基于LLM与规则协同的配置命令知识图谱构建方法

An LLM-Rule Collaborative Method for Constructing Configuration Command Knowledge Graphs

摘要: 随着无线通信网络规模不断扩大和设备功能持续演进，配置命令、参数约束及业务场景之间的关系日益复杂，传统面向人工阅读和关键词检索的配置命令手册难以满足配置知识关联查询与智能化应用需求。为解决这一问题，本文提出一种基于LLM与规则协同的配置命令知识图谱构建方法。该方法构建了语义解析、模式约束与图谱融合相结合的知识建模机制，利用规则与LLM识别配置命令手册中的实体与关系，将非结构化配置知识转换为一致的图谱表示。在基础图谱之上，引入面向业务场景的增量演化机制，通过命令实体对齐、配置序列恢复和有序关系建模，将静态命令知识扩展为能够描述场景执行流程的配置知识网络。最终构建的知识图谱包含5类节点和7类关系，共计15,834个节点和53,964条关系，实现了配置命令的统一表示，为配置命令查询、场景流程检索等提供了结构化知识支撑。

Abstract: The interactions between configuration commands, parameter constraints and service scenarios grow more intricate as networks expand and device functionality advances. However, configuration manuals are written for human reading and keyword lookup, and struggle to support associative queries and intelligent use of configuration information. To solve this, we build a configuration command knowledge graph using LLMs and rule-based checks. Our pipeline combines semantic parsing, schema enforcement, and graph construction. We also design an incremental update mechanism for changes in deployment environments. Through entity alignment, sequence reconstruction, and ordered relationship modeling, static command knowledge is expanded into a network describing execution workflows. The resulting graph comprises 5 node types and 7 relationship types, totaling 15,834 nodes and 53,964 relationships. It provides a standardised view of commands and supports command lookup and workflow tracing.