具有个人信息的对话生成模型研究

赋予聊天机器人个人信息对于提供自然的对话至关重要,对此提出具有个人信息的对话模型,包括问题分类、个人信息回复和开放域对话三个模块。在问题分类模块中,分析测试不同分类方法的效果;个人信息回复模块利用BiLSTM进行语义信息编码,训练采用对比损失函数,同时实验对比多种匹配模型;开放域对话模型以最大互信息为目标函数,减少无意义回复。实验表明该对话模型的自然性、逻辑性、多样性和一致性均优于传统模型。

DIALOGUE GENERATION MODEL WITH PERSONAL INFORMATION

Giving chat robot personal information is essential for providing a more natural dialogue.Therefore,a dialogue model with personal information is proposed,which includes three modules:question classification,personal information reply and open domain dialogue.In the question classification module,the effects of various classification models were analyzed and tested.The personal information reply module encoded the semantic information by BiLSTM,and the training used the contrastive loss function.At the same time,several matching models were compared.The open domain dialogue model used maximum mutual information as the objective function to reduce meaningless replies.The experiments show that the dialogue model in this paper is superior to the traditional model in naturalness,logicality,diversity and consistency.