Lectures

Leveraging Business Q&A with LLMs over Product Knowledge Graphs

In the product information management (PIM) systems domain, we face the challenge of not only managing diverse data but also making it contextually relevant to business needs. Traditional PIM systems often fall short because they cannot provide meaningful contextual knowledge about product information within the same company. They struggle with data heterogeneity and fail to meet modern requirements for comprehensive, integrated information. 

We enhance PIM systems with knowledge graphs, significantly improving decision-making by integrating all data sources into a single source of truth. The knowledge graph uses different ontologies and taxonomies to structure and integrate data from various sources, including internal and external product information. This approach leverages semantic layers to enrich the data, making it contextually relevant. We seamlessly integrate external technical product documentation from competitors’ products through an ETL architecture, transforming unstructured PDF data into structured knowledge using OCR models. 

Business use cases, such as comparing the technical efficiency of internal products with that of external competitor products, demonstrate practical benefits. By integrating generative large language models (LLMs) with knowledge graphs, we enable accurate natural language question answering and specific product retrieval. A simple RAG (Retrieval-Augmented Generation) architecture usually fails in question-answering tasks over a knowledge graph. Therefore, we introduce a Graph RAG approach for PIM systems question answering. This involves steps like entity recognition and linking, subgraph extraction, graph database indexing, and answer retrieval and generation. 

We present PIM systems powered by knowledge graph and the Graph RAG approach, which highlights the system’s capability to answer business questions using natural language. The integration of LLMs with PIM systems knowledge graphs results in context-aware answers to complex business questions in natural language using Graph RAG or query languages using SPARQL. It empowers businesses with comprehensive insights, driving informed decision-making in the dynamic product landscape. 

Contact us

Maraike Heim
Senior Marketing Manager