Integrating knowledge graphs and large language models for querying in an industrial environment

Hočevar, Domen (author), Robnik Šikonja, Marko (mentor), Kenda, Klemen (comentor)

We describe an application that converts Industry 4.0 Administration Shell representations into a knowledge graph and stores the generated knowledge graph and vector embeddings of graph nodes in a GraphDB repository and a vector database, respectively. The application supports user queries in natu- ral language on the stored knowledge graph using a large language model and retrieval-augmented generation (RAG). Two different approaches for retrieval are used: subgraph retrieval and graph query generation. The application also has a front-end, through which users can perform all the aforementioned operations. The application was tested on mock data that represents a simple factory consisting of multiple different types of machines. Its performance was shown on an array of expected queries, showcasing its efficiency and specific strengths of the subgraph retrieval and graph query generation approaches.

• Language: English
• Keywords: knowledge graph, large language model, ChatGPT, retrieval-augmented generation, Industry 4.0
• Work type: Bachelor thesis/paper
• Typology: 2.11 - Undergraduate Thesis
• Organization: FRI - Faculty of Computer and Information Science ; FMF - Faculty of Mathematics and Physics
• Year: 2024
• PID: 20.500.12556/RUL-160705
• COBISS.SI-ID: 209028611
• Publication date in RUL: 03.09.2024

Secondary language

• Language: Slovenian
• Title: Integracija grafov znanja in velikih jezikovnih modelov za povpraševanje v industrijskem okolju

Abstract

Predstavljamo aplikacijo, ki pretvori predstavitve Ovojnice za upravl- janje s sredstvi Industrije 4.0 v graf znanja in shranjuje ustvarjeni graf znanja, skupaj z grafom pa tudi vektorske vložitve grafovih vozlišč, v po- datkovno bazo GraphDB in v vektorsko bazo podatkov. Aplikacija pod- pira uporabniške poizvedbe v naravnem jeziku na shranjenem grafu znanja, kar deluje na osnovi velikega jezikovnega modela in s poizvedovanjem obo- gatenega generiranja (RAG). Uporabljata se dva različna pristopa za prido- bivanje podatkov, pridobivanje podgrafov in generacija poizvedb na grafu. Aplikacija ima tudi uporabniški vmesnik, preko katerega lahko uporabniki izvajajo vse naštete operacije. Aplikacija je testirana na generiranih po- datkih, ki predstavljajo preprosto tovarno, sestavljeno iz več vrst strojev. Rezultati aplikacije so prikazani na množici pričakovanih poizvedb, s čimer smo evalvirali učinkovitost aplikacije in specifične prednosti pristopov prido- bivanja podgrafov in generiranja poizvedb na grafih.

• Keywords: graf znanja, veliki jezikovni model, ChatGPT, Industrija 4.0, s poizvedovanjem obogateno generiranje