语义增强的端到端对话系统
首发时间:2020-02-13
摘要:端到端任务型对话是人工智能领域研究热点,相对于传统流水线方法具有迁移性强、节省人力的优势。现今端到端对话系统通常基于编解码结构构造,结构相对简单,因此存在语义抽取能力差、回复不在词表的问题。本文基于Transformer与Bi-direction Gated Recurrent Unit构造一个层级语义编码器,用于深度抽取对话历史语义信息。并且在解码过程中引入复制机制,将对话历史和知识库信息加入生成源选择中,缓解不在词表问题。In Car Assistant公开数据集的实验验证了我们模型的有效性。
关键词: 计算机应用技术 端到端任务型对话 Transformer Bi-direction Gated Recurrent Unit 复制机制
For information in English, please click here
Semantically enhanced end-to-end task-oriented dialogue systems
Abstract:End-to-end task-oriented dialogue is a research hotspot in the field of artificial intelligence. Compared with the traditional pipeline method, it has the advantages of strong mobility and labor savings. However, nowadays, the structure of end-to-end dialogue is usually based on seq2seq structure, which is relatively simple. However, there are still some problems such as poor semantic extraction ability and out of vocabulary response. This paper constructs a hierarchical semantic extraction encoder based on Transformer and Bi-direction Gated Recurrent Unit to extract the semantic information of dialogue history. In addition, the copy mechanism is introduced in the decoding process, and the dialogue history and knowledge base are added to the selection of the generation source to alleviate the out of vocabulary problem. Experiments with In Car Assistant datasets confirm the validity of our model.
Keywords: Computer Application Technology End-to-end task-oriented dialogue Transformer Bi-direction Gated Recurrent Unit Copy mechanism.
引用
No.****
动态公开评议
共计0人参与
勘误表
语义增强的端到端对话系统
评论
全部评论0/1000