摘要：

Abstract:

【Purpose/significance】Dynamic user portrait model based on danmaku depicts the change law of user behavior characteris⁃

tics, which helps to improve the efficiency and effect of online learning from the perspective of users, content creators and video plat⁃

forms

.

【Method/process】Taking teaching videos of Bilibili as an example, the dynamic user portrait model is constructed based on the

theory of human behavior dynamics to explore the changes of users' behaviors of sending danmaku and watching video in different

time dimensions.

【

Result/conclusion】There are several specific peaks in the time distribution of the user sending danmaku in a day,

which gives the dynamic label of "day learning time preference". In terms of video time series, users are more inclined to send dan⁃

maku at the beginning and end of the title, giving the dynamic label "video progress sending danmaku preference". With the increase

of diversity sets, retained users have the habit of sending regular danmaku interactions.【Innovation/limitation】A dynamic user portrait

model based on danmaku is proposed. Follow-up research will explore the application of the model in other scenarios.