1. Recognition Performance Validation of Weather Map Images by ChatGPT.
- Author
-
Takasuka, Takumi, Takano, Yuki H., Watanabe, Shotaro, and Kumoi, Gendo
- Subjects
METEOROLOGICAL charts ,RECOGNITION (Psychology) ,IMAGE recognition (Computer vision) ,PATTERNMAKING ,CHATGPT - Abstract
Weather map recognition is a complex image recognition task that requires two cognitive processes: the interpretation of symbols and future predictions. Recent advancements in multimodal AI have shown the potential to solve such complex tasks. This research uses ChatGPT, a type of multimodal AI, to validate the recognition performance of weather map images and proposes prompt engineering for more accurate recognition. Weather map recognition is a problem that is also featured in the Common University Entrance Test. If weather maps can be recognized, automatic grading and question generation become possible, leading to learning support. Furthermore, it is expected that weather commentary text can be automatically generated from weather maps. Therefore, weather map image recognition is an important task, but it requires not only recognizing time-series changes in local weather data but also recognizing information on large-scale weather patterns such as pressure pattern and making future predictions. In this research, we use ChatGPT4-Vision, a multimodal AI, to validate the recognition performance of weather map images. We investigate whether it can answer questions about weather maps from university entrance exams and generate weather commentary text from weather map images. By conducting multiple experiments with varying tasks and information in the prompts, and evaluating the accuracy of the generated commentaries, we propose and validate improvements in reading performance through prompt engineering. [ABSTRACT FROM AUTHOR]
- Published
- 2024
- Full Text
- View/download PDF