Keyphrase generation algorithm based on co-occurrence graph

Abstract

In the keyphrase prediction task, it is usually difficult for a single corpus information to help the model learn enough core information, so this paper improves the current keyphrase generation method based on the corpus. From the developments of the co-occurrence, the use of the word conformity is mentioned in the original selection of the relevant text, and the purpose of achieving the maximum utilization of the spending information is achieved. At the same time, this paper proposes the CG-Net model, which uses the GAT network to learn co-present diagrams and implements a replication mechanism, which ultimately surpasses the current optimal method on the NUS test set.