Semantic Retrieval for Product Search in E-Commerce
Fuente:
arXiv
Guardado en:
| Autores principales: | , , , , , |
|---|---|
| Formato: | Preprint |
| Publicado: |
2026
|
| Materias: | |
| Acceso en línea: | |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| _version_ | 1866911738973126656 |
|---|---|
| author | Kothari, Nikhil Samdani, Saksham Mallick, Ritam Gupta, Praveen Vijay, Ankit Kumar, Surender |
| author_facet | Kothari, Nikhil Samdani, Saksham Mallick, Ritam Gupta, Praveen Vijay, Ankit Kumar, Surender |
| contents | Semantic retrieval in e-commerce must handle short, noisy, and colloquial queries over large product catalogs with fine-grained attribute distinctions. We present a Siamese LLM dual-encoder trained through a two-stage pipeline: contrastive learning with a false-negative margin mask to prevent penalization of near-duplicate products, followed by Relative Odds Alignment for Retrieval (ROAR), a preference optimization objective that extends Bradley-Terry to variable-sized graded relevance groups via consecutive odds-ratio margins. The training corpus mirrors this progression - substitute query-product pairs provide coarse semantic supervision in Stage 1 and graded relevance annotations drive fine-grained ranking in Stage 2. The resulting system accurately retrieves exact matches while correctly ordering substitutes and complementary products, with gains confirmed across query-frequency strata and business verticals, and statistical significance validated through live A/B deployment at scale. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2606_01504 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Semantic Retrieval for Product Search in E-Commerce Kothari, Nikhil Samdani, Saksham Mallick, Ritam Gupta, Praveen Vijay, Ankit Kumar, Surender Information Retrieval Machine Learning Semantic retrieval in e-commerce must handle short, noisy, and colloquial queries over large product catalogs with fine-grained attribute distinctions. We present a Siamese LLM dual-encoder trained through a two-stage pipeline: contrastive learning with a false-negative margin mask to prevent penalization of near-duplicate products, followed by Relative Odds Alignment for Retrieval (ROAR), a preference optimization objective that extends Bradley-Terry to variable-sized graded relevance groups via consecutive odds-ratio margins. The training corpus mirrors this progression - substitute query-product pairs provide coarse semantic supervision in Stage 1 and graded relevance annotations drive fine-grained ranking in Stage 2. The resulting system accurately retrieves exact matches while correctly ordering substitutes and complementary products, with gains confirmed across query-frequency strata and business verticals, and statistical significance validated through live A/B deployment at scale. |
| title | Semantic Retrieval for Product Search in E-Commerce |
| topic | Information Retrieval Machine Learning |
| url | https://arxiv.org/abs/2606.01504 |