Abstract: Block grant and project funding are the two main modes for government research funding. Scientific analysis of the differences between the two funding modes is of great significance for the rational design of government research funding mechanisms. Multi-dimensional information contained in the output of these two funding modes from 2009-2018 in chemistry from Japan are taken as research objects Three-dimension analysis framework is built. In terms of methods, deep learning algorithms are introduced, and a training model for research object entity recognition in the field of chemistry is established, which realizes accurate recognition of research object entities and provides a factual basis for the analysis of funding differences. Data analysis results show that block grant and project funding have significant differences in multiple dimensions. The scientific research activities funded by block grant are more concerned in the research problem itself, while the scientific research activities funded by projects are more reflect the selection criteria from scientific community for projects. Policy recommendations of scientifically understand the difference between the two funding modes, rational layout, advancing strengths and avoiding weaknesses to establish efficient government research funding mechanism are proposed combing the data analysis results.