Research and analysis of relational extraction based on bidirectional long and short-term memory networks

This paper explains the architectural design of BiLSTM, which includes word embedding processing in the input layer, bidirectional sequence modeling function in the BiLSTM layer, and attention mechanism layer that can dynamically weigh important information. In the experiment part, use the SemEval 2010 Task 8 dataset for verification. This dataset specifies nine bi-directional relationship types and has 8000 samples for training and 2717 samples for testing. Experiments prove that the model based on the attention mechanism and BiLSTM does well in the accuracy, precision, recall, and F1 value, with accuracy at 0.95 and precision, recall and F1 being 0.94, 0.92, and 0.91, far surpassing traditional CNNs(model accuracy 0.84), LSTM (model accuracy 0.89), and GRN (model accuracy 0.87). In addition, the AUC value of BiLSTM-attention mechanism model is 0.92, which also proves its ability to deal with the problem of sequence data. By combining the bi-directional long short-term memory and attention, it can effectively enhance the accuracy and robustness of the relation extraction, and it can provide good technical ideas and way of thinking for the relation extraction work on natural language.