Elamur Research⌕
☰
home / papers / 2026 / Thesis Proposal: On the Granularity-Robustness Trade-off in Text-Derived Knowledge Graphs

Thesis Proposal: On the Granularity-Robustness Trade-off in Text-Derived Knowledge Graphs

Paper · 2026 · Surawat Pralomram

Summary

Dense-embedding RAG handles paraphrase well, but it packs many attributes into one vector, so extracting a specific value from text (a symptom, a dosage) is unreliable. Knowledge graphs expose attributes directly as nodes and edges. However, exact subgraph matching fails easily when wording, synonyms or extraction errors change the graph's shape. This student thesis proposal argues that the central design question is granularity: how far to split information into fine-grained triples versus merging recurring patterns into single canonical concepts. The author treats KG construction as an encoding task in which the ontology serves as a codebook. By analogy with Huffman coding, he suggests that representational capacity should follow the expected query workload. Attributes requested as answers should stay explicit, while patterns used as query conditions can be compacted. A prototype that aggregates local graph motifs into several alternative views points to better retrieval robustness, at the cost of a larger representation. The work matters because it outlines a route toward retrieval that tolerates linguistic variation and still supports precise attribute queries.

Key points

  • Three research questions guide the thesis:
    • how granularity affects robustness and accuracy of attribute-specific querying (RQ1);
    • how to build KGs from text in a query-driven way (RQ2);
    • how to match queries robustly under structural and lexical variation (RQ3).
  • Each query is divided into two components: a predicate component that must find a match, and a variable component whose value forms the answer. Frequent variables should remain separate nodes, while frequent predicate motifs may be collapsed, e.g. into right_lower_abdominal_pain.
  • The trade-off is likened to database normalization vs. denormalization and to SNOMED CT (fine-grained, compositional) vs. ICD-10 (coarse categories for statistics). Rate-distortion theory is named as a candidate formalism.
  • The prototype brute-forces motif aggregation over 1-hop edges, 1-hop edge pairs and 2-hop paths. The author describes this as a crude Monte Carlo approximation of possible representational views. It raises recall but inflates the representation.
  • The evaluation is small and exploratory. It includes only simple RAG baselines (Appendix A) and does not yet separate predicate matching from value extraction. Future work targets learned, domain-conditioned aggregation operators (e.g. Z′ = WZ) and query-conditioned view selection.

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

🎓Mahidol UniversityBangkok · Thailand3 works · 2 citations

 Fields, topics and concepts

📚 Shelves 4The 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 works

🧭 Topics 1Research subjects and languages the work is about.Knowledge graphs & ontologies · 192 works

Cite

Cite as Surawat Pralomram. 2026. Thesis Proposal: On the Granularity-Robustness Trade-off in Text-Derived Knowledge Graphs. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 4: Student Research Workshop).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.acl-srw.17

@inproceedings{pralomram2026thesis,
  title = {{Thesis Proposal: On the Granularity-Robustness Trade-off in Text-Derived Knowledge Graphs}},
  author = {Surawat Pralomram},
  booktitle = {Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 4: Student Research Workshop)},
  year = {2026},
  doi = {10.18653/v1/2026.acl-srw.17},
  url = {https://doi.org/10.18653/v1/2026.acl-srw.17}
}

