摘要
针对现有意图识别联合模型在专业领域知识图谱问答中容易发生识别领域实体以及问句分类错误的情况,提出一个结合了领域知识图谱的意图识别联合模型。该模型有三步,将领域知识图谱中实体对应的本体标签以及本体间关系导入训练数据集,形成包含本体标签的知识文本以及额外包含本体关系的知识文本图;通过字符级嵌入和位置信息嵌入将包含了本体标签的知识文本转化成嵌入表示并依据知识文本图创建实体关系可视矩阵,明确知识文本各成分的相关程度;将嵌入表示和实体关系可视矩阵输入模型编码层进行模型的训练。以高速列车领域知识图谱为例,经过准确率和召回率的验证,以该方法训练出的模型在高速列车领域问答数据集的意图识别任务上取得了更好的表现。
Aiming at the situation that the existing joint model of intention recognition is prone to identify domain entities and question classification errors in the question answering of professional domain knowledge atlas,a joint model of inten-tion recognition combined with domain knowledge atlas is proposed.The model has three steps.The ontology labels corre-sponding to entities in the domain knowledge map and the relationships between ontologies are imported into the training data set to form the knowledge text containing ontology labels and the knowledge text map containing additional ontology relationships.The knowledge text containing ontology tags is transformed into embedded representation through character level embedding and location information embedding,and the entity relationship visual matrix is created according to the knowledge text graph to clarify the correlation degree of each component of the knowledge text.The embedded represen-tation and entity relationship visual matrix are input into the model coding layer for model training.Taking the high-speed train domain knowledge map as an example,through the verification of accuracy and recall,the model trained by this method has achieved better performance in the intention recognition task of the high-speed train domain question and answer datasets.
作者
马自力
王淑营
张海柱
黎荣
MA Zili;WANG Shuying;ZHANG Haizhu;LI Rong(School of Computing and Artificial Intelligence,Southwest Jiaotong University,Chengdu 611756,China;School of Mechanical Engineering,Southwest Jiaotong University,Chengdu 610031,China)
出处
《计算机工程与应用》
CSCD
北大核心
2023年第6期171-178,共8页
Computer Engineering and Applications
基金
国家重点研发计划(2020YFB1708000)
四川省重大科技专项(2022ZDZX0003)。