Elamur Research⌕
☰
home / papers / 2026 / TLSA: LLM-Guided Text-Label Space Alignment with Contrastive Learning for Generalized Category Discovery

TLSA: LLM-Guided Text-Label Space Alignment with Contrastive Learning for Generalized Category Discovery

Paper · 2026 · Wenxi Xu, Chuan Qin, Xi Chen et al.

Summary

Generalized Category Discovery (GCD) trains a classifier on a small labeled set of known classes plus a large unlabeled pool that also contains unseen classes. The model must recognize the known classes and also form the new ones. The authors identify three weaknesses in current methods:

  • labels are used as bare one-hot IDs;
  • known and novel classes are trained with different objectives, which biases the model toward known classes;
  • close categories blur together in embedding space.

Their framework, TLSA, addresses these with four components:

  • Shared space for texts and labels. A BERT-based dual encoder maps both into one space and is trained with a symmetric contrastive loss plus a ranking term for hard negatives.
  • Naming new classes. Novel clusters are found through clustering and Hungarian matching, and an LLM then gives each one an explicit name.
  • Separating confusable clusters. A graph-based step makes the LLM rename clusters whose labels overlap too much.
  • Balanced pseudo-labeling. Pseudo-labeled examples are drawn from high-, medium- and low-confidence groups.

On four benchmarks, TLSA reports the best average results among GCD methods. Treating labels as meaningful text, rather than as indices, makes discovered categories easier to interpret and helps the model extend to new classes.

Key points

  • Texts and label descriptions go through the same BERT backbone but use separate MLP projection heads. Classification works by matching a text to the most similar label embedding instead of using a fixed classifier head.
  • With 25% known classes, the best confidence-stratified sampling mix was (8, 12, 16) samples (high/medium/low confidence). It reached 74.97 ACC, 65.31 ARI and 85.82 NMI on BANKING, and 76.87 ACC on HWU. Mixed-confidence settings beat single-level ones.
  • The best weight for the hard-negative ranking loss was α = 0.5.
  • Changing the prompting strategy on BANKING raised ACC from 74.97 (Direct) to 76.22 (Self-Refine); Chain-of-Thought gave 76.14.
  • The LLM used was a locally deployed DeepSeek-V3; code is published at github.com/Wenxi-Xu/TLSA.

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

🏢BaiduBeijing · China74 works · 15,263 citations🎓University of Science and Technology of ChinaHefei · China108 works · 112,698 citations🏛️Chinese Academy of SciencesBeijing · China194 works · 23,906 citations🎓University of Chinese Academy of SciencesBeijing · China147 works · 8,480 citations🔬Computer Network Information CenterBeijing · China3 works · 22 citations

 Fields, topics and concepts

📚 Shelves 5The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 worksComputer Vision and Pattern Recognition · 2,869 works

🧭 Topics 1Research subjects and languages the work is about.Language models & pretraining · 3,320 works

Cite

Cite as Wenxi Xu, Chuan Qin, Xi Chen, Chuyu Fang, Yuanchun Zhou and Hengshu Zhu. 2026. TLSA: LLM-Guided Text-Label Space Alignment with Contrastive Learning for Generalized Category Discovery. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.acl-long.869

@inproceedings{xu2026tlsa,
  title = {{TLSA: LLM-Guided Text-Label Space Alignment with Contrastive Learning for Generalized Category Discovery}},
  author = {Wenxi Xu, Chuan Qin, Xi Chen, Chuyu Fang, Yuanchun Zhou and Hengshu Zhu},
  booktitle = {Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
  year = {2026},
  doi = {10.18653/v1/2026.acl-long.869},
  url = {https://doi.org/10.18653/v1/2026.acl-long.869}
}

