An emerging paradigm for scientific decision: the AI evaluation of space science projects

Abstract

In the past six decades, the progress of spaceflight projects has won the admiration of the whole world. However, how to evaluate the values of research projects remains an esoteric and cost effective question. To improve the selections in space science projects, we utilized AI tools to provide an overall framework for broader audience. Our work conducted a three-phased study. We explored space life science research as it is one of the most intensively researched areas in space science. We learned the domain science data and constructed a space science knowledge graph. Subsequently, to better extract semantic features, we introduced SpaceBERT, a pre-trained language model fine-tuned with contrastive learning. We then developed SpaceGL, a deep learning framework tailored for predicting frontier research. Lastly, we prioritized candidate space experimental projects based on AI model and compared with the real results from the science panel judges and the “Lottery model.”