<![CDATA[This study developed and evaluated a BM25-based question-answering system to support final-project information retrieval for Informatics Engineering students at Universitas Islam Sultan Agung. The system used the official Final Project Guidebook and Supervision Logbook as its knowledge corpus and integrated document reconstruction, heading-aware chunking, paragraph overlap, lexical preprocessing, synonym-based query expansion, heading-weighted indexing, BM25 ranking, and source-grounded answer presentation. Evaluation was conducted using 50 manually mapped ground-truth queries and measured through Recall@1, Recall@3, Recall@5, and Mean Reciprocal Rank. The results showed that 41 queries retrieved the relevant passage at rank 1, eight at rank 2, and one at rank 3. Recall@1 reached 82.00%, while Recall@3 and Recall@5 both achieved 100.00%, with an MRR of 0.9067. These findings indicate that the proposed lexical retrieval architecture can reliably recover authoritative academic information while maintaining transparency, interpretability, and source traceability. The remaining limitation concerns first-rank discrimination among procedurally similar passages, suggesting that future improvements should prioritize reranking or hybrid retrieval without sacrificing complete Top-3 coverage. The approach also provides a reproducible baseline for evaluating more complex semantic retrieval methods in academic information environments. ]]>
Copyrights © 2026
