Hierarchical Multi-field Representations for Two-Stage E-commerce Retrieval
Niklas Freymuth, Dong Liu, Thomas Ricatte, Saab Mansour · 2025
Read Hierarchical Multi-field Representations for Two-Stage E-commerce Retrieval by Niklas Freymuth, Dong Liu, Thomas Ricatte, Saab Mansour (2025) 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-2501-18707.txt.
Description
Niklas Freymuth, Dong Liu, Thomas Ricatte, and Saab Mansour speed up shopping search in a 2025 paper. Online stores hold facts about each product in fields such as brand, title, and details. Most search models flatten those fields into one string and lose the structure. Their model, CHARM, keeps the fields apart and layers them by rank.
A special attention pattern lets higher fields shade the reading of lower ones. The model yields both a single summary vector per product and one vector per field. Search then runs in two stages: fast picking with the summary, then fine ranking with the field vectors. On large public store data sets, CHARM matched or beat the top rival systems. It also showed which field each query leaned on.
Catalog record
| CREATOR | Niklas Freymuth, Dong Liu, Thomas Ricatte, Saab Mansour |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 52 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2501-18707