Method and Apparatus for Supporting Decision Making Based on Ontology-Based Expert Knowledge Structuring and Intelligent Reasoning Strategy Selection

Abstract

A method for providing a response strategy based on an intelligent inference strategy selection according to one technical aspect of the present invention is performed in a response strategy providing device and comprises the steps of: collecting domain data included in one or more data sources and preprocessing said domain data into an analyzable form; receiving said user query and obtaining at least a portion of said user query and related information from said domain data based on said user query; selecting an inference strategy including at least one of a plurality of inference methods based on said user query and said related information; and processing said related information according to said selected inference strategy to generate a response strategy.

Description

Technology Field

[0001] The present invention relates to domain data processing and artificial intelligence-based decision support technology. More specifically, it relates to an intelligent inference strategy selection-based response strategy providing technology that collects domain data included in one or more data sources and preprocesses it into an analyzable form, obtains relevant information from the domain data based on a user query, selects an inference strategy including at least one of a plurality of inference methods based on the user query and relevant information, and generates a response strategy thereto, thereby automating the derivation of a response strategy suitable for a situation or query and providing explainability regarding the validity of the response strategy to improve the reliability of decision-making. Background Technology

[0002] In today's business environments of enterprises, public institutions, and various organizations, large-scale domain data from diverse sources—such as public data, sensor and observation data, operational logs, policy documents, reports, and manuals—is being accumulated. Consequently, there is a continuously increasing demand for analysis and decision support to assess situations and derive response strategies based on this domain data. Particularly in fields such as disaster response, facility operations, security monitoring, supply chain management, and policy enforcement, there is a need for intelligent systems that go beyond simply providing statistical summaries or status visualizations. These systems must compare possible response strategies based on given situations or user queries, recommend the optimal strategy, and provide the rationale behind the derived results in an explainable format.

[0003] However, domain data is characterized by a mixture of structured and unstructured data, changes in structure and form during collection and preprocessing, and complex interrelationships between data. Consequently, data often exists in a dispersed, multi-layered silo structure, and the lack of explicit connection structures between data makes it difficult to perform consistent inference or compare strategies for the same situation. Furthermore, when the relationships between various data sources are not clearly expressed, it is difficult to provide the rationale for why response strategies were derived, and there may be limitations in ensuring the reliability of the results.

[0004] Meanwhile, with the proliferation of analysis technologies utilizing large language models, the automation of text-based processing—such as document summarization, question-answering, and explanation generation—has become possible to a certain extent. However, conventional large language model-based analysis has limitations in deriving response strategies by clearly reflecting structural conditions, such as constraints inherent in the data, realistic operational rules, resource limits, and objective functions. For example, when there are rules or prohibitions that must be strictly adhered to in specific situations, it may be advantageous to perform suitability verification through rule-based inference; however, conventional systems lack clear criteria for automatically selecting such inference methods based on queries or situations. Furthermore, while it may be beneficial to predict risk or success probabilities through probability-based inference when sufficient historical data has been accumulated, and numerical optimization-based analysis may be required in situations demanding the optimization of resource allocation, time, and cost, conventional systems often rely on users to manually select the appropriate inference method based on the nature of the problem or depend on a single method, which can lead to reduced accuracy and consistency in analysis results.

[0005] Furthermore, conventional technology has a problem in that even if it provides the analysis results themselves, it fails to provide sufficient explanatory capabilities regarding the data, relationships, or constraints upon which those results were derived. For example, even if a specific response strategy is recommended to a user, if it is not structurally presented how that strategy is linked to specific event information, environmental information, resource information, past cases, or documentary evidence, the user faces limitations in verifying or trusting the results. Prior art literature

[0006] Korean Patent Publication No. 10-0429792 (Date of publication: April 21, 2004) The problem to be solved

[0007] The present invention aims to solve the aforementioned problems. An embodiment according to the present disclosure aims to provide a method and apparatus for providing an intelligent inference strategy selection-based response strategy that can alleviate the burden of manually selecting an inference method according to the nature of the problem, improve the level of automation for deriving response strategies, and enhance the reliability of decision-making by providing explainability regarding the validity of the response strategy. This is achieved by collecting domain data included in one or more data sources and preprocessing it into an analyzable form, obtaining relevant information from the domain data based on a user query, selecting an inference strategy including at least one of a plurality of inference methods based on the user query and the relevant information, processing the relevant information according to the selected inference strategy to generate a response strategy, and then outputting the response strategy along with supporting information indicating the basis for generating the response strategy.

[0008] Another objective of the present invention is to provide a technology for providing response strategies that enables the derivation of response strategies considering complex relationships in domain data and improves the consistency of analysis results, in order to mitigate the problem where it is difficult to obtain user queries and related information based on consistent standards in an environment where structured and unstructured data are mixed, and where the results of deriving response strategies may become unstable because relationships and constraints between data are not explicitly reflected. This is achieved by configuring the technology to generate a domain knowledge structure that structures domain data and to perform a process of searching for and obtaining related information from the domain knowledge structure based on user queries.

[0009] Another embodiment according to the present invention has another objective of providing an explainable response strategy providing technology that can provide the basis for deriving a response strategy to the user and improve the possibility of interpreting and verifying the result, by configuring the technology to provide at least some of the relevant information used in the process of generating the response strategy, documentary evidence information indicating the source of the relevant information, and intermediate inference results generated during the process of generating the response strategy as supporting information, in order to alleviate the problem where it is difficult to verify and ensure the reliability of the result because the conventional method of providing analysis results fails to clearly present the reason for deriving the response strategy or the judgment path.

[0010] However, the problems to be solved in this disclosure are not limited to those mentioned above, and may be expanded in various ways without departing from the spirit and scope of this disclosure. means of solving the problem

[0011] One technical aspect of the present invention proposes a method for providing a response strategy based on an intelligent inference strategy selection. The method for providing a response strategy based on an intelligent inference strategy selection is performed in a response strategy providing device and comprises the steps of: collecting domain data included in one or more data sources and preprocessing the domain data into an analyzable form; receiving the user query and obtaining at least a portion of information related to the user query from the domain data based on the user query; selecting an inference strategy including at least one of a plurality of inference methods based on the user query and the related information; and processing the related information according to the selected inference strategy to generate a response strategy.

[0012] In one embodiment, the step of selecting an inference strategy including at least one of the plurality of inference methods may include: a step of extracting query characteristic information from a user query; and a step of calculating suitability information for each of the plurality of inference methods based on the query characteristic information and related information, and the inference strategy may be selected based on the suitability information.

[0013] In one embodiment, the query characteristic information may include at least one of a coercive expression, a goal-oriented expression, an uncertainty expression, and a temporal expression, and the step of selecting an inference strategy including at least one of the plurality of inference methods may further include the step of selecting at least one of generative inference including rule-based inference, probability-based inference, numerical optimization-based analysis, and graph-based search among the plurality of inference methods according to the query characteristic information.

[0014] In one embodiment, the step of obtaining at least some of the information related to the user query from the domain data may include: a step of structuring the domain data to generate a domain knowledge structure; and a step of searching for and obtaining the information related to the domain knowledge structure based on the user query.

[0015] In one embodiment, the domain knowledge structure may include an ontology and a knowledge graph generated based on the ontology, and the step of generating the domain knowledge structure may include: providing an input including unstructured document data among the domain data to a large language model to extract at least one of a domain concept, a relationship between concepts, and a constraint candidate; and generating an ontology based on the extraction result.

[0016] In one embodiment, the knowledge graph may be composed of graph data in the form of triples based on entities identified from domain data and relationships between entities using an ontology as a schema.

[0017] In one embodiment, together with the response strategy, the method may further include the step of outputting basis information indicating the basis on which the response strategy was generated.

[0018] In one embodiment, the basis information may include at least one of the relevant information used in the process of generating a response strategy, document basis information indicating the source of the relevant information, and at least one of the intermediate inference result produced in the process of generating a response strategy.

[0019] Another technical aspect of the present invention proposes a response strategy providing device based on intelligent inference strategy selection. The response strategy providing device based on intelligent inference strategy selection comprises at least one processor; and a memory for storing instructions. When the instructions are executed by the at least one processor, the processor performs the following operations: collecting domain data included in one or more data sources and preprocessing the domain data into an analyzable form; receiving the user query and obtaining at least a portion of information related to the user query from the domain data based on the user query; selecting an inference strategy including at least one of a plurality of inference methods based on the user query and the related information; and processing the related information according to the selected inference strategy to generate a response strategy. Effects of the invention

[0020] According to various embodiments of the present disclosure, domain data included in one or more data sources is collected and preprocessed into an analyzable form; relevant information is obtained from the domain data based on a user query; an inference strategy including at least one of a plurality of inference methods is selected based on the user query and the relevant information; relevant information is processed according to the selected inference strategy to generate a response strategy; and basis information indicating the basis for generating the response strategy is output along with the response strategy. This provides the effect of mitigating the limitations of the conventional method, where a user manually selects an inference method according to the nature of the problem or relies on a single method of inference. Accordingly, the reliance on manual work in the process of deriving a response strategy is reduced, and the level of automation for generating response strategies to queries or changes in situations can be improved.

[0021] Furthermore, according to various embodiments of the present disclosure, in the process of obtaining relevant information from domain data based on a user query, the domain knowledge structure that structures the domain data can be generated and the relevant information can be retrieved and obtained from the domain knowledge structure. This provides the effect of treating entities, relationships, and constraints associated with the user query in a consistent structure even in an environment where structured and unstructured data are mixed. Accordingly, it is possible to mitigate problems such as unstable analysis results due to the non-explicit reflection of relationships between data, or reduced consistency in strategy derivation due to conflicts in information originating from different data sources.

[0022] In addition, according to various embodiments of the present disclosure, since it can be configured to output supporting information along with a response strategy, the user is provided with the effect of verifying at least some of the relevant information used in the process of generating the response strategy, documentary supporting information indicating the source of the relevant information, and intermediate inference results based on this information. Accordingly, the verifiability of the results is improved, and the reliability of the response strategy can be enhanced in environments requiring explainability, such as regulatory compliance, ensuring safety, or public decision-making.

