship, timestamp, probability, and NAQL execution results to be utilized as reference data for subsequent transaction processing, verification, or quality control, thereby ensuring that NAQL constraint-based semantic block search and relationship management are consistently performed within a single perceptual semantic space. Claim 12 A triple-layer fusion-based perceptual semantic space coordinate system comprises: a user terminal including a display unit and a user input unit; and a server connected to the user terminal via a wired or wireless network; wherein the server includes a communication unit, a storage unit, and a control unit; wherein the control unit receives text or a query input from the user terminal and generates a semantic block through the operation of an LLM-based block generation unit, generates metadata including a block, a relationship, a timestamp, and a probability for the semantic block, maps the semantic block, the relationship, and the NAQL execution result to coordinates in a single perceptual semantic space, and the mapping to the coordinates is performed by generating an integrated semantic vector (z) by weighted fusion of a text embedding vector (f_text), a graph embedding vector (f_graph), and a manifold coordinate (M), wherein z = α·f_text + β·f_graph + γ·M (where α + β + γ = 1, α > 0, β > 0, γ > 0), and wherein the perceptual seman A triple-layer fusion-based perceptual semantic space coordinate system characterized by the fact that spatial coordinates are configured to be movable according to the creation, modification, connection, or disposal of semantic blocks, said coordinate movement occurs in conjunction with changes in relationships between semantic blocks, timestamp changes, or probability changes, said control unit identifies the degree of semantic change by comparing distance or position changes between coordinates at different points in time corresponding to the same semantic block, said degree of semantic change is used as a criterion for determining whether to maintain, update, branch, or discard semantic blocks, and by controlling the creation and execution of transactions regarding semantic blocks or relationships according to the determination result, the consistency and quality of the semantic structure are managed based on changes in perceptual semantic space coordinates. Claim 13 delete

Images