社交網絡網站（SNS）的普及導致了大量與醫療有關的用戶生成內容（UGC）的產生，這些內容可以用於識別不良藥物反應（ADR）。然而，過去關於自動ADR檢測的研究大多需要大量標記資料才能達到不錯的辨識結果，這在建立針對 UGC 的 ADR 提取模型被認為是耗時且效率不佳的。
為了解決這個問題，本研究提出運用LightNER [1]，這是一種少樣本的命名實體識別（NER）方法，利用了預訓練語言模型（即BART）的能力，並引入提示參數來實現少樣本學習。我們的方法首先利用足夠大量的資料集讓 NER 學習知識，來讓提示參數進行學習。接著，該方法僅需要少量標記的 UGC 來微調參數，就可以獲得不錯的 ADR 提取結果。
在實驗中，我們收集了一個名為Migraine Reviews的偏頭痛藥物的評論資料集，並採用了三個足夠大量的資料集，包括一個通用資料集和兩個臨床筆記資料集。實驗結果表明，帶有提示參數的BART有助於在不同領域的之間轉移知識，從而減少了對大量標記 ADR 資料的需求。我們的方法在有限數量的標計資料（例如，5個樣本）上表現出競爭力。
此外，本研究將識別出的實體與生物醫學詞彙標識符相關聯，實現了病友在社群媒體上所使用的 ADR 詞彙與專業醫學詞彙的映射。這個任務是使用大規模語言模型 GPT-3.5-turbo 已經具備的知識能力來進行，實驗證明該方法提升了進行醫學概念規範化的實用性和效率。

The growing popularity of social network sites (SNS) contributes to the abundance of healthcare-related user-generated content (UGC), which can be used for identifying adverse drug reactions (ADR). Most previous works on automatic ADR detection, however, require a large amount of labeled data to reach acceptable recognition performance, deemed labor-intensive and inefficient when building an ADR detection model for UGC.
To address this issue, this study proposes to apply LightNER [1], a few-shot Named Entity Recognition (NER) approach that exploits the power of a pre-trained language model (i.e., BART) and introduces a small set of prompt parameters to enable few-shot learning. Our approach first warms up the prompt parameters for learning knowledge for NER from rich-resource datasets. Then, the approach only requires a few-shot Named Entity Recognition (NER) approach that exploits the power of a pre-trained language model (i.e., BART) and introduces a small set of prompt parameters to enable few-shot learning. Our approach first warms up the prompt parameters for learning knowledge for NER from rich-resource datasets. Then, the approach only requires a few labeled UGC to fine-tune the parameters to achieve satisfactory ADR recognition performance.
In our experiment, we collect a review dataset for a migraine drug called MigraineReviews and adopt three rich-resource datasets, including one general-purpose dataset and two datasets for clinical notes. The experimental results demonstrate that BART with prompt parameters helps transfer knowledge across datasets even if they are of different domains, thus reducing the need for a large amount of labeled ADR data. Our approach performs competitively with a limited size of labeled data (e.g., 5-shots).
Additionally, the study associates recognized entities with biomedical vocabulary identifiers, enabling the mapping of laypeople's ADR terms to professional medical concepts. The use of GPT-3.5-turbo, a large-scale language model, enhances the practicality and efficiency of the model for medical concept normalization in social media posts.

