KR-102964590-B1 - MULTI-MODULE BASED ARTIFICIAL INTELLIGENCE PREDICTION RESULT INTERPRETATION SYSTEM

KR102964590B1KR 102964590 B1KR102964590 B1KR 102964590B1KR-102964590-B1

Abstract

A multi-module-based artificial intelligence prediction result interpretation system is disclosed. The artificial intelligence prediction result interpretation system may include: a first MCP (Model Context Protocol) server that provides dynamic data used as input variables for a DNN (Deep Neural Network) model; a second MCP server that provides static data related to the input variables of the DNN model; a third MCP server that provides expert knowledge corresponding to a user query through RAG (Retrieval-Augmented Generation) in relation to the DNN model; and a LLM (Large Language Model) agent that generates an interpretation report explaining the causal relationship for the prediction result of the DNN model by utilizing the dynamic data and the static data together with the expert knowledge.

Inventors

조흔우
김세훈

Assignees

주식회사 이쓰리

Dates

Publication Date: 20260513
Application Date: 20251021

Claims (7)

In a computer-implemented artificial intelligence prediction result interpretation system, A first MCP (Model Context Protocol) server that provides dynamic data used as input variables for a DNN (Deep Neural Network) model; A second MCP server that provides static data related to the input variables of the above DNN model; A third MCP server that provides expert knowledge corresponding to a user query through RAG (Retrieval-Augmented Generation) in relation to the above DNN model; and A Large Language Model (LM) agent that generates an interpretation report explaining the causal relationship regarding the prediction results of the DNN model using the aforementioned dynamic data and static data together with the aforementioned expertise. Includes, The above LLM agent is, Defining an inference workflow of an LLM agent structure using LangGraph, and controlling the call order and conditions for the first MCP server, the second MCP server, and the third MCP server within the inference workflow. An artificial intelligence prediction result interpretation system characterized by
In paragraph 1, The above-mentioned first MCP server is, As data used by the above DNN model to derive prediction results, data reflecting the current state of a specific location is collected through an external public API (application programming interface), and The above-mentioned second MCP server is, Collecting fixed and private data of specific locations from an internal database or Geographic Information System (GIS) as data used by the above DNN model to derive prediction results. An artificial intelligence prediction result interpretation system characterized by
In paragraph 1, The above-mentioned third MCP server is, Using expertise related to the above DNN model, a corpus consisting of original papers and academic materials is vector-embedded and stored, and Searching for expert knowledge corresponding to the above user query based on vector embeddings An artificial intelligence prediction result interpretation system characterized by
In paragraph 1, The above LLM agent is, Acquiring the dynamic data, the static data, and the expertise by calling the data processing methods implemented in the first MCP server, the second MCP server, and the third MCP server through the MCP protocol. An artificial intelligence prediction result interpretation system characterized by
delete
In paragraph 4, The above LLM agent is, Calculating the complexity score of the above user query and enabling RAG search of the above third MCP server if the complexity score exceeds a predefined threshold. An artificial intelligence prediction result interpretation system characterized by
In paragraph 1, The above LLM agent is, Applying a system prompt including rules regarding the explanation of the prediction basis of the DNN model, causal inference, analysis reliability, and specification of limits to generate the above interpretation report An artificial intelligence prediction result interpretation system characterized by

Description

Multi-Module Based Artificial Intelligence Prediction Result Interpretation System The following description concerns the technology for interpreting artificial intelligence prediction results. The black box problem regarding the results of artificial intelligence (AI) analysis refers to a phenomenon in which it is difficult for humans to clearly understand or explain the process and basis for why or how a complex model, such as artificial intelligence, particularly deep learning, produced an output (result) in response to a specific input. Existing AI-based disaster prediction systems find it difficult to explain the causal relationship between risk indices and influencing factors due to the black box problem regarding AI analysis results. Therefore, even if risk warnings or indices are derived based on artificial intelligence, the intervention of experts is always essential for interpreting the causes. To address this, technology is required to provide reliable interpretation results regarding the causal relationship between independent and dependent variables in AI models. FIG. 1 is a block diagram illustrating an example of the internal configuration of a computer device in an embodiment of the present invention. FIG. 2 illustrates a configuration diagram of a multi-module-based artificial intelligence prediction result interpretation system in one embodiment of the present invention. FIG. 3 illustrates the workflow of a multi-module-based artificial intelligence prediction result interpretation system in one embodiment of the present invention. FIG. 4 illustrates an example of a real-time data collection process (S2) in an embodiment of the present invention. FIG. 5 illustrates an example of a static data collection process (S2) in an embodiment of the present invention. FIG. 6 illustrates an example of the RAG construction process in one embodiment of the present invention. FIG. 7 illustrates an example of a RAG data extraction algorithm in an embodiment of the present invention. FIG. 8 illustrates an example of an inference workflow in an embodiment of the present invention. FIG. 9 illustrates an example screen of an AI wildfire prevention assistant in one embodiment of the present invention. Hereinafter, embodiments of the present invention will be described in detail with reference to the attached drawings. Embodiments of the present invention relate to a technique for interpreting artificial intelligence prediction results. Embodiments including those specifically disclosed in this specification can solve the black box problem of existing DNN models having poor inference grounds and provide detailed explanations based on scientific evidence regarding AI prediction results, thereby dramatically improving the transparency and reliability of the system. An artificial intelligence prediction result interpretation system according to embodiments of the present invention may be implemented by at least one computer device, and an artificial intelligence prediction result interpretation method according to embodiments of the present invention may be performed through at least one computer device included in the artificial intelligence prediction result interpretation system. At this time, a computer program according to one embodiment of the present invention may be installed and run on the computer device, and the computer device may perform an artificial intelligence prediction result interpretation method according to embodiments of the present invention under the control of the run computer program. The above-described computer program may be stored on a computer-readable recording medium to be combined with the computer device to execute the artificial intelligence prediction result interpretation method on the computer. FIG. 1 is a block diagram illustrating an example of a computer device according to an embodiment of the present invention. For example, components of an artificial intelligence prediction result interpretation system according to embodiments of the present invention may be implemented by a computer device (100) illustrated in FIG. 1. As illustrated in FIG. 1, a computer device (100) may include a memory (110), a processor (120), a communication interface (130), and an input/output interface (140) as components for executing an artificial intelligence prediction result interpretation method according to embodiments of the present invention. Memory (110) is a computer-readable recording medium and may include ROM, PROM, EPROM, EEPROM, flash memory (e.g., NAND/NOR), SSD, HDD, magnetic tape, optical recording medium (CD-ROM, DVD, BD), magneto-optical medium, memory card, USB memory, etc. Here, non-perishable mass recording devices such as ROM and disk drives may be included in the computer device (100) as separate permanent storage devices distinct from memory (110). Additionally, an operating system and at least one program code may be stored in memory (110). These software components