Elamur Research⌕
☰
home / papers / 2026 / TaoType: Predicting Fine-Grained Typing Intent for Faster Search

TaoType: Predicting Fine-Grained Typing Intent for Faster Search

Paper · 2026 · Yipeng Yu, Yichen Yuan, Chengxiao Feng et al.

Summary

In mobile shopping search, users often pause between finishing a query and actually pressing "search". This paper, from Alibaba's Taobao & Tmall Group, exploits that gap. After every keystroke, TaoType estimates whether the user will keep typing, submit the query, or pick a dropdown suggestion. When submission looks likely, the app quietly launches the search ahead of time. The model is a compact linear-chain CRF that runs on the device. It combines character-level text features with three groups of context signals: global session state (GLB), suggestion-list statistics (SUG), and typing timing (TMP). It is trained on a week of Taobao logs and tested on the following day. Offline, it outperforms tree ensembles, neural sequence models, and BERT-based baselines while staying very small. In online A/B tests, it cuts response time by roughly 640 ms, at the cost of a modest rise in server queries. The authors present the work as an example of "App Intelligence," meaning lightweight AI built into apps. It offers a way to reduce latency without compressing server-side ranking models.

Key points

  • Three intents are predicted per keystroke: keep typing (intent 1), search via button or keyboard key (intent 2A/2B), or click a suggestion (intent 3). The average delay before submission ranges from about 0.7 s to 3.6 s, depending on platform and submission method.
  • Offline, TaoType reaches precision/recall/accuracy of 0.7936/0.8196/0.7742 with a 78 KB model. The strongest baseline, BERT-CRF, scores 0.7555/0.7827/0.7352 at more than 160 KB. LightGBM, CatBoost, BiLSTM, TinyBERT, and Transformer models also fall behind.
  • The best online configuration (TaoType-o3) reduces response time by 640.51±93.65 ms, with about 4.19% wasted QPS. It beats random pre-search strategies on all metrics, and other configurations trade a smaller latency gain for fewer wasted queries.
  • Features include TEXT (character, language, position, length), GLB (e.g., typing count, deletions), SUG (number of suggestions, top suggestion score), and TMP (typing speed and timing).
  • Limitation: although the features are language-agnostic, the data comes mostly from Chinese-speaking users.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🏢Alibaba GroupHangzhou · China168 works · 15,386 citations

 Fields, topics and concepts

📚 Shelves 2The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksInformation Systems · 1,437 works

Cite

Cite as Yipeng Yu, Yichen Yuan, Chengxiao Feng and Xu Liu. 2026. TaoType: Predicting Fine-Grained Typing Intent for Faster Search. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 6: Industry Track).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.acl-industry.13

@inproceedings{yu2026taotype,
  title = {{TaoType: Predicting Fine-Grained Typing Intent for Faster Search}},
  author = {Yipeng Yu, Yichen Yuan, Chengxiao Feng and Xu Liu},
  booktitle = {Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 6: Industry Track)},
  year = {2026},
  doi = {10.18653/v1/2026.acl-industry.13},
  url = {https://doi.org/10.18653/v1/2026.acl-industry.13}
}

References

14 references, 2 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Zhuojiang Cai, Jingkai Hong, Zhimin Wang, and Feng Lu. 2025. Gazeswipe: Enhancing mobile touchscreen reachability through seamless gaze and finger-swipe integration. In Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems. find
  2. Anna Veronika Dorogush, Vasily Ershov, and Andrey Gulin. 2018. Catboost: gradient boosting with categorical features support. arXiv preprint arXiv:1810.11363. find
  3. Lu Fan, Qimai Li, Bo Liu, Xiao-Ming Wu, Xiaotong Zhang, Fuyu Lv, Guli Lin, Sen Li, Taiwei Jin, and Keping Yang. 2022. Modeling user behavior with graph convolution for personalized product search. In Proceedings of the ACM Web Conference 2022, page 203–212. find
  4. Long Guo, Lifeng Hua, Rongfei Jia, Fei Fang, Binqiang Zhao, and Bin Cui. 2020. Edgedipn: a unified deep intent prediction network deployed at the edge. Proc. VLDB Endow., 14(3):320–328. find
  5. Long Guo, Lifeng Hua, Rongfei Jia, Binqiang Zhao, Xiaobo Wang, and Bin Cui. 2019. Buying or browsing?: Predicting real-time purchasing intent using attention-based deep network with multiple behavior. In Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining… find
  6. Qi Guo and Eugene Agichtein. 2010. Ready to buy or just browsing? detecting web searcher goals from interaction data. In Proceedings of the 33rd International ACM SIGIR Conference on Research and Development in Information Retrieval, page 130–137. find
  7. Xiaoqi Jiao, Yichun Yin, Lifeng Shang, Xin Jiang, Xiao Chen, Linlin Li, Fang Wang, and Qun Liu. 2020. TinyBERT: Distilling BERT for natural language understanding. In Findings of the Association for Computational Linguistics: EMNLP 2020, pages 4163– 4174.
  8. Anil Kag, Igor Fedorov, Aditya Gangrade, Paul Whatmough, and Venkatesh Saligrama. 2022. Efficient edge inference by selective query. In The Eleventh International Conference on Learning Representations. find
  9. Jared Kaplan, Sam McCandlish, Tom Henighan, Tom B. Brown, Benjamin Chess, Rewon Child, Scott Gray, Alec Radford, Jeffrey Wu, and Dario Amodei. 2020. Scaling laws for neural language models. Preprint, arXiv:2001.08361.
  10. Guolin Ke, Qi Meng, Thomas Finley, Taifeng Wang, Wei Chen, Weidong Ma, Qiwei Ye, and Tie-Yan Liu. 2017. Lightgbm: A highly efficient gradient boosting decision tree. In Advances in Neural Information Processing Systems, volume 30. find
  11. John D. Lafferty, Andrew McCallum, and Fernando C. N. Pereira. 2001. Conditional random fields: Probabilistic models for segmenting and labeling sequence data. In Proceedings of the Eighteenth International Conference on Machine Learning, page 282–289. find
  12. Sen Li, Fuyu Lv, Taiwei Jin, Guiyang Li, Yukun Zheng, Tao Zhuang, Qingwen Liu, Xiaoyi Zeng, James Kwok, and Qianli Ma. 2022. Query rewriting in taobao search. In Proceedings of the 31st ACM International Conference on Information & Knowledge Management, page 3262–3271. find
  13. Wenqi Li, Jui-Ching Kuo, Manyu Sheng, Pengyi Zhang, and Qunfang Wu. 2025. Beyond explicit and implicit: How users provide feedback to shape personalized recommendation content. In Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems, pages 1–17. find
  14. Ji Lin, Ligeng Zhu, Wei-Ming Chen, Wei-Chen Wang, and Song Han. 2023. Tiny machine learning: find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/yu-2026-taotype-predicting-fine-grained-typing/