Abstract: Aiming at the problems of lack of context semantic features, insufficient long‑distance dependence information and high entity complexity in the process of entity recognition of grain harvesting machinery maintenance knowledge, a named entity recognition model XLNet—BiLSTM—AFF—CRF for grain harvesting machinery maintenance knowledge based on attention mechanism feature fusion is proposed. Firstly, the generalized autoregressive XLNet pre‑trained model based on Transformer—XL is used as the embedding layer to extract word vectors. Secondly, the Bidirectional Long Short‑Term Memory (BiLSTM) is used to obtain contextual semantic features. Thirdly, the attentional feature fusion (AFF) is used to combine the output of the XLNet layer with the output of the BiLSTM layer to enrich the semantic information of the sequence. Finally, the conditional random field (CRF) model is input to learn the annotation constraint rules to generate the global optimal sequence. Experiments are carried out on the created maintenance corpus. The experimental results show that the accuracy rate, recall rate and F1 value of the proposed model in this paper are 98.4%, 97.6% and 97.9%, respectively, which are higher than those of the comparison model, verifying the effectiveness of the model in this essay.

Key words: grain harvesting machinery, maintenance, named entity recognition, attention mechanism, generalized autoregressive pretraining for language understanding (XLNet)

摘要： 针对谷物收割机械维修实体识别过程中存在上下文语义特征缺失、长距离依赖信息不充足、实体复杂度较高等问题，提出一种引入注意力机制特征融合的谷物收割机械维修知识命名实体识别模型XLNet—BiLSTM—AFF—CRF。该模型采用基于Transformer—XL的广义自回归XLNet预训练模型作为嵌入层提取字向量；然后使用双向长短时记忆网络（BiLSTM）获取上下文语义特征；利用注意力特征融合AFF将XLNet层输出与BiLSTM层输出进行组合，增强序列的语义信息；最后输入条件随机场CRF模型学习标注约束规则生成全局最优序列。在创建的维修语料库上展开试验，结果表明：所提模型的精确率、召回率和F1值分别为98.4%、97.6%和97.9%，均高于对比模型，验证所提模型的有效性。

关键词: 谷物收割机械, 维修, 命名实体识别, 注意力机制, 广义自回归预训练语言模型（XLNet）