論文審定書 i
誌謝 ii
摘要 iii
Abstract iv
Table of Contents vi
List OF Figures ix
List OF Table ix
1. Introduction 1
2. DATASET 6
2.1 CoNLL2003 7
2.2 BC5CDR 8
2.3 N2C2 8
2.4 TwADR-L 9
2.5 SMM4H 2017 - subtask3 9
3. METHODOLOGY 10
3.1 NER Task 10
3.1.1 LightNER 11
3.1.2 Encoder 11
3.1.3 Decoder 12
3.1.4 Guidance Module 14
3.1.5 Transfer Strategy 16
3.2 Concept Normalization 17
3.2.1 Definition 19
3.2.2 Select Top k 19
3.2.3. GPT-3.5-turbo Reranking 20
4. NER Experiments 21
4.1 Experimental Settings 21
4.1.1 Model Settings 21
4.1.2 Evaluation Approach 21
4.1.3 Compared Methods 22
4.1.4 Metrics 22
4.2 Results 22
4.3 Error Analysis 25
4.4. Result in use ChatGPT 27
5. Concept Normalization Experiments 30
5.1 Preliminary Study 30
5.2 Re-rank using ChatGPT 32
5.3 Evaluation 34
5.4 Results 36
6. RELATED WORKS 40
7. CONCLUSION 42
8. REFERENCES 44

[1] X. Chen, L. Li, S. Deng, C. Tan, C. Xu, F. Huang, L. Si, H. Chen, and N. Zhang, “LightNER: A Lightweight Tuning Paradigm for Low-resource NER via Pluggable Prompting,” in Proceedings of the 29th International Conference on Computational Linguistics. Gyeongju, Republic of Korea: International Committee on Computational Linguistics, Oct. 2022, pp. 2374–2387. [Online]. Available: https://aclanthology.org/2022.coling-1.209
[2] J. J. Coleman and S. K. Pontefract, “Adverse drug reactions,” Clinical Medicine, vol. 16, no. 5, pp. 481–485, Oct. 2016. [Online]. Available: https://www.rcpjournals.org/lookup/doi/10.7861/clinmedicine.16-5-481
[3] D.W.Bates,D.J.Cullen,N.Laird,L.A.Petersen,S.D.Small,D.Servi, G. Laffel, B. J. Sweitzer, B. F. Shea, and R. Hallisey, “Incidence of adverse drug events and potential adverse drug events. Implications for prevention. ADE Prevention Study Group,” JAMA, vol. 274, no. 1, pp. 29–34, Jul. 1995.
[4] D.W.Bates,N.Spell,D.J.Cullen,E.Burdick,N.Laird,L.A.Petersen, S. D. Small, B. J. Sweitzer, and L. L. Leape, “The costs of adverse drug events in hospitalized patients. Adverse Drug Events Prevention Study Group,” JAMA, vol. 277, no. 4, pp. 307–311, Jan. 1997.
[5] T.-Y. Wu, M.-H. Jen, A. Bottle, M. Molokhia, P. Aylin, D. Bell, and A. Majeed, “Ten-year trends in hospital admissions for adverse drug reactions in England 1999–2009,” Journal of the Royal Society of Medicine, vol. 103, no. 6, pp. 239–250, Jun. 2010. [Online]. Available: http://journals.sagepub.com/doi/10.1258/jrsm.2010.100113
[6] L. Zhou, D. Zhang, C. C. Yang, and Y. Wang, “Harnessing social media for health information management,” Electronic Commerce Research and Applications, vol. 27, pp. 139–151, Jan. 2018. [Online]. Available: https://linkinghub.elsevier.com/retrieve/pii/S1567422317300960
[7] C. Xiao, P. Zhang, W. Chaovalitwongse, J. Hu, and F. Wang, “Adverse Drug Reaction Prediction with Symbolic Latent Dirichlet Allocation,” Proceedings of the AAAI Conference on Artificial Intelligence, vol. 31, no. 1, Feb. 2017. [Online]. Available: https://ojs.aaai.org/index.php/AAAI/article/view/10717
