摘要 |
The present disclosure relates, according to some embodiments, to a method of temporal bipartite projection for users and objects and a method of link prediction for an unhappened event. The method of the temporal bipartite projection comprises making a sequence of user-object weighted bipartite networks with the user-object weights, identifying the transitions for each user from a first object at a first time to a second object at a second time, assigning the transition weights corresponding to the transitions according to a predetermined rule, summing the transition weights for all users between two objects to obtain the transition tendencies, and constructing a temporal projection graph with the transition tendencies. The method of link prediction comprises constructing the temporal projection graph, identifying the potential transitions for the unhappened event at a target time, assigning the potential transition weights according to a second predetermined rule, summing the potential transition weights to obtain the scores, and ranking the scores of the link prediction. |
主权项 |
1. A method of temporal bipartite projection for users and objects applied, the method comprising:
receiving, at a computer, a user-object dataset, wherein the user-object dataset comprises a set of users, a set of objects, a set of times for which the events between the users and the objects occur; generating, at the computer, a sequence of user-object weighted bipartite networks {Gt, t=1, 2, . . . , T} with a set of user-object weights gn,mt between each pair of an nth user and a mth object at a time t by data processing the user-object dataset; identifying, at the computer, a set of transitions R(n, i, t1, j, t2) for the nth user from ith an object at a first time t1 to a jth object at a second time t2 according to the user-object weighted bipartite networks Gt, wherein t2>t1; assigning, at the computer, a set of transition weights w(n, i, t1, j, t2) corresponding to the transitions R(n, i, t1, j, t2) according to a predetermined rule; summing, at the computer, the transition weights w(n, i, t1, j, t2) for all users to obtain a set of transition tendencies ĝi,j from the ith object and the jth object; and constructing, at the computer, a temporal projection graph Ĝ for all objects with the transition tendencies ĝi,j. |