基于深度学习语义匹配的通用智能问答系统的设计与实现
Design and Implementation of Universal Intelligent FAQ System Based on Deep Learning Semantic Matching
武凌 1黄淑芹 1陈劲松 1周健1
作者信息
- 1. 安徽财经大学 管理科学与工程学院,安徽 蚌埠 233030
- 折叠
摘要
FAQ问答系统有着广泛的实际应用场景,但传统问答系统的搭建通常需要将相关领域的知识转化为一系列的规则和知识图谱,且构建过程重度依赖人力,换个场景或用户都需要大量的重复劳动.针对上述问题并结合Bert模型和向量搜索引擎Faiss,本文设计了一种基于深度学习语义匹配的FAQ问答系统的解决方案,介绍了系统的工作原理和设计流程,可以方便快速地搭建面向特定领域的问答系统,有效地减少了文本预处理的过程,实现了秒级的查询响应.系统测试发现,可以很好地对用户的查询进行语义匹配,并返回正确的答案,为构建各种问答系统提供了一种通用的解决思路.
Abstract
Frequently Asked Questions (FAQ) system has a wide range of application scenarios. However, the construc-tion of systems usually needs to transform the knowledge of related fields into a series of rules and knowledge graph, which is labor-intensive and requires significant repetitive work when changing contexts or users. Combining Bert model and vector search engine Faiss, we design a solution of FAQ system based on deep learning semantic matching. The paper introduces the working principle and design process, which can quickly build a field-specific FAQ system, reduce the text preprocessing pro-cess effectively, and realize second query response. Tested extensively, our system demonstrates an adeptness at semantically matching user queries and providing accurate answers, offering a versatile approach to constructing diverse FAQ systems.
关键词
深度学习/FAQ系统/向量搜索/语义匹配/Bert模型/FaissKey words
deep learning/FAQ system/vector search/semantic matching/Bert model/Faiss引用本文复制引用
基金项目
安徽省高校自然科学研究重点项目(KJ2021A0484)
安徽财经大学校级教研项目(acjyyb2022021)
出版年
2024
国家科技期刊平台
NSTL
万方数据