Event subject extraction method and device combining BIGRU and multi-head attention

• Main Authors: LIU FAQIANG, JI ZHE, NING SHAN, WAN XIN, LIANG RUIQI, SUN XUDONG, LI YAQIANG, WANG YONG
• Format: Patent
• Language: Chinese; English
• Published: 07.02.2023
• Subjects: CALCULATING; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; PHYSICS

The invention discloses a BIGRU and multi-head attention combined event subject extraction method and device. The method comprises the following steps: converting text data into a text input sequence X; utilizing a BiGRU network to obtain a vector XB corresponding to the text input sequence X; performing multi-head attention calculation based on the text input sequence X and the vector XB; and obtaining an event subject extraction result according to the attention calculation result. According to the method, context semantic features are learned by using the BIGRU network, and key feature information in the sequence is captured by introducing a multi-head attention mechanism, so that the working efficiency and accuracy are improved. 本发明公开了一种结合BIGRU和多头注意的事件主体提取方法及装置，包括：把文本数据转化为文本输入序列X；利用BiGRU网络获取文本输入序列X对应的向量XB；基于文本输入序列X与向量XB进行多头注意力计算；依据注意力计算结果，获取事件主体提取结果。本发明使用BIGRU网络学习上下文语义特征，通过引入多头注意力机制捕获序列中的关键特征信息，提高了工作效率及准确性。