The invention relates to a recommendation
algorithm based on adversarial learning and a bidirectional long-short-
term memory network, which comprises the following steps of: 1, predefining a symbol, including A1) defining a
heterogeneous information network, A2) defining a path in the
heterogeneous information network, A3) in the
heterogeneous information network G, defining a node connection sequence from a user u to an article i as a path, and defining that p = [v1, v2,..., vl], and p belongs to P; and 2, modeling as following: S1, modeling an embedded layer, and representing the embedded layer by using an initialized node vector; S2, constructing a
sequence modeling layer, using the vector representation obtained through initialization in the step S1 as input and applying the input to an existing bidirectional LSTM model based on an attention mechanism to optimize vector representation of the node, and learning a
coefficient matrix and an offset vector in the model; S3, setting a prediction layer, and finally calculating the probability; and S4, constructing an adversarial learning model. According to the method, the problem of node relation
noise in the
heterogeneous network isrelieved by learning the adversarial regularization item, adding the adversarial regularization item into a
loss function and optimizing the model, the robustness of node embedding is improved, and the recommendation accuracy is ensured.