End-to-end building field task-based dialogue system based on hierarchical memory network

The invention provides an end-to-end building field task type dialogue system based on a hierarchical memory network. The end-to-end building field task type dialogue system comprises a context encoder used for encoding a dialogue history into a vector which represents a coding vector of an nth word in a dialogue; the hierarchical memory network is used for storing dialogue history and knowledge base information; the memory network pointer is used as a related line Skb in a conditional query knowledge base, a related line Sd in a dialogue memory network and content okb in a related knowledge base; the replication enhancement decoder is used for obtaining specific answer information; according to the method, dialogue history and knowledge base information can be divided into two memory networks, and knowledge is represented in a knowledge line form, so that the memory networks can smoothly perform knowledge reasoning. 本发明提出了一种基于层次型记忆网络的端到端建筑领域任务型对话系统，包括：上下文编码器：用于将对话历史编码成向量表示对话里第n个单词的编码向量；层次型记忆网络：用于存储对话历史和知识库信息