Aiming at the multi-granularity knowledge discovery of users in online community, a method of dynamic user interest portrait based on Three-way Fuzzy Concept Lattice (3WFCL) is proposed to reveal the relationship between long-term and short-term interest drift, which improves the accuracy of the user portrait from a fuzzy perspective. Firstly, the label topic matrix of community users is constructed based on the topic analysis and the dynamic interest modelling is realized. Secondly, the environment migration index is calculated by extracting the topic and sentiment characteristics of the user environment. Meanwhile, interest inheritance is introduced into the calculation of interest characteristics, and interest attenuation vector is established to perceive the short-term interest label set dynamically. Finally, the membership degree of the three-way fuzzy interests is put forward into the multi-feature concept modelling of the long-term and short-term interest labels. Afterwards, the multi-granularity fuzzy association rules are generated to complete the multi-granularity characterization of differentiated interests by constructing the three fuzzy formal contexts. The effectiveness of the proposed method is validated on real data sets. The results show that it is of great significance to utilize the three-way fuzzy concept analysis into user portrait modelling, which is helpful to mine semantic association information and improve the analysis ability of fine-grained interest hierarchy.