In this paper, we present our proposed method for discovering aesthetic notions of the four seasons in Sandaishu, the three imperial anthologies of waka poetry, using approximate inverse model explanations (AIME). Waka, a type of traditional Japanese poetry, reflects the minds of waka poets. By extracting and analyzing the scenes and emotions depicted in waka poetry, it is possible to reveal aesthetic notions of waka poets. We focus on waka poetry on the theme of the four seasons—spring, summer, fall, and winter—in Sandaishu, the three imperial anthologies of waka poetry—Kokin wakashu, Gosen wakashu, and Shui wakashu—. The approximate inverse model explanations (AIME) method enables the interpretation of the results of the tasks of black-box models by generating an inverse operator. In this study, the task of extracting terms that serve as features representing seasons from Sandaishu can be viewed as an inverse problem of a model that predicts seasons based on the terms used in waka poetry. Therefore, by using AIME, which constitutes its approximate inverse operator, it is possible to extract terms that symbolize the seasons from the waka poetry anthology. Extracting terms that symbolize the seasons from the waka poetry anthology reveals aesthetic notions specific to the four seasons. By comparing the results obtained from each Sandaishu, it is possible to discover commonalities and differences in the sense of seasons for each waka poetry anthology. The study of waka poetry from the viewpoint of data science contributes to the development of literary studies by providing objective verification and evidence for prevailing theories and new hypotheses, as well as new insights, such as undiscovered interpretations.