[8] K. Lee, A. Qadir, S. A. Hasan, V. Datla, A. Prakash, J. Liu, and O. Farri, “Adverse Drug Event Detection in Tweets with Semi-Supervised Convolutional Neural Networks,” in Proceedings of the 26th International Conference on World Wide Web. Perth Australia: International World Wide Web Conferences Steering Committee, Apr. 2017, pp. 705–714. [Online]. Available: https://dl.acm.org/doi/10.1145/3038912.3052671
[9] I. Korkontzelos, A. Nikfarjam, M. Shardlow, A. Sarker, S. Ananiadou, and G. H. Gonzalez, “Analysis of the effect of sentiment analysis on extracting adverse drug reactions from tweets and forum posts,” Journal of Biomedical Informatics, vol. 62, pp. 148–158, Aug. 2016. [Online]. Available: https://linkinghub.elsevier.com/retrieve/pii/S1532046416300508
[10] L. Wu, T.-S. Moh, and N. Khuri, “Twitter opinion mining for adverse drug reactions,” in 2015 IEEE International Conference on Big Data (Big Data). Santa Clara, CA, USA: IEEE, Oct. 2015, pp. 1570–1574. [Online]. Available: http://ieeexplore.ieee.org/document/7363922/
[11] S. Wang, Y. Li, D. Ferguson, and C. Zhai, “SideEffectPTM: an unsupervised topic model to mine adverse drug reactions from health forums,” in Proceedings of the 5th ACM Conference on Bioinformatics, Computational Biology, and Health Informatics. Newport Beach California: ACM, Sep. 2014, pp. 321–330. [Online]. Available:
https://dl.acm.org/doi/10.1145/2649387.2649398
[12] H. Yang and C. C. Yang, “Discovering Drug-Drug Interactions and Associated Adverse Drug Reactions with Triad Prediction in Hetero- geneous Healthcare Networks,” in 2016 IEEE International Conference on Healthcare Informatics (ICHI), Oct. 2016, pp. 244–254.
[13] T. Nguyen, M. E. Larsen, B. O’Dea, D. Phung, S. Venkatesh, and H. Christensen, “Estimation of the prevalence of adverse drug reactions from social media,” International Journal of Medical Informatics, vol. 102, pp. 130–137, Jun. 2017. [Online]. Available: https://linkinghub.elsevier.com/retrieve/pii/S1386505617300746
[14] C.-H. Chang, L. Wang, and C. C. Yang, “Constructing Cross-lingual Consumer Health Vocabulary with Word-Embedding from Comparable User Generated Content,” Jun. 2022, iSBN: 2206.11612 Publication Title: arXiv [cs.CL]. [Online]. Available: http://arxiv.org/abs/2206.11612
[15] A. P Tafti, J. Badger, E. LaRose, E. Shirzadi, A. Mahnke, J. Mayer, Z. Ye, D. Page, and P. Peissig, “Adverse Drug Event Discovery Using Biomedical Literature: A Big Data Neural Network Adventure,” JMIR Medical Informatics, vol. 5, no. 4, p. e51, Dec. 2017. [Online]. Available: http://medinform.jmir.org/2017/4/e51/
[16] H. Sampathkumar, X.-w. Chen, and B. Luo, “Mining adverse drug reactions from online healthcare forums using hidden Markov model,” BMC medical informatics and decision making, vol. 14, p. 91, Oct. 2014.