[0023] In addition, according to various embodiments of the present disclosure, evidence information can be configured to be structured into a graph linked to a response strategy and visualized for output, thereby providing the effect of intuitively understanding the decision path from which the response strategy was derived based on relationships. Accordingly, the ease of interpretation of results provided to the user is improved, and the burden required for reviewing analysis results and subsequent decision-making can be reduced.

[0024] The effects obtainable from the present invention are not limited to those mentioned above, and other unmentioned effects will be obvious to those skilled in the art to which the present invention belongs from the description below. Brief explanation of the drawing

[0025] FIG. 1 is a schematic diagram illustrating the overall configuration of a response strategy providing system according to one embodiment of the present invention. FIG. 2 is a flowchart illustrating the overall procedure of a method for providing a response strategy based on an intelligent reasoning strategy selection according to an embodiment of the present invention. FIG. 3 is a block diagram illustrating the detailed configuration of a response strategy providing device according to one embodiment of the present invention. FIG. 4 shows the system architecture of a response strategy providing system according to one embodiment of the present invention. FIG. 5 is a flowchart illustrating a reasoning strategy selection procedure according to one embodiment of the present invention. FIG. 6 is a flowchart illustrating a procedure for generating a domain knowledge structure according to an embodiment of the present invention. FIG. 7 is a flowchart illustrating a response strategy and a procedure for generating basis information according to an embodiment of the present invention. FIG. 8 is a diagram illustrating an example in which a response strategy providing system according to one embodiment of the present invention is applied to a wildfire disaster response scenario. FIG. 9 is a flowchart illustrating a multi-score based inference strategy selection procedure according to an embodiment of the present invention. FIG. 10 is a flowchart illustrating a procedure for synthesizing multiple engine results and resolving constraint-priority collisions according to an embodiment of the present invention. Specific details for implementing the invention

[0026] Hereinafter, embodiments of the present disclosure are described in detail with reference to the drawings so that those skilled in the art can easily practice them. However, the present disclosure may be embodied in various different forms and is not limited to the embodiments described herein. In relation to the description of the drawings, the same or similar reference numerals may be used for identical or similar components. Furthermore, in the drawings and related descriptions, descriptions of well-known functions and configurations may be omitted for clarity and brevity.

[0027] The various embodiments of this document and the terms used therein are not intended to limit the technical features described in this document to specific embodiments, and should be understood to include various modifications, equivalents, or substitutions of said embodiments. In connection with the description of the drawings, similar reference numerals may be used for similar or related components. The singular form of a noun corresponding to an item may include one or more of said items unless the relevant context clearly indicates otherwise. In this document, phrases such as "A or B," "at least one of A and B," "at least one of A or B," "A, B or C," "at least one of A, B and C," and "at least one of A, B, or C" may each include any one of the items listed together in the corresponding phrase, or all possible combinations thereof. Terms such as "first," "second," or "first" or "second" may be used simply to distinguish said components from other said components and do not limit said components in any other aspect (e.g., importance or order). Where any (e.g., 1st) component is referred to as "coupled" or "connected" to another (e.g., 2nd) component, with or without the terms "functionally" or "communicationly," it means that said any component may be connected to said other component directly (e.g., via a wire), wirelessly, or through a third component.

[0028] The term “module” as used in the various embodiments of this document may include a unit implemented in hardware, software, or firmware, and may be used interchangeably with terms such as logic, logic block, component, or circuit, for example. A module may be a component formed integrally, or a minimum unit of said component or a part thereof that performs one or more functions. For example, according to one embodiment, a module may be implemented in the form of an application-specific integrated circuit (ASIC).

[0029] Various embodiments of this document may be implemented as software (e.g., a program) comprising one or more instructions stored in a storage medium (e.g., memory) readable by a machine or device. For example, the processor of the machine or device may call at least one of the one or more instructions stored from the storage medium and execute it. This enables the machine to operate to perform at least one function according to the at least one called instruction. The one or more instructions may include code generated by a compiler or code that can be executed by an interpreter. The storage medium readable by a machine may be provided in the form of a non-transitory storage medium. Here, "non-transitory" simply means that the storage medium is a tangible device and does not contain a signal (e.g., electromagnetic waves), and this term does not distinguish between cases where data is stored semi-permanently and cases where it is stored temporarily in the storage medium.

[0030] According to one embodiment, the method according to the various embodiments disclosed herein may be provided by being included in a computer program product. The computer program product may be traded between a seller and a buyer as a product. The computer program product may be distributed in the form of a device-readable storage medium (e.g., compact disc read-only memory (CD-ROM)), or distributed online (e.g., download or upload) through an application store or directly between two user devices (e.g., smartphones). In the case of online distribution, at least a portion of the computer program product may be temporarily stored or temporarily created on a device-readable storage medium, such as the memory of a manufacturer's server, an application store's server, or a relay server.

[0031] According to various embodiments, each component (e.g., module or program) of the components described above may include a singular or multiple entities, and some of the multiple entities may be separated and placed in other components. According to various embodiments, one or more of the components or operations of the aforementioned components may be omitted, or one or more other components or operations may be added. Generally or additionally, multiple components (e.g., module or program) may be integrated into a single component. In this case, the integrated component may perform one or more functions of each of the multiple components in the same or similar manner as those performed by the corresponding component among the multiple components prior to integration. According to various embodiments, operations performed by the module, program, or other components may be executed sequentially, in parallel, iteratively, or heuristically, or one or more of the operations may be executed in a different order, omitted, or one or more other operations may be added.

[0033] The meanings of the key terms used in this specification are as follows, and each term is defined from a functional perspective to aid in understanding the invention. The following definitions of terms are intended to describe embodiments of the invention and should not be interpreted as unduly limiting the scope of the invention.

[0034] In the present invention, "response strategy" refers to a set of feasible or recommendable response measures, selectable strategy candidates, scenarios, or actions derived according to one or more judgment criteria regarding a user query or a given situation. A response strategy may include not only a single action instruction but also a set of selectable strategies among multiple alternatives, a step-by-step implementation plan, or a list of prioritized strategies, and is not necessarily limited to a specific domain or concrete action.

[0035] In the present invention, "inference strategy" refers to the selection or combination of information processing and inference methods applied to derive a response strategy. An inference strategy does not refer to a single fixed algorithm, but includes processing approaches selected according to the characteristics of a given user query, the structure of related information, and constraints, and may include rule-based inference, probability-based inference, numerical optimization-based analysis, generative inference, or a combination thereof.

[0036] In the present invention, "query characteristic information" refers to semantic or functional attribute information extracted from the content of a user query. The query characteristic information may include judgment cues such as compulsion, goal orientation, level of uncertainty, temporal orientation, or similar factors included in the query, and is utilized as input information for selecting or adjusting inference strategies.

[0037] "Related information" in the present invention refers to information that is semantically or functionally related to a user query among information collected from one or more data sources or obtained through a domain knowledge structure to respond to a user query. Related information may include structured data contained in domain data, unstructured document data, entity and relationship information of a knowledge graph, rule or constraint information, past case information, and combinations thereof. Related information may be selectively used depending on the characteristics of the user query and may be utilized as input information in the process of selecting an inference strategy or generating a response strategy.

[0038] In the present invention, "suitability information" refers to comparison or evaluation information calculated to indicate the extent to which each of a plurality of inference methods is suitable for a user query and related information. Suitability information may be expressed in the form of a quantitative score, relative priority, grade, or similar form, and is not limited to a specific formula or calculation method.

[0039] In the present invention, "domain knowledge structure" refers to an information structure that structurally represents concepts, relationships between concepts, attributes, and constraints included in domain data. The domain knowledge structure may include an ontology, a knowledge graph, or a structure functionally equivalent thereto, and is utilized to search for information related to user queries or as a basis for selecting inference strategies.

[0040] In the present invention, "foundational information" refers to a set of information provided to explain the reason for the generation of a response strategy or the decision path. The foundational information may include relevant information used in deriving the response strategy, documentary evidence indicating the source of said information, intermediate inference results, or a combination thereof, and is provided for the purpose of explaining the validity or reliability of the response strategy.

[0041] In the present invention, "graph structure" refers to a data structure that represents basis information or a domain knowledge structure in the form of nodes and relationships. The graph structure is utilized as a means to visually or structurally represent the relationship between a response strategy and related information, and may include a part of the entire knowledge graph or a partial structure selectively configured to provide basis.

[0043] FIG. 1 illustrates the overall configuration of a response strategy providing system based on intelligent reasoning strategy selection according to one embodiment of the present invention.

[0044] As illustrated in FIG. 1, the response strategy providing system includes a user terminal (10), a response strategy providing device (100), and one or more data sources (200), which can be interconnected via a network. The network may include a wired or wireless communication network, the Internet, or a dedicated network, and is illustrated in FIG. 1 as a cloud for ease of understanding.

[0045] A user terminal (10) is a device for receiving a query from a user and receiving a response strategy or basis information generated from a response strategy providing device (100). The user terminal (10) may be a desktop computer, a laptop computer, a mobile terminal, a tablet terminal, or a terminal functionally equivalent thereto, and the user query may be entered in the form of natural language text, a selection input form, or a combination thereof.

[0046] A response strategy providing device (100) is a core component of the present invention and is configured to generate a response strategy by analyzing domain data collected from a data source (200) based on a user query received from a user terminal (10) and selecting an inference strategy including at least one of a plurality of inference methods.

[0047] The response strategy providing device (100) derives a response strategy based on user query and related information, and, according to an embodiment, can provide the response strategy along with basis information indicating the basis for the generation of the response strategy to the user terminal (10).

[0048] The data source (200) means one or more data providers that provide domain data that the response strategy providing device (100) references to generate a response strategy.

[0049] The data source (200) may include a public database, a sensor and observation system, an operation log storage, a document storage system, or an external information provision system, and may include both structured data and unstructured data. The data source (200) may be configured as a single system or distributed as a plurality of heterogeneous systems.

[0050] The response strategy providing device (100) collects domain data from a data source (200), preprocesses it into an analyzable form, and then obtains user query and related information from the domain data based on a user query received from a user terminal (10). Subsequently, the response strategy providing device (100) selects an inference strategy including at least one of a plurality of inference methods based on the user query and related information, and generates a response strategy by processing the related information according to the selected inference strategy.

