Attributed network embedding based on self-attention mechanism for recommendation method

Our model first extracts item attributes and encodes the attribute set of the item into the item preference space. The preference of different users are different points in the item preference space, and users become the anchor vector in the space. In the item preference space, each anchor vector of the user is used to measure the user’s preference for attributes, and the similarity relationship between the anchor vector and the item vector is constructed through metric learning. In order to make the set of anchor vectors of different users can be reused, the user’s preferences are…