---
title: "ASTraNet at SemEval-2026 Task 13: Not All Code Looks the"
description: "The paper describes the ASTraNet team's entry to SemEval-2026 Task 13, which asks systems to tell… 15 references, cited by 1 paper here."
authors: ["Ruwad Naswan", "Dipit Saha", "Md. Kabir", "Nabiha Tahseen"]
year: 2026
venue: "Proceedings of the 20th International Workshop on Semantic Evaluation (2026)"
source: "https://aclanthology.org/2026.semeval-1.402/"
doi: "10.18653/v1/2026.semeval-1.402"
licence: "CC BY 4.0 (ACL Anthology)"
licence_url: "https://creativecommons.org/licenses/by/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/naswan-2026-astranet-semeval-2026-task-13/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/naswan-2026-astranet-semeval-2026-task-13/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2026](https://md.research.elamur.ai/free/en/papers/2026/page/10/) / ASTraNet at SemEval-2026 Task 13: Not All Code Looks the Same: Multi-View Structural and Semantic Detection of Machine-Generated Code

# ASTraNet at SemEval-2026 Task 13: Not All Code Looks the Same: Multi-View Structural and Semantic Detection of Machine-Generated Code

**Authors:** [Ruwad Naswan](https://md.research.elamur.ai/free/en/authors/ruwad-naswan/), Dipit Saha, Md. Kabir and Nabiha Tahseen · **Year:** 2026 · **Venue:** [Proceedings of the 20th International Workshop on Semantic Evaluation (2026)](https://md.research.elamur.ai/free/en/venues/semeval/) ([SemEval 2026](https://md.research.elamur.ai/free/en/venues/semeval/2026/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2026.semeval-1.402/) · [PDF](https://aclanthology.org/2026.semeval-1.402.pdf) · [DOI 10.18653/v1/2026.semeval-1.402](https://doi.org/10.18653/v1/2026.semeval-1.402) · [OpenAlex](https://openalex.org/W7166866597)

**Citations:** 0 citations (OpenAlex, 2026-10-14) · [1](https://md.research.elamur.ai/free/en/papers/naswan-2026-astranet-semeval-2026-task-13/cited-by/) in our library, of the 149,439 papers whose bibliographies we read

## Summary
The paper describes the ASTraNet team's entry to SemEval-2026 Task 13, which asks systems to tell machine-generated code from human-written code. The test data contain programming languages and domains that never appear in training. Subtask A is binary, with 500K training samples in C++, Python and Java. Subtask C has 900K samples and four classes: human, machine, hybrid and adversarial.

The team trained three code-pretrained encoders, CodeT5p-220M, CodeBERT and UniXcoder. Each was kept frozen and given attention pooling, a supervised contrastive projection head and a focal-loss classifier. Alongside these, they built a structure-first detector: a GGNN over canonicalized Flow-Augmented ASTs (FA-AST) that ignores surface tokens.

After the competition, they added two ensembles:
- **Subtask A:** a layered rank fusion of the three encoders, raising macro F1 from 0.559 to 0.643.
- **Subtask C:** a stacking pipeline that combines neural outputs with LightGBM features and then routes predictions by class priority, raising macro F1 from 0.585 to 0.652.

Validation scores near 0.98 fell sharply on the test set, which shows how severe the distribution shift is. The authors conclude that, under such shift, diversity among models helps more than the strength of any single model.