References

33 references, 7 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Shqiponja Ahmetaj, Vasilis Efthymiou, Ronald Fagin, Phokion G. Kolaitis, Chuan Lei, Fatma Özcan, and Lucian Popa. 2021. Ontology-enriched query answering on relational databases. Proceedings of the AAAI Conference on Artificial Intelligence, 35(17):15247– 15254. find
  2. Sam Buchanan, Druv Pai, Peng Wang, and Yi Ma. 2025. Principles and Practice of Deep Representation Learning. Online. https://ma-lab-berkeley.github.io/ deep-representation-learning-book/. ↗
  3. Sarthak Dash, Gaetano Rossiello, Nandana Mihindukulasooriya, Sugato Bagchi, and Alfio Gliozzo. 2020. Joint entity and relation canonicalization in open knowledge graphs using variational autoencoders. CoRR, abs/2012.04780. — [arXiv:2012.04780](https://arxiv.org/abs/2012.04780) find
  4. Lauren Nicole DeLong, Ramon Fernández Mir, and Jacques D. Fleuriot. 2025. Neurosymbolic ai for reasoning over knowledge graphs: A survey. IEEE Transactions on Neural Networks and Learning Systems, 36(5):7822–7842. find
  5. Nan Du, Kai Chen, Anjuli Kannan, Linh Tran, Yuhui Chen, and Izhak Shafran. 2019. Extracting symptoms and their status from clinical conversations. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 915– 925, Florence, Italy. Association for Computationa… find
  6. Steele Farnsworth, Gabrielle Gurdin, Jorge Vargas, Andriy Mulyar, Nastassja Lewinski, and Bridget T. McInnes. 2022. Extracting experimental parameter entities from scientific articles. Journal of Biomedical Informatics, 126:103970. find
  7. Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich Küttler, Mike Lewis, Wen tau Yih, Tim Rocktäschel, Sebastian Riedel, and Douwe Kiela. 2021. Retrieval-augmented generation for knowledge-intensive nlp tasks. Preprint, arXiv:2005.11401.
  8. Divya Mahajan, Jeffrey J. Liang, Chun-Hao Tsou, and Özlem Uzuner. 2023. Overview of the 2022 n2c2 shared task on contextualized medication event extraction in clinical notes. Journal of Biomedical Informatics, 144:104432. find
  9. Yury A. Malkov and Dmitry A. Yashunin. 2016. Efficient and robust approximate nearest neighbor search using hierarchical navigable small world graphs. CoRR, abs/1603.09320.
  10. Mistral AI. 2025. Mistral-small-3.2-24b-instruct-2506. https://huggingface.co/mistralai/ Mistral-Small-3.2-24B-Instruct-2506. Large language model. ↗
  11. Magdalena Ortiz. 2013. Ontology based query answering: The story so far. In Alberto Mendelzon Workshop on Foundations of Data Management. find
  12. Xiaohan Feng, Xixin Wu, and Helen Meng. 2024. Ontology-grounded automatic knowledge graph construction by llm under wikidata schema. Preprint, arXiv:2412.20942. — [arXiv:2412.20942](https://arxiv.org/abs/2412.20942) find
  13. Yang Peng and Daisy Zhe Wang. 2023. Query-driven knowledge base completion using multimodal path fusion over multimodal knowledge graph. Preprint, arXiv:2212.01923. — [arXiv:2212.01923](https://arxiv.org/abs/2212.01923) find
  14. Aidan Hogan, Eva Blomqvist, Michael Cochez, Claudia d’Amato, Gerard de Melo, Claudio Gutierrez, Sabrina Kirrane, José Emilio Labra Gayo, Roberto Navigli, Sebastian Neumaier, Axel-Cyrille Ngonga Ngomo, Axel Polleres, Sabbir M. Rashid, Anisa Rula, Lukas Schmelzeisen, Juan F. Sequeda, Steffen Staab, an… — Knowledge Graphs
  15. Nils Reimers and Iryna Gurevych. 2019. Sentence-bert: Sentence embeddings using siamese bert-networks. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing. Association for Computational Linguistics.
  16. Yuxin Hou, Eleonora Vig, Michael Donoser, and Loris Bazzani. 2021. Learning attribute-driven disentangled representations for interactive fashion retrieval. In Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), pages 12147– 12157. find
  17. David A. Huffman. 1952. A method for the construction of minimum-redundancy codes. Proceedings of the IRE, 40(9):1098–1101. find
