Research on Personalized Recommendation Algorithm Based on Dynamic Network

At present, personalized recommendation algorithms are primarily based on static networks, and therefore cannot accurately model or effectively provide recommendations in real dynamic networks. To extract the dynamic information of changes in dynamic networks with time and the characteristics of multi-layer networks, a personalized recommendation algorithm based on self-encoding, called “DAER,” is proposed in this work. DAER can dynamically update the input vector of nodes with the change of time, includes the addition of a time coefficient into the loss function, and uses different time dimensions to adjust the loss function for optimal training. Furthermore, the number of hidden layers is expanded according to the changes of nodes in a dynamic network. Finally, the node output vector is used for link prediction, and the recommendation process is realized by sorting the link scores of nodes in reverse order. The results of experiments demonstrate that the proposed DAER algorithm is more effective for recommendation, and has significant application value in the field of personalized recommendation in dynamic networks.