References

33 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Wenbin An, Haonan Lin, Jiahao Nie, Feng Tian, Wenkai Shi, Yaqiang Wu, Qianying Wang, and Ping Chen. 2025. Unleashing the potential of model bias for generalized category discovery. Proceedings of the AAAI Conference on Artificial Intelligence, 39(15):15365– 15373. find
  2. Wenbin An, Wenkai Shi, Feng Tian, Haonan Lin, QianYing Wang, Yaqiang Wu, Mingxiang Cai, Luyan Wang, Yan Chen, Haiping Zhu, and Ping Chen. 2024a. Generalized category discovery with large language models in the loop. In Findings of the Association for Computational Linguistics: ACL 2024, pages 8653–… find
  3. Wenbin An, Feng Tian, Ping Chen, Qinghua Zheng, and Wei Ding. 2023a. New user intent discovery with robust pseudo label training and source domain joint training. IEEE Intelligent Systems, 38(4):21–31. find
  4. Wenbin An, Feng Tian, Wenkai Shi, Yan Chen, Yaqiang Wu, Qianying Wang, and Ping Chen. 2024b. Transfer and alignment network for generalized category discovery. Proceedings of the AAAI Conference on Artificial Intelligence, 38(10):10856–10864. find
  5. Wenbin An, Feng Tian, Qinghua Zheng, Wei Ding, Qianying Wang, and Ping Chen. 2023b. Generalized category discovery with decoupled prototypical network. Proceedings of the AAAI Conference on Artificial Intelligence, 37(11):12527–12535. find
  6. Tom B. Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared Kaplan, Prafulla Dhariwal, Arvind Neelakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal, Ariel Herbert-Voss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel M. Ziegler, Jeffrey Wu, Clemens Winter,… — Language Models are Few-Shot Learners
  7. Iñigo Casanueva, Tadas Temčinas, Daniela Gerz, Matthew Henderson, and Ivan Vulić. 2020. Efficient intent detection with dual sentence encoders. In Proceedings of the 2nd Workshop on Natural Language Processing for Conversational AI, pages 38–45, Online. Association for Computational Linguistics.
  8. Xi Chen, Chuan Qin, Chuyu Fang, Chao Wang, Chen Zhu, Fuzhen Zhuang, Hengshu Zhu, and Hui Xiong. 2024. Job-sdf: A multi-granularity dataset for job skill demand forecasting and benchmarking. Advances in Neural Information Processing Systems, 37:129329–129356. find
  9. Xi Chen, Chuan Qin, Ziqi Wang, Shasha Hu, Chao Wang, Hengshu Zhu, and Hui Xiong. 2026. Beyond the known: An unknown-aware large language model for open-set text classification. In The Fourteenth International Conference on Learning Representations. find
  10. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. BERT: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies,…
  11. Xiaohan Huang, Meng Xiao, Chuan Qin, Qingqing Long, Jinmiao Chen, Yuanchun Zhou, and Hengshu Zhu. 2026. Scihorizon-gene: Benchmarking llm for life sciences inference from gene knowledge to functional understanding. Preprint, arXiv:2601.12805. — [arXiv:2601.12805](https://arxiv.org/abs/2601.12805) find
  12. Feihu Jiang, Chuan Qin, Kaichun Yao, Chuyu Fang, Fuzhen Zhuang, Hengshu Zhu, and Hui Xiong. 2024. find
  13. Enhancing question answering for enterprise knowledge bases using large language models. In Database Systems for Advanced Applications: 29th International Conference, DASFAA 2024, Gifu, Japan, July 2–5, 2024, Proceedings, Part IV, page 273–290, Berlin, Heidelberg. Springer-Verlag. find
  14. H. W. Kuhn. 1955. The hungarian method for the assignment problem. Naval Research Logistics Quarterly, 2(1-2):83–97. find
  15. Stefan Larson, Anish Mahendran, Joseph J. Peper, Christopher Clarke, Andrew Lee, Parker Hill, Jonathan K. Kummerfeld, Kevin Leach, Michael A. Laurenzano, Lingjia Tang, and Jason Mars. 2019. An evaluation dataset for intent classification and out-of-scope prediction. In Proceedings of the 2019 Confer… find
  16. Jinggui Liang, Lizi Liao, Hao Fei, Bobo Li, and Jing Jiang. 2024. Actively learn from LLMs with uncertainty propagation for generalized category discovery. In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologie… find
  17. Ting-En Lin, Hua Xu, and Hanlei Zhang. 2020. Discovering new intents via constrained deep adaptive clustering with cluster refinement. Proceedings of the AAAI Conference on Artificial Intelligence, 34(05):8360–8367. find
