A new neighbourhood-based diffusion algorithm for personalized recommendation

Object ratings in recommendation algorithms are used to represent the extent to which a user likes an object. Most existing recommender systems use these ratings to recommend the top-K objects to a target user. To improve the accuracy and diversity of recommender systems, we proposed a neighbourhood-based diffusion recommendation algorithm (NBD) that distributes the resources among objects using the rating scores of the objects based on the likings of the target user neighbours. Specifically, the Adamic–Adar similarity index is used to calculate the similarity between the target user and other users to select the top K similar neighbours to begin the diffusion process. In this approach, greater significance is put on common neighbours with fewer neighbour nodes. This is to reduce the effect of popular objects. At the end of the diffusion process, a modified redistribution algorithm using the sigmoid function is explored to finally redistribute the resources to the objects. This is to ensure that the objects recommended are personalized to target users. The evaluation has been conducted through experiments using four real-world datasets (Friendfeed, Epinions, MovieLens-100 K, and Netflix). The experiment results show that the performance of our proposed NBD algorithm is better in terms of accuracy when compared with the state-of-the-art algorithms.