[0051] Additionally, the response strategy providing device (100) may, according to the embodiment, generate at least some of the relevant information used in the process of generating the response strategy, document-based information indicating the source of the relevant information, or intermediate inference results as source information, and provide this to the user terminal (10). Accordingly, the user can verify not only the result of the simple response strategy but also the reason why the response strategy was derived.

[0052] The system configuration illustrated in FIG. 1 is merely an example to aid in understanding the present invention, and the technical concept of the present invention is not limited thereto. For example, the response strategy providing device (100) may be implemented as a single server, or as a plurality of servers distributed by function or a cloud-based environment, and the method of connection with the data source (200) may also be expanded in various forms. Such variations and expansions are included within the technical concept and scope of rights of the present invention.

[0054] FIG. 2 is a flowchart illustrating the overall procedure of a method for providing a response strategy based on an intelligent inference strategy selection according to an embodiment of the present invention.

[0055] The procedure according to FIG. 2 includes a processing flow for generating a response strategy that responds to a user query based on domain data, and, according to an embodiment, providing the basis for generating the response strategy.

[0056] Referring to FIG. 2, in step S210, the response strategy providing device (100) collects domain data included in one or more data sources (200). The domain data may be structured data, unstructured data, or a combination thereof, and the data source (200) may include a public database, a sensor and observation system, an operation log storage, a document storage system, or an external information providing system. Additionally, the response strategy providing device (100) may record the time of collection of the domain data, the scope of collection, identification information of the data source (200), and whether collection failed, thereby ensuring the stability of processing in subsequent steps.

[0057] In step S220, the response strategy providing device (100) preprocesses the domain data collected in step S210 into an analyzable form. Preprocessing may include data format conversion, missing value processing, unit conversion, duplicate removal, time synchronization, text normalization, or metadata assignment. For example, sensor observation data may be resampled and normalized based on the time axis, and document data may be structured by separating paragraphs, tables, and numerical information.

[0058] In one embodiment, if the preprocessing result does not satisfy the preset format requirements, the response strategy providing device (100) may identify the cause of the error and request the re-collection of the corresponding domain data, or perform re-preprocessing by changing the preprocessing parameters.

[0059] In step S230, the response strategy providing device (100) receives a user query. The user query may be input from a user terminal (10) and may be in the form of natural language text, a selection input, or a combination thereof. For example, the user may input queries such as "what response is appropriate in the current situation," "how resources should be deployed," or "why that strategy was selected."

[0060] In one embodiment, the response strategy providing device (100) can standardize the user query using a term mapping rule or a thesaurus when the domain term included in the user query is inconsistent with the term system of the domain data.

[0061] In step S240, the response strategy providing device (100) obtains at least some of the user query and related information from the domain data obtained in steps S210 and S220 based on the user query received in step S230. The related information may include data, constraints, case information, or a combination thereof that are semantically or functionally associated with the user query, and may be a part of the domain data that is selectively extracted to respond to the user query, rather than the entire domain data. For example, if the user query is about "resource placement," the related information may include resource location data, availability data, travel time data, and operation constraints.

[0062] In one embodiment, the response strategy providing device (100) can structure domain data to generate a domain knowledge structure. The response strategy providing device (100) can retrieve and obtain relevant information from the domain knowledge structure based on a user query.

[0063] In one embodiment, the domain knowledge structure may include an ontology and a knowledge graph generated based on the ontology. The response strategy providing device (100) may provide an input containing unstructured document data among the domain data to a large language model to extract at least one of a domain concept, a relationship between concepts, and a constraint candidate. The response strategy providing device (100) may generate an ontology based on the extraction result. Here, the knowledge graph may be composed of graph data in the form of triples based on entities identified from the domain data and relationships between entities using the ontology as a schema.

[0064] For example, if the unstructured document data is a manual or a regulation document, the response strategy providing device (100) can extract prohibition conditions or priority conditions within the sentence as constraint candidates and reflect them as constraint attributes of the ontology, and in the knowledge graph, the corresponding constraints can be stored in the form of triples so as to be connected to specific entities or relationships.

[0065] In step S250, the response strategy providing device (100) selects an inference strategy including at least one of a plurality of inference methods based on the user query received in step S230 and the relevant information obtained in step S240. A plurality of inference methods refers to a set of different types of inference or analysis approaches applicable to the same user query, and an inference strategy refers to a selection or combination of at least one of the plurality of inference methods.

[0066] In one embodiment, the response strategy providing device (100) can extract query characteristic information from a user query. The response strategy providing device (100) can calculate suitability information for each of a plurality of inference methods based on the query characteristic information and related information. The response strategy providing device (100) can select an inference strategy based on the suitability information.

[0067] In one embodiment, the query characteristic information may include at least one of a coercive expression, a goal-oriented expression, an uncertainty expression, and a temporal expression. The response strategy providing device (100) may select at least one of generative inference including rule-based inference, probability-based inference, numerical optimization-based analysis, and graph-based search among a plurality of inference methods according to the query characteristic information.

[0068] For example, if a user query includes expressions such as "must," "prohibited," or "not allowed," and a coercive expression is detected, and the relevant information includes a regulation or rule, the response strategy providing device (100) can calculate the suitability information of the rule-based inference to be greater than or equal to a preset threshold and select an inference strategy that includes the rule-based inference.

[0069] In another example, if a user query includes expressions related to "probability of success," "risk," or "probability of occurrence," the response strategy providing device (100) can calculate the suitability information of the probability-based inference to be greater than or equal to a preset threshold and select an inference strategy including the probability-based inference.

[0070] As another example, if a user query includes expressions related to "minimum cost," "shortest time," or "optimal resource placement," the response strategy providing device (100) can calculate the suitability information of the numerical optimization-based analysis to be greater than or equal to a preset threshold and select an inference strategy including the numerical optimization-based analysis.

[0071] In another example, when a user query requests "basis," "reason," or "related document," or when the related information is composed of unstructured document data, the response strategy providing device (100) can calculate the suitability information of generative inference including graph-based search to be greater than or equal to a preset threshold, and select an inference strategy including generative inference including graph-based search.

[0072] In step S260, the response strategy providing device (100) generates a response strategy by processing the relevant information obtained in step S240 according to the inference strategy selected in step S250.

[0073] A response strategy may include a set of actionable or recommendable response measures, selectable strategy candidates, scenarios, or actions, and may be generated as a single result or multiple results assigned a priority. For example, a response strategy providing device (100) may generate a response strategy that excludes strategy candidates that violate a constraint by determining whether the constraint has been violated according to an inference strategy including rule-based inference, and may generate a response strategy that places strategy candidates with a risk level lower than a preset standard at a higher priority according to an inference strategy including probability-based inference.

[0074] For example, if an inference strategy including numerical optimization-based analysis is selected, the response strategy providing device (100) may place a response strategy in which the objective function value is smaller than a preset standard at a higher priority. For another example, if a generative inference including graph-based search is selected, the response strategy providing device (100) may generate a description of the response strategy based on document phrases retrieved from relevant information.

[0075] In step S270, the response strategy providing device (100) outputs the response strategy along with basis information indicating the basis on which the response strategy was generated. The basis information may include at least some of the relevant information used in the process of generating the response strategy, document basis information indicating the source of the relevant information, and at least one of the intermediate inference results produced during the process of generating the response strategy. For example, the response strategy providing device (100) may provide the user terminal (10) with the basis information including an identifier of a data item used to derive the response strategy, the title and location information of a regulation document as document basis information, and a result of determining whether a constraint is satisfied or a risk level is calculated as an intermediate inference result.

[0076] In one embodiment, the response strategy providing device (100) can generate a graph structure representing a basis relationship by linking a response strategy with basis information. The response strategy providing device (100) can visualize and output the graph structure. In the graph structure, nodes may include a response strategy, related information, document basis information, or intermediate inference results, and edges may represent a relationship that contributed to the derivation of the response strategy or a relationship that satisfied a constraint.

[0077] According to the procedure according to FIG. 2, the response strategy providing device (100) generates a response strategy by selecting an inference strategy based on the characteristics of a user query and relevant information obtained from domain data, and provides supporting information along with the response strategy, thereby alleviating the burden on the user or operator to manually select an inference method and improving the explainability of the result of deriving the response strategy.

[0079] FIG. 3 is a configuration diagram illustrating the detailed configuration of a response strategy providing device according to one embodiment of the present invention.

[0080] The response strategy providing device (100) according to FIG. 3 includes functional blocks for generating a response strategy in response to a user query, and is configured so that the processing of domain data, selection of inference strategies, generation of response strategies, and provision of basis information are performed in a consistent flow.

[0081] Referring to FIG. 3, the response strategy providing device (100) includes a communication unit (110), a data processing unit (120), an inference strategy processing unit (130), a response strategy generating unit (140), and a storage unit (150). The configuration illustrated in FIG. 3 is an example to aid in understanding the present invention, and the components may be combined, separated, or functionally rearranged depending on the implementation environment.

[0082] The communication unit (110) is configured to perform data transmission and reception between a user terminal (10) and a data source (200). The communication unit (110) receives a user query from the user terminal (10), receives domain data from the data source (200), and can transmit basis information generated according to a response strategy and an embodiment generated by the response strategy generation unit (140) to the user terminal (10). For example, the communication unit (110) can store a transmission log including identification information of the user query, the time of reception, and the transmission status in the storage unit (150).

[0083] The data processing unit (120) is configured to preprocess domain data into an analyzable form and to generate a domain knowledge structure. As illustrated in FIG. 3, the data processing unit (120) includes a domain data preprocessing function and a knowledge graph generation function.

[0084] The data processing unit (120) can preprocess domain data received from the data source (200) by normalizing, formatting, processing missing values, time synchronization, or text normalization. For example, sensor observation data can be converted into standard units based on the time axis, operation logs can be standardized based on event types, and unstructured document data can be broken down into paragraphs, tables, and numerical items to be structured.

[0085] In one embodiment, the data processing unit (120) can structure domain data to generate a domain knowledge structure. The domain knowledge structure may include an ontology and a knowledge graph generated based on the ontology.

[0086] In one embodiment, the data processing unit (120) may provide an input containing unstructured document data among the domain data to a large language model to extract at least one of a domain concept, a relationship between concepts, and a constraint candidate. The data processing unit (120) may generate an ontology based on the extraction result. Here, the knowledge graph may be composed of graph data in the form of triples based on entities identified from the domain data and relationships between entities using the ontology as a schema.