## Key points
- **FA-AST pipeline:**
  - Code is parsed with Tree-sitter, trying seven grammars and keeping the parse with the fewest error nodes.
  - Node types are mapped to 137 unified labels that cover 267 language-specific types.
  - The graph uses 8 edge types and is encoded by a 4-layer, 256-d GGNN.
- **Subtask A test macro F1:**
  - FA-AST: 0.493.
  - CodeT5p-220M: 0.559.
  - Rank-fusion ensemble (post-competition): 0.643. It exploits low inter-model Spearman correlation between the encoders (r=0.35–0.57).
- **Subtask C ensemble:**
  - LightGBM is trained on TF-IDF character 2–4-grams, 32 stylometric features and 40 StarCoder2-3B statistics (perplexity, GLTR ranks, entropy).
  - A LightGBM meta-learner then combines these predictions with the neural ones over 17 meta-features.
  - Threshold routing checks the adversarial class first.
- **Distribution shift:**
  - Subtask A validation macro F1 was 0.980 for both CodeT5p and FA-AST, versus 0.559 and 0.493 on test.
  - Subtask C fell from 0.810 on validation to 0.585 on test.
- **Error analysis:** most errors involve brief code fragments of fewer than 20 tokens drawn from domains absent in training, or terse competitive-programming solutions whose style resembles LLM output.

*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

- [Bangladesh University of Engineering and Technology](https://md.research.elamur.ai/free/en/universities/bangladesh-university-of-engineering-and-technology/) — University, Dhaka, Bangladesh · 7 works · 308 citations

📚 Shelves (5): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 18,614 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 17,965 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 17,450 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 17,296 works · [🗄️ Information Systems](https://md.research.elamur.ai/free/en/shelves/computer-science/information-systems/) · 1,730 works

## Cite

Cite as: Ruwad Naswan, Dipit Saha, Md. Kabir and Nabiha Tahseen. 2026. ASTraNet at SemEval-2026 Task 13: Not All Code Looks the Same: Multi-View Structural and Semantic Detection of Machine-Generated Code. Proceedings of the 20th International Workshop on Semantic Evaluation (2026).