  18. Aixin Liu, Bei Feng, Bing Xue, Bingxuan Wang, Bochao Wu, Chengda Lu, Chenggang Zhao, Chengqi Deng, Chenyu Zhang, Chong Ruan, et al. 2025. Deepseek-v3 technical report. Preprint, arXiv:2412.19437. find
  19. Xingkun Liu, Arash Eshghi, Pawel Swietojanski, and Verena Rieser. 2021. Benchmarking Natural Language Understanding Services for Building Conversational Agents, pages 165–183. Springer Singapore, Singapore. find
  20. Aman Madaan, Niket Tandon, Prakhar Gupta, Skyler Hallinan, Luyu Gao, Sarah Wiegreffe, Uri Alon, Nouha Dziri, Shrimai Prabhumoye, Yiming Yang, Shashank Gupta, Bodhisattwa Prasad Majumder, Katherine Hermann, Sean Welleck, Amir Yazdanbakhsh, and Peter Clark. 2023. Self-refine: iterative refinement with… find
  21. Sridhama Prakhya, Vinodini Venkataram, and Jugal Kalita. 2017. Open set text classification using CNNs. In Proceedings of the 14th International Conference on Natural Language Processing (ICON-2017), pages 466–475, Kolkata, India. NLP Association of India. find
  22. Chuan Qin, Xin Chen, Chengrui Wang, Pengmin Wu, Xi Chen, Yihang Cheng, Jingyi Zhao, Meng Xiao, Xiangchao Dong, Qingqing Long, Boya Pan, Han Wu, Chengzan Li, Yuanchun Zhou, Hui Xiong, and Hengshu Zhu. 2025a. Scihorizon: Benchmarking ai-for-science readiness from scientific data to large language mode… find
  23. Chuan Qin, Chuyu Fang, Kaichun Yao, Xi Chen, Fuzhen Zhuang, and Hengshu Zhu. 2025b. Cotr: Efficient job task recognition for occupational information systems with class-incremental learning. ACM Trans. Manage. Inf. Syst., 16(2). find
  24. Chuan Qin, Le Zhang, Yihang Cheng, Rui Zha, Dazhong Shen, Qi Zhang, Xi Chen, Ying Sun, Chen Zhu, Hengshu Zhu, and Hui Xiong. 2025c. A comprehensive survey of artificial intelligence techniques for talent analytics. Proceedings of the IEEE, 113(2):125–171. and Denny Zhou. 2022. Chain-of-thought promp… find
  25. Hanlei Zhang, Hua Xu, Ting-En Lin, and Rui Lyu. 2021. Discovering new intents with deep aligned clustering. Proceedings of the AAAI Conference on Artificial Intelligence, 35(16):14365–14373. find
  26. Hanlei Zhang, Hua Xu, Xin Wang, Fei Long, and Kai Gao. 2024. A clustering framework for unsupervised and semi-supervised new intent discovery. IEEE Transactions on Knowledge and Data Engineering, 36(11):5468–5481. find
  27. Henry Peng Zou, Siffi Singh, Yi Nian, Jianfeng He, Jason Cai, Saab Mansour, and Hang Su. 2025. Glean: Generalized category discovery with diverse and quality-enhanced llm feedback. Preprint, arXiv:2502.18414. — [arXiv:2502.18414](https://arxiv.org/abs/2502.18414) find
  28. Wenkai Shi, Wenbin An, Feng Tian, Yan Chen, Yaqiang Wu, Qianying Wang, and Ping Chen. 2024. A unified knowledge transfer network for generalized category discovery. Proceedings of the AAAI Conference on Artificial Intelligence, 38(17):18961–18969. find
  29. Zhiheng Song, Jingshuai Zhang, Chuan Qin, Chao Wang, Chao Chen, Longfei Xu, Kaikui Liu, Xiangxiang Chu, and Hengshu Zhu. 2026. Mobility-bench: A benchmark for evaluating route-planning agents in real-world mobility scenarios. Preprint, arXiv:2602.22638. — [arXiv:2602.22638](https://arxiv.org/abs/2602.22638) find
  30. Kai Tang, Junbo Zhao, Xiao Ding, Runze Wu, Lei Feng, Gang Chen, and Haobo Wang. 2024. Learning geometry-aware representations for new intent discovery. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 5641–5654, Bangkok, Thaila… find
  31. Zhenyu Tong, Chuan Qin, Chuyu Fang, Kaichun Yao, Xi Chen, Jingshuai Zhang, Chen Zhu, and Hengshu Zhu. 2025. From missteps to mastery: Enhancing low-resource dense retrieval through adaptive query generation. In Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.1,… find
  32. Sagar Vaze, Kai Han, Andrea Vedaldi, and Andrew Zisserman. 2022. Generalized category discovery. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 7492–7501. find
  33. Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Brian Ichter, Fei Xia, Ed H. Chi, Quoc V. Le, find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/xu-2026-tlsa-llm-guided-text-label/