A Joint Extraction Method of Entity Relations in Aquaculture Long Text Using N-Gram Fusion

To solve the problem of misjudgment and loss of valid information caused by a large amount of irrelevant information in aquaculture long text, a joint extraction method of entity relations based on N-Gram fusion was proposed. Firstly, the multi-model fusion algorithm is used to extract the text matrix feature map based on BERT initialization, and then the cascading BiLSTM is used to extract the deep features. After that, the features of the long text slice matrix preprocessed by fusion N-Gram algorithm are extracted layer by layer, and the relative and absolute positions of slice matrix are modeled. The experimental results on the self-constructed aquaculture long text data set and SKE public data set show significant improvements compared with the benchmark model. The experimental results show that this method can fully acquire and process the semantic information in aquaculture long text, and effectively improve the accuracy and integrity of entity relation extraction.