COMPLEX EVENT EXTRACTION ALGORITHM BASED ON DEEP EMBEDDING DECOUPLING

Complex events are descriptions of events with multiple subjects, objects and behaviors. Compound sentence is the most typical representative, complex events have problems such as sentence nesting, element dispersion and sharing. As a result, the conventional event element extraction algorithm cannot completely extract the corresponding nested event elements. To deal with the above problem, this paper proposes a multi-domain oriented complex event extraction algorithm based on deep embedded decoupling (CEEN-DEP). First, build Roberta-BigruAttention network to fuse character features, time series features, interaction features and embed semantic features of long text complex events. Then being aimed at nested attributes for complex events, define a common tag system for multidomain complex event elements and divide complex event elements into common elements and domain elements. At the same time, considering the dependence between coupling tags, the conditional random field is introduced, and feedback training is conducted to complete the joint extraction of coupling elements. The comparative experiments show that the algorithm can decouple the nested elements of complex events and complete the reorganization of corresponding event elements. The accuracy rate, recall rate and f1 value are relatively stable under different sample sizes, and they are maintained in the range of 80%-95%, which proves the effectiveness and robustness of the algorithm.