[0087] The data processing unit (120) can store the domain knowledge structure in the storage unit (150) so that it can be repeatedly used in the subsequent user query processing process.

[0088] The inference strategy processing unit (130) is configured to select an inference strategy based on user queries and related information. As illustrated in FIG. 3, the inference strategy processing unit (130) includes a query characteristic analysis function and an inference strategy selection function. The inference strategy processing unit (130) receives a user query from the communication unit (110) and can obtain at least some of the user query and related information based on domain data or domain knowledge structures provided by the data processing unit (120). The related information may include data, constraints, case information, or a combination thereof that is semantically or functionally associated with the user query.

[0089] In one embodiment, the inference strategy processing unit (130) can extract query characteristic information from a user query. The query characteristic information may include at least one of a coercive expression, a goal-oriented expression, an uncertainty expression, and a temporal expression. The inference strategy processing unit (130) can calculate suitability information for each of a plurality of inference methods based on the query characteristic information and related information. Based on the suitability information, the inference strategy processing unit (130) can select an inference strategy among a plurality of inference methods that includes at least one of generative inference including rule-based inference, probability-based inference, numerical optimization-based analysis, and graph-based search.

[0090] For example, if a coercive expression included in a user query is detected and a regulation or rule is included in the related information, the inference strategy processing unit (130) can calculate the suitability information of the rule-based inference to be greater than or equal to a preset threshold and select an inference strategy including the rule-based inference.

[0091] As another example, if the user query includes expressions related to probability or risk, the inference strategy processing unit (130) can calculate the suitability information of the probability-based inference to be greater than or equal to a preset threshold and select an inference strategy that includes the probability-based inference.

[0092] As another example, when the user query includes an optimization purpose, the inference strategy processing unit (130) can calculate the suitability information of the numerical optimization-based analysis to be greater than or equal to a preset threshold and select an inference strategy that includes the numerical optimization-based analysis.

[0093] As another example, when a user query requires document-based evidence or related information is structured around unstructured document data, the inference strategy processing unit (130) can calculate the suitability information of generative inference including graph-based search to be greater than or equal to a preset threshold and select an inference strategy including generative inference including graph-based search.

[0094] The response strategy generation unit (140) is configured to generate a response strategy by processing relevant information according to the inference strategy selected by the inference strategy processing unit (130), and to provide basis information according to the embodiment. As shown in FIG. 3, the response strategy generation unit (140) includes a response strategy generation function, basis information, and graph visualization function.

[0095] The response strategy generation unit (140) can generate a response strategy by processing relevant information according to the selected inference strategy, and the response strategy may include a set of actionable or recommendable response measures, selectable strategy candidates, scenarios, or actions. For example, if an inference strategy including rule-based inference is selected, the response strategy generation unit (140) can generate a response strategy that excludes strategy candidates that violate constraints by determining whether the constraints are violated, and if an inference strategy including probability-based inference is selected, the response strategy generation unit (140) can generate a response strategy that places strategy candidates with a risk level lower than a preset standard at the top of the priority list. If an inference strategy including numerical optimization-based analysis is selected, the response strategy generation unit (140) can place response strategies with an objective function value smaller than a preset standard at the top of the priority list. If a generative inference including graph-based search is selected, the response strategy generation unit (140) can generate a description of the response strategy using document phrases retrieved from relevant information.

[0096] In one embodiment, the response strategy generation unit (140) may output a response strategy along with basis information indicating the basis for the generation of the response strategy. For example, the basis information may include at least some of the relevant information used in the process of generating the response strategy, document basis information indicating the source of the relevant information, and at least one of the intermediate inference result produced during the process of generating the response strategy. The response strategy generation unit (140) may generate a graph structure indicating the basis relationship by linking the response strategy and the basis information. The response strategy generation unit (140) may visualize the graph structure and output it to the user terminal (10). At this time, in the graph structure, nodes may include the response strategy, relevant information, document basis information, or intermediate inference result, and edges may represent a relationship in which the response strategy is supported by the corresponding node or a relationship in which constraints are satisfied.

[0097] The storage unit (150) is configured to store information used or generated during the processing of the response strategy providing device (100). The storage unit (150) can store at least some of domain data, preprocessed domain data, domain knowledge structure, related information, query characteristic information, suitability information, inference strategy, response strategy, basis information, and intermediate inference results.

[0098] The storage unit (150) can enable reuse-based processing when the same or similar user query is re-entered by storing the processing history and the result of generating the response strategy for each user query together.

[0099] The response strategy providing device (100) according to FIG. 3 is configured to preprocess domain data and generate a domain knowledge structure in a data processing unit (120), select an inference strategy based on user queries and related information in an inference strategy processing unit (130), generate a response strategy according to the selected inference strategy in a response strategy generating unit (140), and provide supporting information, thereby alleviating the burden on a user or operator to manually select an inference method according to the nature of the problem and improving the explainability of the result of deriving the response strategy.

[0101] FIG. 4 shows the system architecture of a response strategy providing system according to one embodiment of the present invention.

[0102] The architecture according to FIG. 4 includes a data collection and normalization layer, an ontology automatic generation layer, a knowledge graph automatic generation layer, an analysis and inference layer, and a strategy recommendation and rationale visualization layer, each layer being functionally implemented by a component of the corresponding strategy providing device (100) shown in FIG. 3.

[0103] Referring to FIG. 4, the hierarchical system architecture illustrated in FIG. 4 is a representation of the functional blocks of the response strategy providing device (100) illustrated in FIG. 3, which are logically decomposed. In the actual implementation, the data processing unit (120), the inference strategy processing unit (130), and the response strategy generation unit (140) execute the pipeline of each layer, the communication unit (110) provides an external interface for input and output between layers, and the storage unit (150) operates as a storage medium for sharing outputs between layers.

[0104] For example, the refined data generated in the data normalization pipeline of FIG. 4, the ontology schema generated in the ontology generation pipeline, the knowledge graph generated in the KG generation pipeline, the embeddings and document search results generated in the Graph RAG pipeline, and the evidence graph generated in the evidence graph builder can be stored in the storage unit (150), and subsequently, the inference strategy processing unit (130) or the response strategy generation unit (140) can refer to and reuse them during the user query processing process.

[0105] Layer 1, the data collection and normalization layer, includes a data collector and a data normalization pipeline. The data collector is configured to collect domain data from external data, public data, and expert knowledge, including a collector and a crawler. The data normalization pipeline is configured to perform schema transformation, quality verification, preprocessing, and normalization processing to store data in a refined structured data repository, a refined unstructured data repository, and a vector repository. The data collection and normalization functions performed in this layer are implemented by the data processing unit (120) of FIG. 3. When a data format mismatch is detected, the data normalization pipeline can unify the schema by applying pre-set transformation rules, and classify data that does not meet the standard value in the quality verification result as a target for re-preprocessing.

[0106] The vector storage labeled "VectorDB" in Layer 1 can be implemented as a vector storage area within the storage unit (150) of FIG. 3. For example, the data processing unit (120) can break down unstructured document data into paragraphs or numerical items to structure it, then generate an embedding vector for each document fragment and store it in the vector storage area of ​​the storage unit (150). When processing a user query, the inference strategy processing unit (130) can generate an embedding vector of the user query, search for document fragments in the vector storage area of ​​the storage unit (150) that have a similarity level greater than or equal to a preset threshold, and obtain the searched document fragments as related information.

[0107] Layer 2, the automatic ontology generation layer, includes an ontology generation pipeline based on a large language model. The ontology generation pipeline includes ontology candidate extraction, ontology schema verification, and storage, and the ontology candidate extraction is configured to include concept analysis, relationship extraction, and constraint extraction. The automatic ontology generation function performed in this layer is implemented by the domain knowledge structure generation function of the data processing unit (120) of FIG. 3. Ontology schema verification may include a procedure to check whether there is a circular relationship between concepts or whether there is a conflict of constraints, and if the result of the check does not satisfy the pre-set verification criteria, the data processing unit (120) may perform re-extraction by changing the input range provided to the large language model.

[0108] Layer 3, the knowledge graph automatic generation layer, includes a KG generation pipeline. The KG generation pipeline is configured to include entity and relationship extraction, mapping interpretation, RDF triple generation, metadata annotation, and KG storage, and the generated knowledge graph is configured to be stored in a graph database. The knowledge graph generation and storage functions performed in this layer are implemented by the data processing unit (120) of FIG. 3. The graph database can be implemented as a graph data storage area implemented within the storage unit (150) of FIG. 3. For example, the data processing unit (120) can configure entities identified from domain data and relationships between entities into graph data in the form of triples and store them in the graph data storage area of ​​the storage unit (150), and the inference strategy processing unit (130) can map terms included in a user query to entities to perform a graph search and obtain the graph search results as relevant information.

[0109] Layer 4 analysis and inference layer includes KG Reasoning and Graph RAG pipelines. The Graph RAG pipeline includes document search, embedding generation and vector storage, graph search, and large language model-based explanation generation, and "inference strategy selection" and "inference application" are performed as core components of the processing flow for user queries. The query characteristic analysis, suitability information calculation, inference strategy selection, and related information acquisition functions performed in this layer are implemented by the inference strategy processing unit (130) of FIG. 3, and processing and response strategy generation according to the selected inference strategy can be implemented by the response strategy generation unit (140) of FIG. 3.

[0110] For example, if a user query includes a coercive expression and the related information includes a rule document or rule, the inference strategy processing unit (130) may calculate the suitability information of the rule-based inference to be greater than or equal to a preset threshold and select an inference strategy that includes the rule-based inference. In this case, the response strategy generation unit (140) may apply the rule-based inference to determine whether there is a violation of the constraints and generate a response strategy by excluding the candidate strategy that violates the constraints.

[0111] If the user query includes risk or probability, the inference strategy processing unit (130) can calculate the suitability information of the probability-based inference to be greater than or equal to a preset threshold and select an inference strategy that includes the probability-based inference. In this case, the response strategy generation unit (140) can generate a response strategy that places strategy candidates with a risk lower than a preset standard at the top of the priority list.

