An effective deep learning method with multi-feature and attention mechanism for recognition of Chinese rice variety information

In the process of Chinese rice variety information named entity recognition, traditional methods cannot extract potential semantic information from data and cannot capture long-distance dependence. So, this paper proposes a Chinese rice variety information named entity recognition method based on a bidirectional long short-term memory network and conditional random field (BiLSTM-CRF), which combines radical features, word segmentation boundary features, and multi-head attention mechanism. First, the radical features and word segmentation boundary features are encoded and integrated into a pre-trained character vector as the model embedding to solve the disadvantage of the lack of semantic information. Then, the multi-head attention mechanism is introduced to assist the bidirectional long short-term memory network (BiLSTM) in acquiring long-distance context-dependence. Finally, a conditional random field (CRF) is used to realize character-level sequence annotation and then realize the named entity recognition task of Chinese rice variety information. The experimental results show that this model’s precision, recall, and F1-score are 95.78%, 97.07%, and 96.42%, respectively. The three evaluation indices are better than those of the other models. The model proposed in this paper can effectively identify Chinese rice variety information entities and provides method support for the subsequent construction of a Chinese rice variety information knowledge graph.