Search

JP-2026075489-A - Information processing device, information processing method, and program

JP2026075489AJP 2026075489 AJP2026075489 AJP 2026075489AJP-2026075489-A

Abstract

[Problem] To provide a technology that can generate easily understandable explanations for the output results of a model. [Solution] An information processing device according to one aspect of the present disclosure includes: a first creation unit that creates model information including the output result of a model that predicts a target variable from explanatory variables; a second creation unit that creates a prompt instructing the generation of an explanation for the output result based on the model information and domain knowledge of the task performed by the model; and a generation unit that generates information representing an explanation for the output result based on the prompt and generative AI realized by a machine learning model including a large-scale language model. [Selection Diagram] Figure 2

Inventors

  • 金田 龍哉
  • 松井 哲郎
  • 島崎 祐一

Assignees

  • 富士電機株式会社

Dates

Publication Date
20260508
Application Date
20241022

Claims (11)

  1. A first creation unit creates model information including the output results of a model that predicts the dependent variable from the explanatory variables, A second generation unit creates a prompt that instructs the generation of an explanation for the output result based on the model information and the domain knowledge of the task the model performs, A generation unit generates information representing an explanation for the output result based on the prompt and a generative AI implemented by a machine learning model including a large-scale language model. An information processing device having
  2. The second creation unit described above is: The information processing device according to claim 1, which generates the prompt by setting each piece of information included in the model information and the domain knowledge to a pre-created template.
  3. The information processing apparatus according to claim 2, wherein the template includes a natural language sentence instructing the generation of a predetermined explanation for the output result, and a variable portion in which each piece of information included in the model information and the domain knowledge are respectively set.
  4. The information processing apparatus according to claim 3, wherein the template includes at least one of a mathematical formula used for numerical calculations performed by the generating AI and program code for a predetermined process performed by the generating AI.
  5. The information processing apparatus according to any one of claims 2 to 4, wherein the model information includes the output result, the names of the explanatory variables, the names of the target variables, and the names of the model type or the type of algorithm that implements the model.
  6. It has an acquisition unit that acquires user input information representing instructions for summarizing the explanation of the output result, or instructions for additional explanations, questions, or predetermined instructions for the explanation of the output result, The generating unit is The information processing apparatus according to claim 1, which generates information representing a summary of the explanation for the output result, or information representing the additional explanation, information representing the answer to the question, or information that satisfies the predetermined instructions, based on the user input information and the generating AI.
  7. The second creation unit described above is: The information processing device according to claim 1, which generates a prompt instructing an explanation for the result of comparing the output result with the actual value represented by the actual data, based on the model information, the domain knowledge, and the actual data obtained from the target of the task.
  8. The information processing apparatus according to claim 1, wherein the domain knowledge includes a natural language sentence describing the task and a natural language sentence describing the input and output data of the model.
  9. The information processing apparatus according to claim 8, wherein the domain knowledge includes natural language sentences describing the characteristics of the input/output data.
  10. A first creation procedure for creating model information that includes the output results of a model that predicts the dependent variable from the explanatory variables, A second creation procedure for creating a prompt that instructs the generation of an explanation for the output result based on the model information and the domain knowledge of the task the model performs, A generation procedure that generates information representing an explanation for the output result based on the prompt and a generative AI implemented by a machine learning model including a large-scale language model, A method of information processing performed by a computer.
  11. A first creation procedure for creating model information that includes the output results of a model that predicts the dependent variable from the explanatory variables, A second creation procedure for creating a prompt that instructs the generation of an explanation for the output result based on the model information and the domain knowledge of the task the model performs, A generation procedure that generates information representing an explanation for the output result based on the prompt and a generative AI implemented by a machine learning model including a large-scale language model, A program that causes a computer to execute something.

Description

This disclosure relates to an information processing device, an information processing method, and a program. Artificial intelligence (AI), machine learning, statistical methods, etc., are being used to perform specific tasks. For example, Patent Document 1 discloses a technology that uses a model trained through machine learning to detect anomalies in a monitored system. On the other hand, in recent years, there have been cases where explanations for the output results of models are required. For example, users may want to know how the output values of a model that forecasts electricity demand are calculated, or what the reasons are for increases or decreases in the forecast values compared to recent electricity demand figures. Japanese Patent Publication No. 2024-41512 This figure shows an example of the hardware configuration of the model output result explanation device according to this embodiment.This figure shows an example of the functional configuration of the model output result explanation device according to this embodiment.This flowchart shows an example of the model output result explanation process according to this embodiment.This figure shows an example of domain knowledge (part 1).This figure shows an example of domain knowledge (part 2).This figure shows an example of a prompt template.This figure shows an example of a prompt (part 1).This figure shows an example of a prompt (part 2).This figure shows an example of a prompt (part 3).This figure shows an example of a prompt (part 4).This figure shows an example of a prompt (number 5).This figure shows an example (part 1) of explanatory information for the model output results.This figure shows an example (part 2) of explanatory information for the model output results.This figure shows an example (part 3) of explanatory information regarding the model output results.This figure shows an example (part 4) of explanatory information regarding the model output results. The following describes in detail one embodiment of the present invention with reference to the drawings. The following embodiment describes a model output result explanation device 10 capable of generating an explanation that is easily understandable to the user for the output results of a model that performs some task (e.g., classification tasks such as anomaly detection, regression tasks such as prediction, etc.). Note that the task is not limited to a specific task; the following embodiment can be applied to any predetermined task performed by a model. Here, a model refers to a program or other entity that has been learned, trained, or created to perform a certain task using artificial intelligence, machine learning, statistical methods, etc. Generally, a model can be expressed as y = f(x) using some function f. x and y are generally expressed in the form x = ( x₁ , ..., xₙ ), y = ( y₁ , ..., y ₩M ), etc., where each xₙ is called an explanatory variable and each y₩ m is called a dependent variable. N is the number of explanatory variables, and M is the number of dependent variables. Note that data collected from the object of the task (e.g., plant, equipment, machinery, etc.) may contain various variables, but the variables used as explanatory and dependent variables are defined or determined, for example, by the user or model designer. Because a model takes multiple explanatory variables as input, even if the same output value is obtained, the reasons may differ. For example, in a model where the predicted value of electricity demand is the dependent variable, various variables such as temperature, time of day, and day of the week become explanatory variables. Therefore, even if the same predicted value is obtained, there are various patterns, such as electricity demand being high due to high temperatures and air conditioning demand, electricity demand being high due to low temperatures and heating demand, or electricity demand being high due to factories operating during the daytime on a weekday. Since the number of these patterns is enormous, it is difficult to prepare explanatory statements for all of them. Furthermore, for example, if the predicted value or prediction error suddenly increases or decreases compared to recent actual electricity demand, it may be necessary to investigate the reason. This is because some special reason (e.g., extreme weather, local events, etc.) can cause a significant change in the predicted value or an increase in the prediction error. In such cases, it is common to investigate the cause by comparing and examining the data with normal conditions. However, even if a highly explanatory model (e.g., multiple regression model, decision tree, etc.) is used, users cannot obtain a clear explanation without knowledge of both the task's domain and the model. In other words, in this case, the user is required to have a highly specialized knowledge of both the task's domain and the model. Therefore, even if a general explanation of the model's outpu