@inproceedings{singh-das-2026-cultrag,
    title = "{C}ult{RAG} at {S}em{E}val-2026 Task 7: Hybrid Sparse-Dense Retrieval with Entity-Centric Knowledge Bases for Cultural {MCQ} Answering",
    author = "Singh, Aditya  and
      Das, Rickarya",
    editor = "Kochmar, Ekaterina  and
      Ghosh, Debanjan  and
      North, Kai  and
      Komachi, Mamoru",
    booktitle = "Proceedings of the 20th {I}nternational {W}orkshop on {S}emantic {E}valuation (2026)",
    month = jul,
    year = "2026",
    address = "San Diego, California, USA",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/ingest-acl-workshops/2026.semeval-1.393/",
    pages = "3137--3142",
    ISBN = "979-8-89176-414-9",
    abstract = "We developed CultRAG, a trust-weighted Retrieval-Augmented Generation system for BLEnD Track 2 (SemEval-2026 Task 7), targeting culturally grounded multiple-choice QA across 30 countries. Built on Llama-3.1-8B-Instruct, the six-phase pipeline integrates entity extraction via spaCy, hybrid BM25+FAISS retrieval with Reciprocal Rank Fusion, country-aware filtering, keyword-based intent detection, tiered prompt routing, anti-leak quality filtering to suppress answer-anchoring artifacts, and trust-weighted document reranking with source-credibility tiers. Ablation analysis across eight cumulative configurations and per-country decomposition identify which components contribute and where retrieval helps versus hurts, informing future directions for confidence-conditioned selective retrieval."
}