![[2602.22219] Comparative Analysis of Neural Retriever-Reranker Pipelines for Retrieval-Augmented Generation over Knowledge Graphs in E-commerce Applications](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.22219] Comparative Analysis of Neural Retriever-Reranker Pipelines for Retrieval-Augmented Generation over Knowledge Graphs in E-commerce Applications
Summary
This article presents a comparative analysis of neural retriever-reranker pipelines for retrieval-augmented generation (RAG) in e-commerce applications, highlighting advancements in integrating knowledge graphs with generative models.
Why It Matters
The study addresses critical challenges in applying RAG to structured knowledge graphs, which is essential for enhancing information retrieval in e-commerce and other domains. By improving retrieval accuracy and contextual grounding, the findings can significantly impact the development of domain-specific AI assistants.
Key Takeaways
- RAG enhances factual accuracy by integrating external knowledge sources with generative models.
- The study evaluates multiple retriever-reranker configurations on a production-scale e-commerce dataset.
- Experimental results show significant improvements in retrieval metrics over existing benchmarks.
- The findings provide a practical framework for deploying RAG systems in various domains.
- Addressing the integration of structured data with generative models is crucial for future AI applications.
Computer Science > Information Retrieval arXiv:2602.22219 (cs) [Submitted on 14 Dec 2025] Title:Comparative Analysis of Neural Retriever-Reranker Pipelines for Retrieval-Augmented Generation over Knowledge Graphs in E-commerce Applications Authors:Teri Rumble, Zbyněk Gazdík, Javad Zarrin, Jagdeep Ahluwalia View a PDF of the paper titled Comparative Analysis of Neural Retriever-Reranker Pipelines for Retrieval-Augmented Generation over Knowledge Graphs in E-commerce Applications, by Teri Rumble and 3 other authors View PDF HTML (experimental) Abstract:Recent advancements in Large Language Models (LLMs) have transformed Natural Language Processing (NLP), enabling complex information retrieval and generation tasks. Retrieval-Augmented Generation (RAG) has emerged as a key innovation, enhancing factual accuracy and contextual grounding by integrating external knowledge sources with generative models. Although RAG demonstrates strong performance on unstructured text, its application to structured knowledge graphs presents challenges: scaling retrieval across connected graphs and preserving contextual relationships during response generation. Cross-encoders refine retrieval precision, yet their integration with structured data remains underexplored. Addressing these challenges is crucial for developing domain-specific assistants that operate in production environments. This study presents the design and comparative evaluation of multiple Retriever-Reranker pipelines for knowledg...
