Position-Aware Sequential Attention for Accurate Next Item Recommendations
Timur Nabiev, Evgeny Frolov · 2026
Read Position-Aware Sequential Attention for Accurate Next Item Recommendations by Timur Nabiev, Evgeny Frolov (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-2602-21052.txt.
Description
In-order tip models usually add place tags at the input, but that trick only weakly teaches order. Timur Nabiev and Evgeny Frolov argue the flaw runs deeper in this 2026 arXiv paper. With no place signals, focus is unchanged by reordering, and added tags tangle place with item meaning while fading in deep nets. Their answer is a kernel self-focus tool.
A learnable place kernel works purely in place space, untangled from meaning likeness, and tunes focus weights straight. Applied per focus block, it allows tunable multi-scale in-order model work. Tests on standard next-item trials show steady gains over strong base cases. Readers get a cleaner way to teach sequence models about order.
Catalog record
| CREATOR | Timur Nabiev, Evgeny Frolov |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 40 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-21052