This paper presents a novel context-sensitive ranking algorithm, called ActiveRec, for providing flexible movie recommendations. Typically, ActiveRec can recommend movies to a user according to a specific movie type, or to a group of users satisfying their common interests. Firstly, ActiveRec constructs a multipartite graph where the nodes represent users, movies, and joint information, respectively. And then, a biased random-walk is performed to obtain the similarity between the request vector and every node on the graph. Based on the similarities, ActiveRec sorts out all movies that meet the user requirements and notify the user of a Top-N list. Additionally, a time-decay model following the Ebbinghaus forgetting curve is introduced to imitate the decay process of the importance of users' feedbacks when computing the edges' weights in the graph. Extensive experiments are performed on a real dataset to evaluate the performance. The results demonstrate that ActiveRec not only satisfies the requirements of flexible recommendations, but also achieves higher performance compared to the existing works.
展开▼