[17] F. Christopoulou, T. T. Tran, S. K. Sahu, M. Miwa, and S. Ananiadou, “Adverse drug events and medication relation extraction in electronic health records with ensemble deep learning methods,” Journal of the American Medical Informatics Association, vol. 27, no. 1, pp. 39–46, Jan. 2020. [Online]. Available: https://academic.oup.com/jamia/article/27/1/39/5544735
[18] X. Yang, J. Bian, R. Fang, R. I. Bjarnadottir, W. R. Hogan, and Y. Wu, “Identifying relations of medications with adverse drug events using recurrent convolutional neural networks and gradient boosting,” Journal of the American Medical Informatics Association, vol. 27, no. 1, pp. 65–72, Jan. 2020. [Online]. Available: https://academic.oup.com/jamia/article/27/1/65/5555856
[19] E.-d. El-allaly, M. Sarrouti, N. En-Nahnahi, and S. Ouatik El Alaoui, “MTTLADE: A multi-task transfer learning-based method for adverse drug events extraction,” Information Processing & Management, vol. 58, no. 3, p. 102473, May 2021. [Online]. Available: https://linkinghub.elsevier.com/retrieve/pii/S0306457320309626
[20] X. L. Li and P. Liang, “Prefix-Tuning: Optimizing Continuous Prompts for Generation,” in Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). Online:AssociationforComputationalLinguistics,Aug.2021, pp. 4582–4597. [Online]. Available: https://aclanthology.org/2021.acl- long.353
[21] B. Lester, R. Al-Rfou, and N. Constant, “The Power of Scale for Parameter-Efficient Prompt Tuning.” Online and Punta Cana, Dominican Republic: Association for Computational Linguistics, Nov. 2021, pp. 3045–3059. [Online]. Available: https://aclanthology.org/2021.emnlp-main.243
[22] M. Lewis, Y. Liu, N. Goyal, M. Ghazvininejad, A. Mohamed, O. Levy, V. Stoyanov, and L. Zettlemoyer, “BART: Denoising Sequence-to- Sequence Pre-training for Natural Language Generation, Translation, and Comprehension,” in Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics. Online: Association for Computational Linguistics, Jul. 2020, pp. 7871–7880. [Online]. Available: https://aclanthology.org/2020.acl-main.703
[23] M. Neumann, D. King, I. Beltagy, and W. Ammar, “ScispaCy: Fast and Robust Models for Biomedical Natural Language Processing,” in Proceedings of the 18th BioNLP Workshop and Shared Task. Florence, Italy: Association for Computational Linguistics, Aug. 2019, pp. 319–327. [Online]. Available: https://aclanthology.org/W19-5034
[24] E. F. Tjong Kim Sang and F. De Meulder, “Introduction to the CoNLL- 2003 Shared Task: Language-Independent Named Entity Recognition,” in Proceedings of the Seventh Conference on Natural Language Learning at HLT-NAACL 2003, 2003, pp. 142–147. [Online]. Available: https://aclanthology.org/W03-0419
[25] J. Li, Y. Sun, R. J. Johnson, D. Sciaky, C.-H. Wei, R. Leaman, A. P. Davis, C. J. Mattingly, T. C. Wiegers, and Z. Lu, “BioCreative V CDR task corpus: a resource for chemical disease relation extraction,” Database, vol. 2016, p. baw068,2016. [Online]. Available: https://academic.oup.com/database/article-lookup/doi/10.1093/database/baw068
[26] S. Henry, K. Buchan, M. Filannino, A. Stubbs, and O. Uzuner, “2018 n2c2 shared task on adverse drug events and medication extraction in electronic health records,” Journal of the American Medical Informatics Association, vol. 27, no. 1, pp. 3–12, Jan. 2020. [Online]. Available: https://academic.oup.com/jamia/article/27/1/3/5581277
[27] A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. Kaiser, and I. Polosukhin, “Attention is
All you Need,” in Advances in Neural Information Processing Systems, vol. 30. Curran Associates, Inc., 2017. [Online]. Available: https://papers.nips.cc/paper/2017/hash/3f5ee243547dee91fbd053c1c4a845aa- Abstract.html