[0112] If the user query includes a goal-oriented expression such as minimum cost or shortest time, the inference strategy processing unit (130) can calculate the goodness-of-fit information of the numerical optimization-based analysis to be greater than or equal to a preset threshold and select an inference strategy that includes the numerical optimization-based analysis. In this case, the response strategy generation unit (140) can place a response strategy with an objective function value smaller than a preset standard at a higher priority.

[0113] When a user query requires document evidence or explainability, the inference strategy processing unit (130) may calculate the suitability information of generative inference including graph-based search to be greater than or equal to a preset threshold and select an inference strategy including generative inference including graph-based search. In this case, the inference strategy processing unit (130) may obtain relevant information by performing document search in the vector storage area of ​​the storage unit (150) and performing graph search in the graph data storage area of ​​the storage unit (150), and the response strategy generation unit (140) may generate an explanation of the response strategy using document evidence information included in the relevant information. The LLM Explainer of FIG. 4 may be implemented as a process in which the response strategy generation unit (140) generates a part of the explanation or evidence information of the response strategy.

[0114] Layer 5, the strategy recommendation and reasoning visualization layer, includes a strategy recommender, a reasoning graph builder, and a service output. The strategy recommender is configured to generate a response strategy using the results of the analysis and inference layer as input, and the reasoning graph builder is configured to construct a reasoning graph that represents the reasoning for the generated response strategy as nodes and relationships. The service output is configured to visualize the response strategy and the reasoning graph and provide them to the user. The functions of generating a response strategy, generating reasoning information, generating a graph structure, and visualizing the graph structure output performed in this layer are implemented by the response strategy generation unit (140) of FIG. 3, and transmission and reception with the user terminal (10) are implemented by the communication unit (110) of FIG. 3.

[0115] For example, when a response strategy is generated as "Move resource A to location B," the response strategy generation unit (140) may include resource A availability data, travel time data, constraint determination results, and document basis information containing the constraints as basis information, and in the graph structure generated by the basis graph builder, the response strategy node may be visualized by being connected to the "Resource A Availability" node, the "Travel Time Criteria Satisfaction" node, and the "Regulatory Document Basis" node. At this time, the service output may be interpreted to mean that the communication unit (110) transmits the response strategy and basis information to the user terminal (10), and the user terminal (10) displays it on the screen.

[0116] The implementation relationship between the layer of Fig. 4 and the component of Fig. 3 can be more clearly identified in a user query processing scenario. For example, when a user query such as "respond in the shortest time while satisfying constraints" is input from a user terminal (10), the communication unit (110) receives the user query and transmits it to the inference strategy processing unit (130).

[0117] The inference strategy processing unit (130) can obtain relevant information by referring to domain knowledge structures and refined data stored in the storage unit (150), extract coercive expressions and goal-oriented expressions from query characteristic information, and then calculate fit information to select an inference strategy including numerical optimization-based analysis and rule-based inference.

[0118] The response strategy generation unit (140) generates a response strategy by processing relevant information according to the selected inference strategy, and may include the result of determining the constraint satisfaction of rule-based inference and the result of calculating the objective function value of numerical optimization-based analysis as intermediate inference results in the basis information.

[0119] The response strategy generation unit (140) generates and outputs basis information including document basis information and graph structure, and the communication unit (110) can transmit the response strategy and basis information to the user terminal (10).

[0120] According to the system architecture of Fig. 4, data collection and normalization, creation of domain knowledge structures, selection and application of inference strategies, creation of response strategies, and visualization of evidence are performed in a stepwise linked manner. Accordingly, an inference strategy suitable for the characteristics of the user query is selected, and evidence is provided along with the response strategy, thereby improving the explainability and reliability of the results.

[0122] FIG. 5 is a flowchart illustrating a reasoning strategy selection procedure according to one embodiment of the present invention.

[0123] The procedure according to FIG. 5 includes a processing flow for extracting query characteristic information from a user query, calculating suitability information for each of a plurality of inference methods based on the user query and related information, and selecting an inference strategy based on the suitability information. The procedure illustrated in FIG. 5 can be applied as a detailed processing of the inference strategy selection step illustrated in FIG. 2 and can be performed by the inference strategy processing unit (130) illustrated in FIG. 3.

[0124] Referring to FIG. 5, in step S510, the response strategy providing device (100) receives a user query from a user terminal (10). The user query may be input in the form of natural language text, a selection input, or a combination thereof. For example, the user query may be in the form of "what response is appropriate," "what is the priority for reducing risk," or "what is the optimal placement that satisfies the constraints."

[0125] In step S520, the response strategy providing device (100) normalizes the user query into an analyzable form. The normalization of the user query may include at least one of sentence segmentation, stop word processing, expression unification, or domain term standardization. For example, if a domain term included in the user query is inconsistent with an entity name of a domain knowledge structure, the response strategy providing device (100) may convert the term in the user query into a standard term by applying a synonym mapping rule.

[0126] In step S530, the response strategy providing device (100) extracts query characteristic information from a user query. The query characteristic information may include at least one of a coercive expression, a goal-oriented expression, an uncertainty expression, and a temporal expression. For example, if the user query includes "must," "prohibited," or "not allowed," a coercive expression may be extracted; if it includes "minimum cost" or "shortest time," a goal-oriented expression may be extracted; if it includes "possibility," "risk," or "probability," an uncertainty expression may be extracted; and if it includes "when," "deadline," or "immediately," a temporal expression may be extracted.

[0127] In step S540, the response strategy providing device (100) calculates the configuration characteristics of the user query and related information. Here, the related information refers to data, constraints, case information, or a combination thereof that is semantically or functionally associated with the user query, and includes at least some of the information obtained from domain data or domain knowledge structures to respond to the user query. For example, if the related information includes a rule document or a set of rules, a rule-centric characteristic may be calculated; if the related information includes sensor values ​​or numeric tables, a numeric data-centric characteristic may be calculated; and if the related information includes entities and relationships of a knowledge graph, a graph data-centric characteristic may be calculated.

[0128] In step S550, the response strategy providing device (100) calculates suitability information for each of the plurality of inference methods based on query characteristic information and related information. The plurality of inference methods include rule-based inference, probability-based inference, numerical optimization-based analysis, and generative inference including graph-based search. The suitability information is a value indicating the degree to which each inference method is suitable for application to a user query, and can be calculated for each inference method. Here, the suitability information can be normalized to a preset score range, and only inference methods whose suitability information is above a preset threshold can be restricted to be included as components of the inference strategy.

[0129] In step S560, the response strategy providing device (100) selects an inference strategy based on the suitability information calculated in step S550. The inference strategy may be determined to include at least one of a plurality of inference methods, may be composed of a single inference method, or may be composed of a combination of a plurality of inference methods. For example, if a coercive expression is extracted and the related information has rule-centered characteristics, an inference strategy including rule-based inference may be selected; if a goal-oriented expression is extracted and the related information has numeric data-centered characteristics, an inference strategy including numeric optimization-based analysis may be selected; if an uncertainty expression is extracted, an inference strategy including probability-based inference may be selected; and if a document-based requirement is included or the related information has unstructured document-centered characteristics, an inference strategy including generative inference including graph-based search may be selected.

[0130] In step S570, the response strategy providing device (100) outputs the selected inference strategy to a subsequent processing step. The selected inference strategy can be used as a criterion for processing relevant information in the response strategy generation procedure. For example, if the selected inference strategy includes rule-based inference and numerical optimization-based analysis, the response strategy can be generated in the subsequent step by determining whether a constraint is violated through rule-based inference and applying numerical optimization-based analysis to a set of candidates satisfying the constraint, thereby sorting candidates whose objective function value is smaller than a preset criterion to a higher priority.

[0131] According to the procedure according to FIG. 5, the response strategy providing device (100) can automatically determine a processing path that matches the nature of the problem by selecting a suitable inference strategy among a plurality of inference methods by considering the expression of the user query and the configuration of related information, and has the characteristic of improving the quality and consistency of the response strategy generation.

[0133] FIG. 6 is a flowchart illustrating a procedure for generating a domain knowledge structure according to an embodiment of the present invention.

[0134] The procedure according to FIG. 6 includes a processing flow that generates a domain knowledge structure including an ontology and a knowledge graph based on domain data, and stores and prepares it so that it can be used for searching for relevant information in response to a user query. The procedure illustrated in FIG. 6 can be applied as a detailed processing step for the relevant information acquisition step and the domain data preprocessing step illustrated in FIG. 2, and can be performed by the data processing unit (120) illustrated in FIG. 3.

[0135] Referring to FIG. 6, in step S610, the response strategy providing device (100) obtains domain data from one or more data sources (200). The domain data may include structured data and unstructured document data, and may include operational logs, sensor observations, public data, and expert knowledge documents. For example, the response strategy providing device (100) records the source, collection time, and data format of the domain data as metadata so that it can be used as document evidence information when generating evidence information later.

[0136] In step S620, the response strategy providing device (100) selects and preprocesses unstructured document data among the domain data. The preprocessing of unstructured document data may include at least one of paragraph separation, extraction of table structure, extraction of numerical items, text normalization, removal of duplicate sentences, or assignment of document identifiers. For example, when a regulation document is input, the response strategy providing device (100) may break down paragraphs into clause units and structure thresholds or criteria presented in a table by extracting them as numerical items.

[0137] In step S630, the response strategy providing device (100) generates and provides input to be provided to a large language model. The input may include some or all of the preprocessed unstructured document data and may include instruction information that specifies the extraction target range and output format for generating a domain knowledge structure. For example, the input may include a prompt template that distinguishes and outputs domain concepts, relationships between concepts, and constraint candidates.

[0138] In step S640, the response strategy providing device (100) extracts at least one of a domain concept, a relationship between concepts, and a candidate constraint from the output of a large language model. A domain concept represents a meaningful object or attribute in the domain, a relationship between concepts represents a connection or dependency between concepts, and a candidate constraint may represent a condition or prohibition condition for a specific response strategy to be allowed. For example, a "resource" concept and a "location" concept may be extracted, a relationship such as "resources are placed in a location" may be extracted, and a candidate constraint such as "placement is possible only if the risk level is below a threshold value" may be extracted.

