Towards Operational Data Analytics Chatbots -- Virtual Knowledge Graph is All You Need

2025-06-27

AI Summary

To address the challenges of querying NoSQL telemetry data—namely, query complexity, implicit relational semantics, and poor real-time performance—in Datacenter Operations Data Analytics (ODA), this paper proposes an end-to-end chatbot system that synergistically integrates a lightweight Virtual Knowledge Graph (VKG) with Large Language Models (LLMs). The system bypasses costly physical knowledge graph construction by dynamically mapping NoSQL schemas into queryable VKGs and leverages LLMs to accurately translate natural language questions into SPARQL queries. Its key innovation lies in a VKG-LLM co-reasoning mechanism, which drastically reduces graph maintenance overhead and query latency. Experiments demonstrate a substantial improvement in query accuracy—from 25% to 92.5%—and an 85% reduction in average latency (from 20.36 s to 3.03 s). The VKG consumes only 179 MiB of memory, enabling real-time deployment and interactive multi-variable time-series analysis.

Technology Category

• Data Mining & Knowledge Management: Intelligent Query Processing
• Natural Language Processing: Question Answering
• Search and Optimization: Distributed Search

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information

Abstract

With generative artificial intelligence challenging computational scientific computing, data centers are experiencing unprecedented growth in both scale and volume. As a result, computing efficiency has become more critical than ever. Operational Data Analytics (ODA) relies on the collection of data center telemetry to improve efficiency, but so far has been focusing on real-time telemetry data visualization and post-mortem analysis. However, with NoSQL databases now serving as the default storage backend to support scalability, querying this data is challenging due to its schema-less nature, which requires domain knowledge to traverse relationships between data sources. Ontologies and Knowledge Graphs (KGs) can capture these relationships, but traditional KGs are costly to scale and have not been widely applied to multivariate timeseries. Virtual Knowledge Graphs (VKGs) offer a lightweight alternative by generating query-specific graphs at runtime. In this work, we present a full end-to-end ODA chatbot system that uses a Large Language Model (LLM) to generate SPARQL queries, utilizing VKG for data retrieval. This approach achieves 92.5% accuracy compared to 25% with direct NoSQL queries. The proposed methodology optimizes VKG construction and LLM inference, cutting previous work average query latency by 85% (from 20.36s to 3.03s) and keeping VKG sizes under 179 MiB. This performance makes the tool suitable for deployment and real-time interaction with ODA end-users.

Problem

• Challenges in querying schema-less NoSQL data centers efficiently
• Need for scalable knowledge graphs for multivariate timeseries data
• Improving real-time interaction in Operational Data Analytics chatbots

Innovation

• Uses Virtual Knowledge Graphs for lightweight data retrieval
• Leverages LLM for SPARQL query generation
• Optimizes VKG construction and LLM inference