Smart Maritime Literature Recommendation System Using Retrieval-Augmented Generation

Abstract

This study presents the development and evaluation of a Smart Literature Recommendation System (SRLC) based on Retrieval-Augmented Generation (RAG) technology to support final project (Tugas Akhir) research at Politeknik Pelayaran Surabaya. Unlike conventional keyword-based catalog search (OPAC), the SRLC employs semantic search using Gemini embedding vectors stored in ChromaDB to understand the meaning and context of user queries rather than relying on exact keyword matches. The system was populated with 6,640 bibliographic records comprising 6,261 entries from the Politeknik Pelayaran Surabaya public library catalog and 379 publications from faculty Google Scholar profiles. Each record was embedded as a high-dimensional vector, enabling natural language queries such as topic descriptions to retrieve semantically relevant literature with AI-generated explanations of relevance. The system was built using FastAPI, ChromaDB, Gemini 2.5 Flash, and React 18, deployed on a VPS accessible to cadets. Information retrieval evaluation using 15 ground truth queries yielded Precision@5 of 0.55, Recall@10 of 1.00, and Mean Reciprocal Rank of 0.90, demonstrating strong recall and ranking performance with moderate precision. The system provides cadets with 24-hour access to intelligent literature discovery, addressing the limitation of keyword-based search and restricted librarian availability. Results indicate that RAG-based recommendation systems can effectively enhance academic library services in specialized educational institutions