[0139] In step S650, the response strategy providing device (100) generates and verifies an ontology based on the extraction result of step S640. The ontology may include a schema that structurally defines domain concepts, relationships between concepts, and constraints. Verification of the ontology may include at least one of whether a definition of a relationship type is missing, whether a prohibited relationship exists, whether there is a constraint conflict, or whether a circular relationship exists. For example, if a constraint conflict is detected as a result of verification, the response strategy providing device (100) may store the conflicting constraint candidates separately or disable them by applying a priority rule.

[0140] In step S660, the response strategy providing device (100) identifies entities and relationships between entities from domain data. An entity may refer to an individual instance corresponding to a domain concept of an ontology, and relationships between entities may be represented as connections between entities according to relationship types defined in the ontology. For example, a specific resource identifier may be identified as an entity, and a relationship in which the resource is placed at a specific location may be identified. As an example, during the process of identifying entities and relationships, if a person with the same name or duplicate entities is detected, the response strategy providing device (100) may perform merging processing according to an identifier rule or an attribute similarity criterion.

[0141] In step S670, the response strategy providing device (100) constructs and stores a knowledge graph based on entities and relationships. The knowledge graph can be composed of graph data in the form of triples based on entities identified from domain data and relationships between entities, using an ontology as a schema. For example, a triple of "Entity A, Relationship R, Entity B" can be generated and stored as graph data. As an example, during the storage process of the knowledge graph, the response strategy providing device (100) may store the source document identifier and extraction time of the triple together as metadata to be used for generating evidence information later.

[0142] In step S680, the response strategy providing device (100) stores the ontology and knowledge graph as a domain knowledge structure and prepares it for use in searching for relevant information regarding a user query. The storage of the domain knowledge structure may be performed by writing it to a storage unit (150), and the preparation for searching may include a process of creating at least one of an entity index, a relationship index, or a triple index. As an example, the preparation for searching may include a process of creating a pre-mapping table to quickly generate mapping candidates when a term included in the user query is inconsistent with a concept or entity of the domain knowledge structure.

[0143] According to the procedure according to FIG. 6, the response strategy providing device (100) has the characteristic of extracting domain concepts, relationships, and constraint candidates using a large language model and generating an ontology, constructing a knowledge graph using the ontology as a schema, and storing it as a domain knowledge structure so that the acquisition of relevant information for user queries is performed in a structured manner. In addition, since the domain knowledge structure can be stored and indexed for reuse, it provides the effect of improving the efficiency of searching for relevant information in the processing of repetitive user queries.

[0145] FIG. 7 is a flowchart illustrating the procedure for generating a response strategy and outputting basis information according to an embodiment of the present invention.

[0146] The procedure according to FIG. 7 includes a processing flow that takes a selected inference strategy as input to generate a response strategy corresponding to a user query, and constructs and outputs basis information indicating the basis on which the response strategy was generated.

[0147] The procedure illustrated in FIG. 7 can be applied as a detailed processing step for the response strategy generation step and the basis information output step illustrated in FIG. 2, and can be performed by the response strategy generation unit (140) illustrated in FIG. 3, and transmission and reception with the user terminal (10) can be performed by the communication unit (110).

[0148] Referring to FIG. 7, in step S710, the corresponding strategy providing device (100) receives a selected inference strategy as input. The selected inference strategy may include at least one of a plurality of inference methods and represents a processing path selected based on the characteristics of the user query and the configuration of related information. The selected inference strategy may be determined to include at least one of generative inference including rule-based inference, probability-based inference, numerical optimization-based analysis, or graph-based search.

[0149] In step S720, the response strategy providing device (100) obtains relevant information in response to a user query. The relevant information includes information that is semantically or functionally related to the user query, among information collected from one or more data sources or obtained through a domain knowledge structure. The response strategy providing device (100) may search for and obtain relevant information by referring to a domain knowledge structure including refined data, an ontology, or a knowledge graph stored in a storage unit (150). For example, the relevant information may include document-based information, and the document-based information may include a document identifier, document phrase, collection time, or data source indicating the source of the relevant information.

[0150] In step S730, the response strategy providing device (100) processes relevant information by applying a selected inference strategy. If the selected inference strategy includes rule-based inference, the response strategy providing device (100) can determine whether a candidate strategy is acceptable based on rules or constraints included in the relevant information. If the selected inference strategy includes probability-based inference, the response strategy providing device (100) can calculate the risk of the candidate strategy based on probability variables or risk indicators included in the relevant information.

[0151] If the selected inference strategy includes numerical optimization-based analysis, the response strategy providing device (100) can calculate an objective function value based on numerical parameters included in the relevant information and perform optimization calculations. If the selected inference strategy includes generative inference including graph-based search, the response strategy providing device (100) can perform graph search using entities and relationships of a knowledge graph and perform document search in a vector repository to collect candidate grounds corresponding to the user query.

[0152] In step S740, the response strategy providing device (100) generates an intermediate inference result produced during the process of applying the inference strategy. The intermediate inference result may include at least one of the following: a result of determining the satisfaction of constraints in rule-based inference, a risk calculation value in probability-based inference, an objective function value or a set of explored candidate solutions in numerical optimization-based analysis, path information in graph-based search results, or document search results. For example, the intermediate inference result may be used as key material to explain the process by which the response strategy was derived when constructing the evidence information later.

[0153] In step S750, the response strategy providing device (100) generates a response strategy based on an intermediate inference result. The response strategy may include a set of feasible or recommendable response measures, selectable strategy candidates, scenarios, or actions, and may be generated as a single result or multiple results assigned a priority. For example, a response strategy may be generated by excluding violation candidates based on the result of determining the satisfaction of constraints in rule-based inference, and arranging candidates whose objective function value in numerical optimization-based analysis is smaller than a preset criterion to a higher priority.

[0154] In step S760, the response strategy providing device (100) configures basis information representing the basis on which the response strategy was generated. The basis information may include at least some of the relevant information used in the process of generating the response strategy, document basis information representing the source of the relevant information, and at least one of the intermediate inference results produced during the process of generating the response strategy. The response strategy providing device (100) may generate a graph structure representing basis relationships by linking the components of the response strategy and the basis information. For example, in the graph structure, nodes may include the response strategy, relevant information, document basis information, or intermediate inference results, and edges may represent relationships or constraint satisfaction relationships that contributed to the derivation of the response strategy. For example, each node of the graph structure may be configured to be connected to a document identifier or knowledge graph triple so that basis tracking is possible.

[0155] In step S770, the response strategy providing device (100) outputs the response strategy and basis information. The response strategy providing device (100) can transmit the response strategy and basis information to the user terminal (10) through the communication unit (110), and the user terminal (10) can display the response strategy and basis information on a screen. For example, the basis information may include visualization results of a graph structure, and the user can check document basis information or intermediate inference results associated with the response strategy by selecting nodes of the graph structure.

[0156] According to the procedure according to FIG. 7, the response strategy providing device (100) generates a response strategy by applying a selected inference strategy, and outputs together the relevant information, document basis information, and intermediate inference result used in the process of generating the response strategy as basis information, thereby providing the basis for deriving the response strategy and improving the explainability and reliability of the result.

[0158] FIG. 8 shows a scenario in which a response strategy providing system according to one embodiment of the present invention is applied to a wildfire disaster response system.

[0159] In the scenario illustrated in FIG. 8, domain data including structured and unstructured data is collected and normalized, an ontology is automatically generated using a large language model, a knowledge graph is automatically generated using the ontology as a schema, a response strategy is generated by selecting and applying an inference strategy including multiple inference methods according to a user query, and the basis of the response strategy is visualized and provided to the user. Each function illustrated in FIG. 8 can be implemented by the data processing unit (120), inference strategy processing unit (130), response strategy generation unit (140), storage unit (150), and communication unit (110) of the response strategy providing device (100) illustrated in FIG. 3.

[0160] Referring to FIG. 8, the response strategy providing system may use data (structured and unstructured) and expert knowledge regarding the wildfire disaster response domain as input. For example, the domain data may include the location of the wildfire, helicopter location, fuel quantity, availability of deployment, firefighting team deployment and personnel information, freshwater source information, wildfire response manual, past wildfire case reports, and expert opinion memos. Such domain data may be collected from one or more data sources (200), and the response strategy providing device (100) may preprocess it into an analyzable form. The collection and preprocessing functions are performed by the data processing unit (120) of FIG. 3, and storage areas corresponding to the refined structured data storage, time series storage, unstructured data storage, and vector storage may be provided by the storage unit (150) of FIG. 3.

[0161] The "data collector and normalization (layer 1)" illustrated in Fig. 8 can be implemented as a process that preprocesses domain data of the wildfire disaster response domain through normalization, format conversion, missing value handling, time synchronization, or text normalization. For example, the location of the wildfire, the location of the helicopter, and the location of the firefighting team can be synchronized along a time axis and stored in a time-series storage, and status information such as fuel quantity and deployment availability can be stored in a structured data storage with a standard field structure. Document data such as wildfire response manuals and past case reports can be decomposed into paragraphs, tables, and numerical items and stored in an unstructured data storage. As an example, the result of decomposing document data into paragraphs can be converted into embedding vectors and stored in a vector storage, which can then be used to retrieve relevant information during user query processing.

[0162] The "automatic ontology generation (layer 2)" illustrated in Fig. 8 can be implemented as a process for generating an ontology among the domain knowledge structures for the wildfire disaster response domain. The ontology may include a schema that structurally defines domain concepts, relationships between concepts, and constraints. Fig. 8 illustrates concepts such as wildfire incident, fire line location, wind speed, helicopter, firefighting team, firefighting strategy, and water reservoir as examples of the ontology, and relationships such as "wildfire incidents are affected by wind speed," "helicopters draw water from water reservoirs," and "firefighting teams require accessible terrain conditions." Additionally, examples of constraints such as "if the wind speed exceeds a preset threshold of 8 m / s, the success rate of aerial firefighting decreases," and "if the helicopter distance value exceeds a preset threshold of 20 km, the dispatch time is delayed" are illustrated. This automatic ontology generation function can be implemented by the data processing unit (120) of FIG. 3 using a large language model to extract domain concepts, relationships, and constraint candidates from unstructured document data, and by generating and verifying an ontology based on the extraction results.