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@inproceedings{naswan2026astranet,
  title = {{ASTraNet at SemEval-2026 Task 13: Not All Code Looks the Same: Multi-View Structural and Semantic Detection of Machine-Generated Code}},
  author = {Ruwad Naswan, Dipit Saha, Md. Kabir and Nabiha Tahseen},
  booktitle = {Proceedings of the 20th International Workshop on Semantic Evaluation (2026)},
  year = {2026},
  doi = {10.18653/v1/2026.semeval-1.402},
  url = {https://doi.org/10.18653/v1/2026.semeval-1.402}
}
```

- [← previous arXiv2Table: Toward Realistic Benchmarking and Evaluation for LLM-Based Literature-Review Table Generation](https://md.research.elamur.ai/free/en/papers/wang-2026-arxiv2table-realistic-benchmarking-evaluation-llm/)
- [next → Attention Sinks Are Provably Necessary in Softmax Transformers: Evidence from Trigger-Conditional Tasks](https://md.research.elamur.ai/free/en/papers/milo-2026-attention-sinks-are-provably-necessary/)
- [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/10/)

## References

15 references, 4 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Aylin Caliskan-Islam, Richard Harang, Andrew Liu, Arvind Narayanan, Clare Voss, Fabian Yamaguchi, and Rachel Greenstadt. 2015. De-anonymizing programmers via code stylometry. In Proceedings of the 24th USENIX Security Symposium, pages 255–270. [find](https://scholar.google.com/scholar?q=Aylin+Caliskan%2DIslam%2C+Richard+Harang%2C+Andrew+Liu%2C+Arvind+Narayanan%2C+Clare+Voss%2C+Fabian+Yamaguchi%2C+and+Rachel+Greenstadt%2E+2015%2E+De%2Danonymizing+programmers+via+code+stylometry%2E+In+Proceedings+of+the+24t)
2. Zhangyin Feng, Daya Guo, Duyu Tang, Nan Duan, Xiaocheng Feng, Ming Gong, Linjun Shou, Bing Qin, Ting Liu, Daxin Jiang, and Ming Zhou. 2020. Code-BERT: A pre-trained model for programming and natural languages. In Findings of the Association for Computational Linguistics: EMNLP 2020, pages 1536–1547.… [find](https://scholar.google.com/scholar?q=Zhangyin+Feng%2C+Daya+Guo%2C+Duyu+Tang%2C+Nan+Duan%2C+Xiaocheng+Feng%2C+Ming+Gong%2C+Linjun+Shou%2C+Bing+Qin%2C+Ting+Liu%2C+Daxin+Jiang%2C+and+Ming+Zhou%2E+2020%2E+Code%2DBERT%3A+A+pre%2Dtrained+model+for+programming+and+natural+l)
3. Sebastian Gehrmann, Hendrik Strobelt, and Alexander Rush. 2019. [GLTR: Statistical detection and visualization of generated text](https://md.research.elamur.ai/free/en/papers/gehrmann-2019-gltr-statistical-detection-visualization-generated/). In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics: System Demonstrations, pages 111–116, Florence, Italy. Association for Computat…
4. Daya Guo, Shuai Lu, Nan Duan, Yanlin Wang, Ming Zhou, and Jian Yin. 2022. [UniXcoder: Unified cross-modal pre-training for code representation](https://md.research.elamur.ai/free/en/papers/guo-2022-unixcoder-unified-cross-modal-pre/). In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 7212–7225. Association for Computat…
5. Vaibhavi Kalgutkar, Ratinder Kaur, Hugo Gonzalez, Natalia Stakhanova, and Alina Matyukhina. 2019. Code authorship attribution: Methods and challenges. ACM Computing Surveys, 52(1):1–36. [find](https://scholar.google.com/scholar?q=Vaibhavi+Kalgutkar%2C+Ratinder+Kaur%2C+Hugo+Gonzalez%2C+Natalia+Stakhanova%2C+and+Alina+Matyukhina%2E+2019%2E+Code+authorship+attribution%3A+Methods+and+challenges%2E+ACM+Computing+Surveys%2C+52%281%29%3A1%E2%80%9336%2E)
6. Anton Lozhkov, Raymond Li, Loubna Ben Allal, and 1 others. 2024. StarCoder 2 and The Stack v2: The next generation. arXiv preprint arXiv:2402.19173. [find](https://scholar.google.com/scholar?q=Anton+Lozhkov%2C+Raymond+Li%2C+Loubna+Ben+Allal%2C+and+1+others%2E+2024%2E+StarCoder+2+and+The+Stack+v2%3A+The+next+generation%2E+arXiv+preprint+arXiv%3A2402%2E19173%2E)
7. Daniil Orel, Dilshod Azizov, and Preslav Nakov. 2025a. CoDet-M4: Detecting machine-generated code in multi-lingual, multi-generator and multi-domain settings. In Findings of the Association for Computational Linguistics: ACL 2025, pages 10570–10593, Vienna, Austria. Association for Computational Lin… [find](https://scholar.google.com/scholar?q=Daniil+Orel%2C+Dilshod+Azizov%2C+and+Preslav+Nakov%2E+2025a%2E+CoDet%2DM4%3A+Detecting+machine%2Dgenerated+code+in+multi%2Dlingual%2C+multi%2Dgenerator+and+multi%2Ddomain+settings%2E+In+Findings+of+the+Association+for+Comput)
