Abstract: 【Purpose/significance】Media-based think tanks is the latest state of the integration of think tanks and media,and have natural advantages in guiding public opinion. Exploring the influencing factors and its relative importance of the public opinion guid⁃ance of media-based think tanks will provide theoretical support for the targeted construction of media-based think tank's public opin⁃ion guidance.【Method/process】Firstly, this paper puts forward the influencing factor system of public opinion guidance of mediabased think tanks; Then,the data of influencing factors,public opinion guidance and its component forces are collected through the questionnaire,and then the prediction model is established by using machine learning method (XGBoost algorithm). Finally,the relative importance of the influencing factors is obtained by calculating the gain.【Result/conclusion】It is found that the public opinion guid⁃ance and its component forces of media-based think tanks are affected by multiple factors of subject,object,media and field dimension.
Subject dimension factors have the greatest impact on the public opinion guidance of media-based think tanks,followed by the field di⁃mension. The degree of coordination and cooperation of public opinion guides is the key factor affecting the public opinion guidance of media-based think tanks.【Innovation/limitation】This paper innovatively puts forward the influencing factor system of public opinion guidance of media-based think tanks,and uses machine learning method to solve the relative importance of influencing factors by es⁃tablishing a prediction model. Expanding data and trying other algorithms are the direction of further improvement.