[0163] The "KG automatic generation (layer 3)" illustrated in FIG. 8 can be implemented as a process of generating a knowledge graph using an ontology as a schema. The knowledge graph can be composed of graph data in the form of triples based on entities identified from domain data and relationships between entities. FIG. 8 illustrates entities corresponding to wildfire, wind, helicopter, firefighting team, and freshwater reservoir as examples of a knowledge graph, and illustrates the influence relationship between wildfire and wind, the freshwater possibility relationship between helicopter and freshwater reservoir, the field access relationship between firefighting team and wildfire, and the topographic influence relationship between wildfire and region. This KG automatic generation function can be implemented as a process in which the data processing unit (120) of FIG. 3 identifies entities and relationships, generates RDF triples, and stores them in a graph database, and the graph database can be provided by the storage unit (150) of FIG. 3.

[0164] The "multiple inference-based analysis and result integration (layer 4)" illustrated in FIG. 8 can be implemented as an analysis layer for generating a response strategy by selecting an inference strategy that includes multiple inference methods based on a user query and processing relevant information according to the selected inference strategy. In this process, the response strategy providing device (100) receives a user query and can obtain at least some of the user query and related information from domain data or domain knowledge structures based on the user query, and can select an inference strategy that includes at least one of multiple inference methods based on the user query and related information. For example, if the user query is entered in the form of "What is the response method to extinguish a wildfire that occurred in region A?", the response strategy providing device (100) can obtain relevant information including a wildfire response manual, past similar cases, the location of the wildfire, wind speed, and the current location and availability status of helicopters and firefighting teams through document search and graph search. Additionally, the response strategy providing device (100) may select an inference strategy including at least one of rule-based inference, probability-based inference, numerical optimization-based analysis, and graph-based search, depending on the characteristics of the user query. At this time, the function of selecting the inference strategy may be implemented by the inference strategy processing unit (130) of FIG. 3.

[0165] The "strategy and action recommendation and rationale visualization" illustrated in FIG. 8 can be implemented as a layer that generates a response strategy by processing relevant information according to a selected inference strategy and provides the user with rationale information indicating the basis for the generated response strategy. The response strategy providing device (100) can generate an intermediate inference result that calculates whether constraints are satisfied, risk level, objective function value, or graph search path by applying the selected inference strategy, and can generate a response strategy based on the intermediate inference result. For example, if a constraint is applied such that the success rate of aerial firefighting decreases when the wind speed exceeds a preset threshold value, the response strategy providing device (100) can generate a response strategy that lowers the weight of aerial firefighting or prioritizes recommending other firefighting strategies. Additionally, if a constraint is applied such that the dispatch time is delayed when the distance value between the helicopter and the freshwater reservoir exceeds a preset threshold value, the response strategy providing device (100) can generate a response strategy that prioritizes the deployment of a helicopter with higher accessibility to the freshwater reservoir. These functions for generating response strategies and configuring rationale information can be implemented by the response strategy generating unit (140) of FIG. 3.

[0166] As illustrated in FIG. 8, the response strategy providing device (100) may output a response strategy along with evidence information indicating the basis on which the response strategy was generated. The evidence information may include at least some of the relevant information used in the process of generating the response strategy, document evidence information indicating the source of the relevant information, and at least one of the intermediate inference results produced during the process of generating the response strategy. FIG. 8 illustrates an evidence graph as an example of evidence information, and the evidence graph may be provided in a form that visualizes the affected elements, intermediate calculation results, and inference results. The user can verify the basis for deriving the response strategy and verify the reliability of the response strategy through the evidence graph, reports, and data references. For example, the user can select a specific node in the evidence graph to view document evidence information or knowledge graph triples corresponding to that node.

[0167] According to the scenario of FIG. 8, the present invention is characterized by improving the explainability and reliability of results by selecting an inference strategy suitable for a user query to generate a response strategy in a domain where various data types and multiple constraints are combined, such as in wildfire disaster response, and by providing the basis for generating the response strategy. In addition, since a domain knowledge structure including an ontology and a knowledge graph can be generated, stored, and reused, it provides the effect of improving the efficiency of retrieving relevant information in the processing of repetitive queries.

[0169] FIG. 9 is a flowchart illustrating a procedure for selecting an inference strategy based on multiple scores according to an embodiment of the present invention.

[0170] The embodiment illustrated in FIG. 9 aims to improve the objectivity, reproducibility, and explainability of the inference strategy selection process by numerically quantifying the characteristics of user queries and related information, and by selecting an inference strategy through normalization and threshold comparison. The function according to FIG. 9 can be implemented by the inference strategy processing unit (130) illustrated in FIG. 3.

[0171] Referring to FIG. 9, in step S910, the response strategy providing device (100) receives a user query and related information obtained in response to the user query as input. Here, the related information may include a domain knowledge structure, document data, past case data, sensor data, or calculation result data, and may be composed of information semantically related to the user query.

[0172] In step S920, the response strategy providing device (100) extracts query characteristic information from a user query and calculates a query characteristic score based on the query characteristic information. For example, the query characteristic score may be calculated based on at least one of the frequency of coercive expressions included in the user query, the presence or absence of goal-oriented expressions, the ratio of uncertainty expressions, and the inclusion or absence of temporal expressions. At this time, the query characteristic score may be expressed as a continuous value between 0 and 1 or as a pre-set score range.

[0173] In step S930, the response strategy providing device (100) calculates a data characteristic score based on the configuration of relevant information. For example, the data characteristic score may be calculated based on at least one of the ratio of structured data included in the relevant information, whether time-series data is included, the amount of unstructured document data, and the density of knowledge graph nodes and relationships. Through this, it is possible to numerically determine whether a situation is suitable for simple document-based analysis, or for graph-based exploration or numerical optimization-based analysis.

[0174] In step S940, the response strategy providing device (100) calculates a raw fit score for each of the plurality of inference methods by reflecting the query characteristic score and the data characteristic score. For example, for each generative inference method including rule-based inference, probability-based inference, numerical optimization-based analysis, and graph-based search, a raw fit score may be calculated by applying a predefined combination of weights. In this step, since each inference method is evaluated according to the same score criteria, the relative fit between the inference methods can be compared.

[0175] In step S950, the response strategy providing device (100) normalizes the raw fitness scores calculated for each inference method into a preset score range. For example, even if the raw fitness scores have different distributions, consistency in score interpretation can be ensured by normalizing them between 0 and 1 based on the minimum and maximum values. Through this, the problem of a specific inference method being excessively selected or excluded due to differences in the scale of the raw scores can be mitigated.

[0176] In step S960, the response strategy providing device (100) selects candidate inference methods by comparing a normalized fit score with a preset threshold. For example, it may be configured to select only inference methods with a normalized fit score of 0.7 or higher as candidates. By explicitly performing a threshold comparison in this way, a minimum quality standard can be guaranteed in the inference strategy selection process.

[0177] In step S970, the response strategy providing device (100) constructs an inference strategy by combining one or more selected candidate inference methods. For example, a composite inference strategy may be constructed by applying graph-based search and rule-based inference in parallel, or by performing graph-based verification after document-based generative inference. Through this, the reliability and stability of the response strategy may be improved compared to when relying on a single inference method.

[0178] In step S980, the response strategy providing device (100) outputs the configured inference strategy to the subsequent response strategy generation step. The output inference strategy can then be used in the process of generating a response strategy by processing relevant information and configuring the basis information of the response strategy.

[0179] According to an additional embodiment of FIG. 9, the inference strategy selection process does not rely on simple rule-based branching or empirical judgment, but is performed through normalization and threshold comparison by numerically evaluating query characteristics and data characteristics, thereby providing the effect of improving the objectivity, reproducibility, and explainability of the inference strategy selection. In addition, since it is possible to configure an inference strategy by combining multiple inference methods, it provides the technical effect of generating a response strategy that is more suitable for user queries even in complex domain environments.

[0181] FIG. 10 is a flowchart illustrating a multiple engine result synthesis procedure according to one embodiment of the present invention.

[0182] The embodiment illustrated in FIG. 10 aims to improve the stability and reliability of multi-engine-based inference by receiving candidate results generated by each of the multiple inference engines as input, resolving conflicts between the candidate results according to constraint priority, and then combining the objective function value and a penalty for constraint violation to produce a final composite result. The function according to FIG. 10 can be implemented by the inference strategy processing unit (130) and the corresponding strategy generation unit (140) illustrated in FIG. 3.

[0183] Referring to FIG. 10, in step S1010, the response strategy providing device (100) receives candidate results produced from a plurality of inference engines as input. The candidate results may include response strategy candidates, response action candidates, or a priority list of response strategy candidates.

[0184] For example, a plurality of inference engines may include at least two of a rule-based inference engine, a numerical optimization-based analysis engine, a probability-based inference engine, or a generative inference engine including a graph-based search.

[0185] In step S1020, the response strategy providing device (100) converts engine-specific candidate results into a common representation and standardizes them. Standardization may include processing to align the candidate results into a set of fields of a common schema. For example, if the candidate results have different attribute configurations such as "resource, location, time, cost, risk," the response strategy providing device (100) may correct missing fields or convert units according to the common schema to convert them into a comparable form. As an example, during the standardization process, qualitative descriptions included in the candidate results may be mapped to a pre-set set of labels and used for subsequent constraint evaluation.

[0186] In step S1030, the response strategy providing device (100) configures a set of constraints based on domain knowledge structures and related information. The constraints may include allowable conditions, prohibitive conditions, threshold conditions, or precedence conditions. For example, the availability of a specific resource, the feasibility of execution within a specific time window, the maintenance of a specific risk threshold value or lower, or compliance with specific regulations may be configured as constraints. As an example, the constraints may be configured by referencing constraints defined in an ontology and relationships stored in a knowledge graph.

[0187] In step S1040, the response strategy providing device (100) generates a list of constraint priorities by determining the priority for each constraint. The constraint priorities may be determined according to the nature of the constraints. For example, constraints related to safety or regulatory compliance may be placed at the highest priority, and constraints related to cost reduction may be placed at a lower priority. As an example, the constraint priorities may be predefined by a domain-specific policy table and may be dynamically adjusted based on whether there is a mandatory expression or a goal-oriented expression of the user query.