8. Daniil Orel, Dilshod Azizov, Indraneil Paul, Yuxia Wang, Iryna Gurevych, and Preslav Nakov. 2026a. [AICD bench: A challenging benchmark for AI-generated code detection](https://md.research.elamur.ai/free/en/papers/orel-2026-aicd-bench-challenging-benchmark-ai/). In Proceedings of the 19th Conference of the European Chapter of the Association for Computational Linguistics (Volume 1: Long Paper…
9. Daniil Orel, Dilshod Azizov, Indraneil Paul, Yuxia Wang, Iryna Gurevych, and Preslav Nakov. 2026b. [SemEval-2026 task 13: Detecting machine-generated code with multiple programming languages, generators, and application scenarios](https://md.research.elamur.ai/free/en/papers/orel-2026-semeval-2026-task-13-detecting/). In Proceedings of the 20th International Workshop on Semantic Evaluati…
10. Daniil Orel, Indraneil Paul, Iryna Gurevych, and Preslav Nakov. 2025b. Droid: A resource suite for AI-generated code detection. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 31251–31277. [find](https://scholar.google.com/scholar?q=Daniil+Orel%2C+Indraneil+Paul%2C+Iryna+Gurevych%2C+and+Preslav+Nakov%2E+2025b%2E+Droid%3A+A+resource+suite+for+AI%2Dgenerated+code+detection%2E+In+Proceedings+of+the+2025+Conference+on+Empirical+Methods+in+Natural+La)
11. Hung Pham, Huyen Ha, Van Tong, Dung Hoang, Duc Tran, and Tuyen Ngoc Le. 2024. MAGECODE: Machine-generated code detection method using large language models. IEEE Access. [find](https://scholar.google.com/scholar?q=Hung+Pham%2C+Huyen+Ha%2C+Van+Tong%2C+Dung+Hoang%2C+Duc+Tran%2C+and+Tuyen+Ngoc+Le%2E+2024%2E+MAGECODE%3A+Machine%2Dgenerated+code+detection+method+using+large+language+models%2E+IEEE+Access%2E)
12. Dequan Wang, Evan Shelhamer, Shaoteng Liu, Bruno Olshausen, and Trevor Darrell. 2021. TENT: Fully test-time adaptation by entropy minimization. In Proceedings of the 9th International Conference on Learning Representations (ICLR 2021). [find](https://scholar.google.com/scholar?q=Dequan+Wang%2C+Evan+Shelhamer%2C+Shaoteng+Liu%2C+Bruno+Olshausen%2C+and+Trevor+Darrell%2E+2021%2E+TENT%3A+Fully+test%2Dtime+adaptation+by+entropy+minimization%2E+In+Proceedings+of+the+9th+International+Conference+on+Le)
13. Wenhan Wang, Ge Li, Bo Ma, Xin Xia, and Zhi Jin. 2020. Detecting code clones with graph neural network and flow-augmented abstract syntax tree. In Proceedings of the 27th IEEE International Conference on Software Analysis, Evolution and Reengineering (SANER 2020), pages 261–271. IEEE. [find](https://scholar.google.com/scholar?q=Wenhan+Wang%2C+Ge+Li%2C+Bo+Ma%2C+Xin+Xia%2C+and+Zhi+Jin%2E+2020%2E+Detecting+code+clones+with+graph+neural+network+and+flow%2Daugmented+abstract+syntax+tree%2E+In+Proceedings+of+the+27th+IEEE+International+Conference)
14. Yue Wang, Hung Le, Akhilesh Deepak Gotmare, Nghi D.Q. Bui, Junnan Li, and Steven C.H. Hoi. 2023. CodeT5+: Open code large language models for code understanding and generation. arXiv preprint arXiv:2305.07922. [find](https://scholar.google.com/scholar?q=Yue+Wang%2C+Hung+Le%2C+Akhilesh+Deepak+Gotmare%2C+Nghi+D%2EQ%2E+Bui%2C+Junnan+Li%2C+and+Steven+C%2EH%2E+Hoi%2E+2023%2E+CodeT5%2B%3A+Open+code+large+language+models+for+code+understanding+and+generation%2E+arXiv+preprint+arXiv%3A23)
15. Xiaodan Xu, Chao Ni, Xinrong Guo, Shaoxuan Liu, Xiaoya Wang, Kui Liu, and Xiaohu Yang. 2025. Distinguishing LLM-generated from human-written code by contrastive learning. ACM Transactions on Software Engineering and Methodology, 34(91):1–31. [find](https://scholar.google.com/scholar?q=Xiaodan+Xu%2C+Chao+Ni%2C+Xinrong+Guo%2C+Shaoxuan+Liu%2C+Xiaoya+Wang%2C+Kui+Liu%2C+and+Xiaohu+Yang%2E+2025%2E+Distinguishing+LLM%2Dgenerated+from+human%2Dwritten+code+by+contrastive+learning%2E+ACM+Transactions+on+Software+)

*Updated 31 October 2026*