This paper proposes a semi-supervised automatic value recognition and prediction method to accomplish user group value modeling and unknown user value prediction modeling. Firstly, the LRCFP feature model is proposed, and SOM clustering algorithm is used to subdivide the social media users; thus the user group portrait is constructed The CH index is used to verify the validity of the LRCFP feature model and the SOM clustering model. Then, the decision tree model is used to predict the unknown user value. The authors take Bilibili social media platform as an example to illustrate the validity of the proposed modeling and method. User value is divided into five grades, and user characteristics of each grade are analyzed.