Design And Implementation Of Joint Extraction Of Entities And Relations Based On Relation Splitting

With the development of the Internet,massive amounts of data are born.As one of the information forms that people contact most,text data contains a lot of valuable information.It is very important to obtain valuable key information from complex and redundant text data.As one of the basic tasks of information extraction,extraction of entity and relation is of great significance.In this context,this paper proposes a joint extraction model of entity and relation based on relation splitting.With the development of deep learning technology,joint extraction of entity and relation has made great progress.At present,the mainstream of joint extraction of entity and relation mostly adopts the idea of sequence annotation,but the existing annotation strategy has some shortcomings such as too many annotation times or overlapping labels.Aiming at this problem,this paper proposes a labeling strategy based on relation splitting,which can not only reduce labels overlap problem but also reduce complexity.Then,A joint extraction model of entity and relation based on relation splitting is proposed.n order to improve the information extraction ability of the model,multi-head self-attention mechanism is applied.In the decoder layer,tag embedding technology is applied to map different relations into continuous vector space,so that the model can annotate different relations sequentially at the same time.Tag smoothing and cost sensitive learning techniques are applied to the loss function of the model,so as to optimize the model convergence process and reduce the problem of label category imbalance caused by relation splitting.In order to verify the performance of the model,this paper selects two open datasets and designs four experiments to prove the accuracy of the model.Finally,this paper applies the proposed model to the field of power grid fault.A power grid fault domain dataset is build.Then verifies the effectiveness of the model in the field of power grid fault text through experiments,and completes the construction of power grid fault domain knowledge graph through steps such as knowledge extraction,processing,storage and visualization.