Low complexity customized video playlist generation

We consider the problem of customizing video playlist where a system for constructing a playlist composed of a sequence of videos that attain a specific user profile is proposed. Each video item is represented as a collection of weighted attributes that represent the accompanying features. The proposed solution allows users to construct a sequence of video items by imposing global and transitional constraints. Transitional constraints deal with the similarity between successive video objects while global constraints ensure that the retrieved video attributes attain the proposed user profile. The retrieved sequence of videos is constructed by considering it as a solution to a constrained sequence retrieval problem where constraints represent user preferences. The constrained retrieval problem is then modeled as a network flow model which leads to integer program. Relaxation techniques are then used for complexity reduction. Experiments demonstrate the flexibility of the proposed model in constructing the playlist according to user needs.