Multimedia data mining (MDM) is one of the focused problems in current multimedia research domain. During MDM, based on user's annotations of corresponding media, it is crucial to represent those annotations correctly and completely to discover expected latent information. In other words, knowledge representation plays an important role in MDM. Existing representation models can express only static attributions of objects. But there are many important dynamic features in multimedia objects, especially in videos that should be represented correctly. In order to solve this problem, this paper proposes a new knowledge representation model - concept hypotaxis semantic network, whose significant contribution is to enable users to take the continuous and dynamic semantic features of media into account in MDM. Qualitative analyses and evaluations indicate that this model can be more flexible and general than others that only reflect static features of objects.