SOIRP: Subject-Object Interaction and Reasoning Path based joint relational triple extraction by table filling

Abstract

Joint relational triple extraction methods based on table filling have gained considerable attention in recent years due to remarkable effectiveness and capabilities of extracting relational triples from complicated sentences. However, most of the existing methods disregard the impact of the information interaction between the subject and object in the relational triple and the intrinsic association among relational triples on relational overlapping triples extraction, resulting in the inability to extract the relational triples hidden in the sentence. Furthermore, previous sequence annotation methods failed to consider the case of entity nesting. Therefore, there is still room for improving the relational overlap problem and entity nesting problem. To address this issue, we propose a new joint relational triple extraction model based on table filling, SOIRP, which contains a Subject-Object Interaction (SOI) module consisting of several different interactions and a Reasoning Path Extraction (RPE) module based on the idea of decoder module in Transformer. We refine the subject-object interaction information captured by the former and the path inference information captured by the latter through multiple iterations. In addition, we introduce a new table filling method and decoding strategy, which can capture all entities in each sentence and align the boundary tokens of entity pairs for each relation. We evaluate our proposed model on two public datasets and the experimental results demonstrate that our model outperforms the baseline model. The code of our model can be available at: https://github.com/valentiner/SOIPR.