[0188] In step S1050, the response strategy providing device (100) evaluates whether there is a constraint violation for each candidate result and detects conflicts between candidate results. The constraint violation evaluation can be implemented as a process that determines whether each attribute of a candidate result satisfies a constraint condition. Conflict detection may include cases where candidate results presented by different engines simultaneously request the same resource or request mutually exclusive actions within the same time interval. For example, conflict detection can be performed by verifying resource dependency or exclusive relationships using the relationships in a knowledge graph.

[0189] In step S1060, the response strategy providing device (100) selects synthetic target candidates by resolving conflicts according to constraint priority. Constraint priority-based conflict resolution can be performed by first excluding candidates that violate high-priority constraints, and for conflicts with low-priority constraints, by searching for alternative candidates or mitigable adjustment values. For example, even if two candidates conflict in terms of cost, a candidate that satisfies safety constraints may be selected first. For example, a violation of a high-priority constraint may be treated as an immediate exclusion rule, and a violation of a low-priority constraint may be treated as a penalty.

[0190] In step S1070, the response strategy providing device (100) calculates a final score by combining a constraint violation penalty with an objective function value. The objective function value may include at least one of cost, time, risk, or efficiency indicators, and the constraint violation penalty may be weighted according to the priority of each constraint. For example, the final score may be calculated as "the sum of the objective function value + penalty term," and the penalty term may be designed to be proportional to the degree to which the degree of constraint violation exceeds a reference value. Violations of high-priority constraints may be treated as virtually unselectable by assigning a very large penalty value, and violations of low-priority constraints may be assigned a small penalty value so as to be interchangeable with the objective function value.

[0191] In step S1080, the response strategy providing device (100) outputs a synthetic result based on the final score. The synthetic result may include a final response strategy, a list of final response actions, or a list of multiple candidates assigned priorities. For example, along with the synthetic result, the response strategy providing device (100) may include, as part of the basis information, which constraints played a decisive role in conflict resolution and which penalty terms contributed significantly to the final score.

[0192] According to an additional embodiment of FIG. 10, even if candidate results generated by multiple engines based on different criteria exist, the candidate results are standardized into a common representation, conflicts are resolved according to constraint priority, and a synthetic result is produced based on a final score by combining the objective function value and the constraint violation penalty, thereby providing the effect of improving safety, policy compliance, and result consistency compared to a single-engine-based result. In addition, unlike a method that simply excludes constraint violations, since a combination of penalties based on priority is applied, the balance between the domain objective and the constraint is quantitatively reflected, and it has the technical feature of being able to perform reproducible synthetic decisions even under complex conditions.

[0194] The device described above may be implemented as a hardware component, a software component, and / or a combination of a hardware component and a software component. For example, the device and components described in the embodiments may be implemented using one or more general-purpose or special-purpose computers, such as, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable gate array (FPGA), a programmable logic unit (PLU), a microprocessor, or any other device capable of executing and responding to instructions. The processing unit may execute an operating system (OS) and one or more software applications executed on said operating system. Additionally, the processing unit may access, store, manipulate, process, and generate data in response to the execution of the software. For ease of understanding, the processing unit may be described as being used as a single unit, but those skilled in the art will understand that the processing unit may include multiple processing elements and / or multiple types of processing elements. For example, the processing unit may include multiple processors or one processor and one controller. In addition, other processing configurations, such as parallel processors, are also possible.

[0195] Software may include computer programs, code, instructions, or a combination of one or more of these, and may configure a processing unit to operate as desired or command the processing unit independently or collectively. Software and / or data may be permanently or temporarily embodied in any type of machine, component, physical device, virtual equipment, computer storage medium or device, or transmitted signal wave so as to be interpreted by the processing unit or to provide instructions or data to the processing unit. Software may be distributed over networked computer systems and may be stored or executed in a distributed manner. Software and data may be stored on one or more computer-readable recording media.

[0196] The method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded on a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, etc., either alone or in combination. The program instructions recorded on the medium may be those specifically designed and configured for the embodiment, or they may be those known and available to those skilled in the art of computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes; optical recording media such as CD-ROMs and DVDs; magneto-optical media such as floptical disks; and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, and flash memory. Examples of program instructions include machine code, such as that generated by a compiler, as well as high-level language code that can be executed by a computer using an interpreter, etc. The hardware devices described above may be configured to operate as one or more software modules to perform the operation of the embodiment, and vice versa.

[0197] Although the embodiments have been described above with reference to limited examples and drawings, those skilled in the art can make various modifications and variations from the description above. For example, suitable results can be achieved even if the described techniques are performed in a different order than described, and / or the components of the described system, structure, device, circuit, etc. are combined or assembled in a form different from described, or replaced or substituted by other components or equivalents.

[0198] Therefore, other implementations, other embodiments, and equivalents to the claims also fall within the scope of the claims set forth below.

[0199] Although specific embodiments have been described in the detailed description of this document, it will be obvious to those skilled in the art that various modifications are possible within the scope of this document. Explanation of the symbols

[0200] 10 : User terminal 100: Response strategy providing device 110 : Communications Department 120 : Data processing unit 130 : Inference Strategy Processing Unit 140 : Response Strategy Generation Section 150 : Storage unit 200 : Data source

Claims

Claim 1 A method for providing a response strategy performed in a response strategy providing device, comprising: collecting domain data included in one or more data sources and preprocessing said domain data into an analyzable form; receiving a user query and obtaining at least a portion of information related to said user query from said domain data based on said user query; extracting query characteristic information from said user query and calculating a query characteristic score based on said query characteristic information; calculating a data characteristic score based on the composition of said related information; for each of a plurality of inference methods, calculating a raw fitness score by reflecting said query characteristic score and said data characteristic score; normalizing said raw fitness score to a preset score range; selecting one or more candidate inference methods in which the normalized fitness score is greater than or equal to a preset threshold; configuring an inference strategy by combining said one or more candidate inference methods; obtaining candidate results processed with said related information from each of one or more inference engines corresponding to each of said one or more candidate inference methods; standardizing said candidate results into a common representation; configuring a set of constraints based on a domain knowledge structure and said related information; and for each constraint A method for providing a response strategy based on intelligent inference strategy selection, comprising: a step of determining priority; a step of detecting conflicts between candidate results by evaluating whether constraints are violated for each of the candidate results; a step of selecting a candidate for synthesis by resolving the conflicts according to the priority for each constraint condition; and a step of generating a response strategy from the candidate for synthesis based on a final score calculated by combining a constraint violation penalty with an objective function value. Claim 2 delete Claim 3 In claim 1, the plurality of inference methods includes at least one of generative inference including rule-based inference, probability-based inference, numerical optimization-based analysis, and graph-based search, and the query characteristic information includes at least one of coercive expression, goal-oriented expression, uncertainty expression, and temporal expression, and a method for providing a response strategy based on intelligent inference strategy selection. Claim 4 A method for providing a response strategy based on an intelligent inference strategy selection, wherein, in claim 1, the step of obtaining at least some of the information related to the user query from the domain data comprises: a step of structuring the domain data to generate a domain knowledge structure; and a step of searching for and obtaining related information from the domain knowledge structure based on the user query. Claim 5 In claim 4, the domain knowledge structure comprises an ontology and a knowledge graph generated based on the ontology, and the step of generating the domain knowledge structure comprises: a step of providing an input including unstructured document data among the domain data to a large language model to extract at least one of a domain concept, a relationship between concepts, and a candidate constraint; a step of verifying the extraction result by checking whether the relationship between concepts is circular or whether the constraints conflict; and a step of performing re-extraction by changing the input range provided to the large language model if the verification result does not satisfy a pre-set verification criterion, and generating an ontology based on the extraction result if the verification result satisfies the verification criterion. A method for providing a response strategy based on an intelligent inference strategy selection. Claim 6 In claim 5, the knowledge graph is composed of triple-shaped graph data based on the selection of an intelligent inference strategy, based on entities identified from domain data and relationships between entities using an ontology as a schema. Claim 7 A method for providing a response strategy based on an intelligent reasoning strategy selection, wherein, in claim 1, the method further comprises: a step of generating basis information representing the basis on which the response strategy was generated; a step of generating a graph structure representing a basis relationship by linking the components of the response strategy and the basis information; and a step of outputting the response strategy and the basis information, wherein the output basis information includes a visualization result of the graph structure. Claim 8 A method for providing a response strategy based on an intelligent inference strategy selection, wherein, in claim 7, the above-mentioned basis information includes at least one of the relevant information used in the process of generating a response strategy, document basis information indicating the source of the relevant information, and at least one of the intermediate inference result produced in the process of generating a response strategy, wherein the nodes in the graph structure include the response strategy, relevant information, document basis information, or intermediate inference result, and the edges in the graph structure indicate a relationship or constraint satisfaction relationship that contributed to the derivation of the response strategy. Claim 9 A computer-readable storage medium storing a program for executing the method of any one of paragraphs 1, 3 through 8 on a computer. Claim 10 At least one processor; The method includes memory for storing instructions, wherein when the instructions are executed by the at least one processor, the processor performs the following operations: collecting domain data included in one or more data sources and preprocessing the domain data into an analyzable form; receiving a user query and obtaining at least some of the information related to the user query from the domain data based on the user query; extracting query characteristic information from the user query and calculating a query characteristic score based on the query characteristic information; calculating a data characteristic score based on the composition of the related information; for each of a plurality of inference methods, calculating a raw fitness score by reflecting the query characteristic score and the data characteristic score; normalizing the raw fitness score to a preset score range; selecting one or more candidate inference methods in which the normalized fitness score is greater than or equal to a preset threshold; configuring an inference strategy by combining the one or more candidate inference methods; obtaining candidate results processed from the related information from each of one or more inference engines corresponding to each of the one or more candidate inference methods; standardizing the candidate results into a common representation; and the domain knowledge structure and the A response strategy providing device based on intelligent inference strategy selection, which performs the following operations: constructing a set of constraints based on relevant information; determining a priority for each constraint; evaluating whether there is a constraint violation for each of the candidate results to detect a conflict between the candidate results; resolving the conflict according to the priority for each constraint to select a candidate for synthesis; and generating a response strategy from the candidate for synthesis based on a final score calculated by combining a constraint violation penalty with an objective function value.

Images