  18. Herve Jégou, Matthijs Douze, and Cordelia Schmid. 2011. Product quantization for nearest neighbor search. IEEE Transactions on Pattern Analysis and Machine Intelligence, 33(1):117–128. find
  19. Vladimir Karpukhin, Barlas Oguz, Sewon Min, Patrick Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, and Wen-tau Yih. 2020. Dense passage retrieval for open-domain question answering. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 6769–6781, Onli…
  20. Indradyumna Roy, Venkata Sai Baba Reddy Velugoti, Soumen Chakrabarti, and Abir De. 2022. Interpretable neural subgraph matching for graph retrieval. Proceedings of the AAAI Conference on Artificial Intelligence, 36(7):8115–8123. find
  21. G. Sanders and S. Shin. 2001. Denormalization effects on performance of rdbms. In Proceedings of the 34th Annual Hawaii International Conference on System Sciences ( HICSS-34)-Volume 3 - Volume 3, HICSS ’01, page 3013, USA. IEEE Computer Society. find
  22. Jeffrey Sardina, John D. Kelleher, and Declan O’Sullivan. 2024. A survey on knowledge graph structure and knowledge graph embeddings. Preprint, arXiv:2412.10092. — [arXiv:2412.10092](https://arxiv.org/abs/2412.10092) find
  23. Claude E. Shannon. 1959. Coding theorems for a discrete source with a fidelity criterion. In IRE National Convention Record, volume 7, pages 142–163. find
  24. Juhee Son, Yeon Seonwoo, Seunghyun Yoon, James Thorne, and Alice Oh. 2024. Multi-hop database reasoning with virtual knowledge graph. In Proceedings of the 1st Workshop on Knowledge Graphs and Large Language Models (KaLLM 2024), pages 1–11, Bangkok, Thailand. Association for Computational Linguistic… find
  25. Qwen Team. 2024. Qwen2.5: A party of foundation models. find
  26. James Thorne, Majid Yazdani, Marzieh Saeidi, Fabrizio Silvestri, Sebastian Riedel, and Alon Halevy. 2021. Database reasoning over text. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processin… find
  27. Anna Vikström, Mikael Nyström, Hans Ahlfeldt, Lars-Erik Strender, and Gunnar H. Nilsson. 2010. Views of diagnosis distribution in primary care in 2.5 million encounters in stockholm: a comparison between icd-10 and snomed ct. Informatics in Primary Care, 18(1):17–29. find
  28. Mengtian Yin, Llewellyn Tang, Chris Webster, Shen Xu, Xiongyi Li, and Huaquan Ying. 2023. An ontology-aided, natural language-based approach for multi-constraint bim model querying. Journal of Building Engineering, 76:107066. find
  29. Yaodong Yu, Kwan Ho Ryan Chan, Chong You, Chaobing Song, and Yi Ma. 2020. Learning diverse and discriminative representations via the principle of maximal coding rate reduction. In Advances in Neural Information Processing Systems, volume 33, pages 9422–9434. Curran Associates, Inc. find
  30. Bowen Zhang and Harold Soh. 2024. Extract, define, canonicalize: An llm-based framework for knowledge graph construction. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 9820–9836. Association for Computational Linguistics.
  31. Jing Zhang, Bo Chen, Lingxi Zhang, Xirui Ke, and Haipeng Ding. 2021. Neural, symbolic and neural-symbolic reasoning on knowledge graphs. AI Open, 2:14–35. find
  32. Shunyu Zhang, Yaobo Liang, Ming Gong, Daxin Jiang, and Nan Duan. 2022. Multi-view document representation learning for open-domain dense retrieval. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 5990–6000, Dublin, Ireland. As… find
  33. Fengbin Zhu, Wenqiang Lei, Chao Wang, Jianming Zheng, Soujanya Poria, and Tat-Seng Chua. 2021. Retrieving and reading: A comprehensive survey on open-domain question answering. Preprint, arXiv:2101.00774. — [arXiv:2101.00774](https://arxiv.org/abs/2101.00774)

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/pralomram-2026-thesis-proposal-granularity-robustness-trade/