---
格式版本: 2
标题: "UniDot: A Unified Network for Sequence Modeling and Feature Interaction in Large-scale Recommendation"
原文链接: "https://arxiv.org/abs/2608.16797"
发布日期: "2026-08-17"
发布时间校准状态: "found"
发布时间需复核: "否"
发布时间来源: "rule:scrape:strict_html_body"
发布时间证据: "citation_date: 2026/08/17"
发布时间校准原因: "规则确认唯一严格发布时间，来源 scrape:strict_html_body"
发布时间校准置信度: "high"
发布时间候选数量: 6
发布时间严格候选数量: 6
发布时间原页读取状态: ""
发布时间未找到原因: ""
发布时间校准时间: "2026-08-18T19:33:43+08:00"
发布时间仲裁状态: "skipped"
发布时间仲裁尝试次数: 0
发布时间仲裁耗时毫秒: 0
发现时间: "2026-08-18T19:33:29+08:00"
入库时间: "2026-08-18T11:33:52.402Z"
来源平台: "arXiv 学术论文搜索"
搜索渠道: "source_template"
搜索词: "https://arxiv.org/search/?query=latency&searchtype=all"
匹配关键词:
  - "latency"
  - "AI"
相关厂家:
  []
相关专家:
  []
内容类型: "网页"
抓取工具: "Free Fetch + Defuddle"
清洗工具: "Defuddle Markdown + Defuddle/Readability 正文提取"
原始附件:
  []
AI优质: "否"
AI打分: 14
AI分档: "非优质"
AI质检状态: "不通过"
AI打分理由: "论文为推荐系统算法研究，与超节点/AI Rack/机柜级AI基础设施完全无关，仅命中'latency'搜索词但无相关技术或商业内容。"
AI质检模型: "tx-deepseek-v4-flash"
AI质检时间: "2026-08-18T19:34:50+08:00"
AI主题相关性: 0
AI来源权威性: 5
AI新颖性: 2
AI技术细节: 2
AI商业部署信号: 0
AI完整性: 5
AI摘要: "UniDot提出一种统一网络架构，将特征交互与用户行为序列建模整合到同一token空间，通过并行token-mixing与sequence-retrieval总线及FM Highway实现深度融合。"
AI摘要模型: "ali-deepseek-v4-flash"
AI摘要时间: "2026-09-07T03:25:04.820Z"
采集批次: "2026年8月18日15点52分05秒"
采集批次ID: "20260818-155205-090"
去重键: "https://arxiv.org/abs/2608.16797"
---

## Computer Science > Information Retrieval

## Title:UniDot: A Unified Network for Sequence Modeling and Feature Interaction in Large-scale Recommendation

Authors:[Rongcheng Lin](https://arxiv.org/search/cs?searchtype=author&query=Lin,+R), [Yan Sun](https://arxiv.org/search/cs?searchtype=author&query=Sun,+Y), [Jamey Zhang](https://arxiv.org/search/cs?searchtype=author&query=Zhang,+J), [Guanglei Xiong](https://arxiv.org/search/cs?searchtype=author&query=Xiong,+G), [Ivan Ji](https://arxiv.org/search/cs?searchtype=author&query=Ji,+I), [Xianjie Chen](https://arxiv.org/search/cs?searchtype=author&query=Chen,+X), [Shujian Bu](https://arxiv.org/search/cs?searchtype=author&query=Bu,+S)

[View PDF](https://arxiv.org/pdf/2608.16797) [HTML (experimental)](https://arxiv.org/html/2608.16797v1)

> Abstract:Industrial recommenders rely on two model families that have evolved largely independently: feature-interaction models over multi-field user/item features, and sequential models over user-behavior histories. Production systems couple them only loosely. To unify the two, we present UniDot, a novel architecture for post-click conversion prediction built from the factorization-machine (FM) point of view: the embedding inner product---which powers collaborative filtering and lets a recommender generalize to unseen user--item pairs---is the same primitive as attention's query dot key scoring, so a single dot-product of tokens can underlie both feature interaction and sequence modeling. UniDot tokenizes non-sequential fields and multi-domain behavioral sequences into one shared token space and stacks a single macro-block in which a token-mixing bus and a sequence-retrieval bus (item tokens cross-attending the histories) run in parallel and exchange state each layer through an MLP-Mixer fusion, while an FM Highway carries explicit per-layer dot-product interactions around the residual stack directly to the classifier. The sequence side is embedded once per forward pass and shared by all consumers, bounding inference latency. Trained with a dual sparse/dense (Adagrad + Muon) optimizer, an auxiliary conversion-delay head, and multi-path mutual learning, UniDot finished as the runner-up on the Industrial track of the TAAC KDD Cup 2026.

| Subjects: | Information Retrieval (cs.IR); Artificial Intelligence (cs.AI) |
| --- | --- |
| Cite as: | [arXiv:2608.16797](https://arxiv.org/abs/2608.16797) \[cs.IR\] |
|  | (or [arXiv:2608.16797v1](https://arxiv.org/abs/2608.16797v1) \[cs.IR\] for this version) |
|  | [https://doi.org/10.48550/arXiv.2608.16797](https://doi.org/10.48550/arXiv.2608.16797) |
| Journal reference: | KDD 2026 UniRec Workshop |

## Submission history

From: Rongcheng Lin \[[view email](https://arxiv.org/show-email/3a71dc8e/2608.16797)\]  
**\[v1\]** Mon, 17 Aug 2026 16:52:56 UTC (73 KB)

[Which authors of this paper are endorsers?](https://arxiv.org/auth/show-endorsers/2608.16797) | Disable MathJax ([What is MathJax?](https://info.arxiv.org/help/mathjax.html))