摘要
为解决对话关系抽取任务中实体间关联语义信息稀疏、获取核心语义和触发线索困难等问题,提出一种新型的对话关系抽取模型。在对话文本中融入抽象语义表示来增强对话的核心语义,以解决在对话关系提取过程中出现的语义缺失和逻辑纠缠问题;引入全局对话交互机制,通过对关键线索的捕捉来改善对话中有效信息稀疏的问题;通过增加明确的结构信息来进一步丰富实体间的关系特征,使模型能够更好地理解对话文本。实验结果表明:相较于基线模型BERTs,文中提出的模型在数据集DialogRE上的F 1和F 1C分别提升了5.5%和6.2%;相比于序列模型CNN、LSTM和BiLSTM,在对话关系抽取中准确率提高9%以上,效果显著。文中模型在复杂对话场景中的泛化能力更好,鲁棒性更强。
In order to solve the problems of sparse inter entity association semantic information and difficulty in acquiring core semantics and trigger cues,the paper presents a novel dialogue relation extraction model.Firstly,the abstract semantic representation is incorporated into the dialogue text to enhance the core semantics of the dialogue,so as to solve the semantic loss and logical entanglement in the process of dialogue relation extraction.Secondly,the global dialogue interaction mechanism is introduced to capture key clues in order to solve the problem of the sparse effective information in dialogue.Finally,the explicit structural information is added to further enrich the relationship between entities,which enables the model to better understand the dialogue text.The experimental results are as follows.Compared with the baseline model BERTs,the F 1 and F 1C of the proposed model in data set DialogRE are improved by 5.5%and 6.2%respectively.Compared with the sequence models CNN,LSTM and BiLSTM,the accuracy rate of the dialogue relation extraction is increased by more than 9%,having an obvious effect.Therefore,the model proposed in this paper has better generalization ability and robustness in complex dialogue scenes.
作者
李博博
荆心
仲尧
LI Bobo;JING Xin;ZHONG Yao(School of Computer Science and Engineering,Xi’an Technological University,Xi’an 710021,China)
出处
《西安工业大学学报》
CAS
2023年第5期503-511,共9页
Journal of Xi’an Technological University
基金
陕西省科技厅重点研发计划(2022GY 048)。
关键词
对话关系抽取
抽象语义表示
全局对话交互
关系特征
dialogue relation extraction
abstract meaning representation
global dialogue interaction
relationship feature