| 摘 要: 聊天机器人依据其对话生成技术主要分为检索模型和生成模型，但这两种模型各有优劣。通过设计一种混合检索与生成的混合对话模型，保留了两种模型的优点并尽量避免其缺点。混合模型既具有检索模型的信息性，又具有生成模型的多样性，人机对话更加流畅、生动。使用Elasticsearch搜索服务器大大降低了系统回复生成的时间，回复速度提高了10%。此外，考虑到聊天机器人在对话过程中难以保持一致的个性，采用PostKS(Posterior Knowledge Selection)模型嵌入个性属性，实现了个性化的智能聊天机器人系统。文中提出的混合模型实现了个性化的回复且具有较快的回复速度，更加符合用户的需求。 |
| 关键词: 聊天机器人；Elasticsearch；PostKS模型；个性化 |
|
中图分类号: TP311.1
文献标识码: A
|
| 基金项目: 国家自然科学基金面上项目(61872074). |
|
| Design and Implementation of Personalized Chatbot System based on Retrieval and Generation Hybrid Model |
|
KONG Fanheng, GAO Yongqi, ZHANG Zishuai, RUAN Junhao, FENG Shi
|
(School of Computer Science and Engineering, Northeastern University, Shenyang 110819, China)
kongfanheng426@163.com; 905323742@qq.com; 1060876095@qq.com; 954369920@qq.com; fengshi@stu.neu.edu.cn
|
| Abstract: According to their dialogue generation technology, chatbots can be mainly divided into retrieval model and generation model, which have both strengths and weaknesses. This paper proposes to design a hybrid dialogue model of retrieval and generation, to retain the strengths of the two models while avoiding as many weaknesses as possible. The hybrid model not only has the information of retrieval model but also has the diversity of generating model, so the man-machine dialogue is more smooth and vivid. Using the Elasticsearch search server greatly reduces the system reply generation time and the response speed is increased by 10%. In addition, considering that it is difficult for chatbots to maintain consistent personalities in the process of dialogue, PostKS (Posterior Knowledge Selection) model is used to embed personality attributes to realize the personalized intelligent chatbot system. The proposed hybrid modelr realizes personalized reply and has a faster reply speed, which is more in line with the needs of users. |
| Keywords: chatbot; Elasticsearch; PostKS model; personalization |
