Consumer Price Index Dynamics Unveiled: Approximate Inverse Model Explanations (AIME) of News Text Data and Consumer Price Index Data

This study describes a method for extracting explanations for changes in the Consumer Price Index (CPI) using approximate inverse model explanations (AIME) for news text data. AIME is an Explainable AI (XAI) technique that can explain model behavior and estimation results by deriving approximate inverse operators of AI and machine learning models. In this study, we assumed that there is an AI or machine learning model (black box model) that predicts CPI from news text data, and we found that it is possible to derive an explanation for CPI fluctuations by deriving its approximate inverse operator. In this method, news data are decomposed into words, weighted and quantified using term frequency-inverse document frequency (TF-IDF) as the explanatory variable, and CPI as the objective variable, and an approximate inverse operator is constructed using AIME, This method makes it possible to extract words that contribute to the rise and fall of the CPI. By realizing this method, it is possible to use it for investment decisions and individual consumption behavior that consider CPI fluctuations, and it can also be used as a decision-making tool for the Japanese government's economic policy decisions.