基于自注意力机制的多特征融合槽抽取模型  

Multidimensional features slot filling model based on self-attention mechanism

作　　者：牛迪 Niu Di(School of Economics,Zhejiang University,Hangzhou 310027,China)

机构地区：[1]浙江大学经济学院,浙江杭州310027

出　　处：《南京理工大学学报》2022年第1期69-75,共7页Journal of Nanjing University of Science and Technology

基　　金：国家自然科学基金(71673249)。

摘　　要：在任务型人机对话系统中,槽抽取任务是至关重要的一个环节。为了提高槽抽取模型的识别准确率,该文提出了一种利用自注意力机制融合文本的多特征维度特征的方法。该方法在常规的双向长短期记忆模型(Bi-directional long short-term memory,Bi-LSTM)和条件随机场(Conditional random field,CRF)的槽抽取模型基础框架之上,引入了自注意力机制融合多维度的特征,如意图特征、词特征、词典特征、字符特征等。通过融合这些特征,槽抽取模型的槽提取能力得到了提升。实验结果证明,该方法的槽抽取F1值能达到97.99%,比常规模型最大提升了3%。Slot filling is one of the most important tasks in task-oriented dialogue system.In order to improve the performance of slot filling model,this paper proposes a method to merge multidimensional features using self-attention mechanism.Based on the basic framework of Bi-directional long short-term memory(Bi-LSTM)and conditional random field(CRF),self-attention mechanism is introduced to merge multidimensional features,such as intention feature,word feature,dictionary feature,character feature,etc.By merging these features,the performance of slot filling model can by improved.Experimental results show that the F1 value of slot filling model is 97.99%,which is 3%higher than the conventional methods.

关 键 词：槽抽取 自注意力 双向长短期记忆模型 多特征融合 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]