Domain-Adaptive and Scalable Dense Retrieval for Content-Based Recommendation

Mritunjay Pandey · 2026

Read Domain-Adaptive and Scalable Dense Retrieval for Content-Based Recommendation by Mritunjay Pandey (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2602-00899.txt.

Description

Mritunjay Pandey casts content-based recommendation as recommendation-as-retrieval, where an intent signal retrieves the top-K relevant items via semantic similarity. Sparse keyword matching like BM25 breaks under vocabulary mismatch when intent barely overlaps product metadata. His two-tower bi-encoder, fine-tuned on Amazon Reviews 2023 Fashion, trains on 50,000 sampled interactions. FAISS HNSW indexing plus an ONNX Runtime pipeline with INT8 quantization serves results efficiently. Over 826,402 catalog items, a review-to-title benchmark shows Recall@10 more than doubles over BM25, meeting practical latency needs.

Catalog record

CREATORMritunjay Pandey
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE34778
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2602.00899

Browse the full library · About the Tower of Babel