[28] H. Yan, T. Gui, J. Dai, Q. Guo, Z. Zhang, and X. Qiu, “A Unified Generative Framework for Various NER Subtasks,” in Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1: Long Papers). Online: Association for Computational Linguistics, Aug. 2021, pp. 5808–5822. [Online]. Available: https://aclanthology.org/2021.acl-long.451
[29] P. Liu, W. Yuan, J. Fu, Z. Jiang, H. Hayashi, and G. Neubig, “Pre-train, Prompt, and Predict: A Systematic Survey of Prompting Methods in Natural Language Processing,” Jul. 2021, iSBN: 2107.13586 Publication Title: arXiv [cs.CL]. [Online]. Available: http://arxiv.org/abs/2107.13586
[30] J.-Y. Huang, W.-P. Lee, and K.-D. Lee, “Predicting Adverse Drug Reactions from Social Media Posts: Data Balance, Feature Selection and Deep Learning,” Healthcare, vol. 10, no. 4, p. 618, Mar. 2022. [Online]. Available: https://www.mdpi.com/2227-9032/10/4/618
[31] Q. Wei, Z. Ji, Z. Li, J. Du, J. Wang, J. Xu, Y. Xiang, F. Tiryaki, S. Wu, Y. Zhang, C. Tao, and H. Xu, “A study of deep learning approaches for medication and adverse drug event extraction from clinical text,” Journal of the American Medical Informatics Association, vol. 27, no. 1, pp. 13–21, Jan. 2020. [Online]. Available: https://academic.oup.com/jamia/article/27/1/13/5499225
[32] Fung, K. W., & Bodenreider, O. (2005). Utilizing the UMLS for semantic mapping between terminologies. AMIA ... Annual Symposium proceedings. AMIA Symposium, 2005, 266–270.
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC1560893/
[33] Kalyan KS, Sangeetha S. BertMCN: Mapping colloquial phrases to standard medical concepts using BERT and highway network. Artif Intell Med. 2021 Feb;112:102008. doi: 10.1016/j.artmed.2021.102008. Epub 2021 Jan 7. PMID: 33581833.
https://pubmed.ncbi.nlm.nih.gov/33581833/
[34] Brennan PF, Aronson AR. Towards linking patients and clinical information: detecting UMLS concepts in e-mail. J Biomed Inform. 2003 Aug-Oct;36(4-5):334-41. doi: 10.1016/j.jbi.2003.09.017. PMID: 14643729.
https://pubmed.ncbi.nlm.nih.gov/14643729/
[35] Keselman, A., Smith, C. A., Divita, G., Kim, H., Browne, A. C., Leroy, G., & Zeng-Treitler, Q. (2008). Consumer health concepts that do not map to the UMLS: where do they fit?. Journal of the American Medical Informatics Association : JAMIA, 15(4), 496–505. https://doi.org/10.1197/jamia.M2599
[36] Tutubalina E, Miftahutdinov Z, Nikolenko S, Malykh V. Medical concept normalization in social media posts with recurrent neural networks. Journal of Biomedical Informatics. 2018 Aug;84:93-102. DOI: 10.1016/j.jbi.2018.06.006. PMID: 29906585.
https://europepmc.org/article/med/29906585
[37] Kung, Tiffany H., et al. "Performance of ChatGPT on USMLE: Potential for AI-assisted medical education using large language models." PLoS digital health 2.2 (2023): e0000198.
https://journals.plos.org/digitalhealth/article?id=10.1371/journal.pdig.0000198&fbclid=IwAR3U9R0A5QwAVMACFfgA79EFYWu32uFE8upittW5ZEb9qaNSZyWXpxdnJU4
[38] Khan, Rehan Ahmed, et al. "ChatGPT-Reshaping medical education and clinical management." Pakistan Journal of Medical Sciences 39.2 (2023): 605.
https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10025693/
[39] Wei, Xiang, et al. "Zero-shot information extraction via chatting with chatgpt." arXiv preprint arXiv:2302.10205 (2023).
https://arxiv.org/abs/2302.10205
[40] Abeed Sarker and others, Data and systems for medication-related text classification and concept normalization from Twitter: insights from the Social Media Mining for Health (SMM4H)-2017 shared task, Journal of the American Medical Informatics Association, Volume 25, Issue 10, October 2018, Pages 1274–1283, https://doi.org/10.1093/jamia/ocy114
[41] Belousov M , Dixon W, Nenadic G. Using an ensemble of linear and deep learning models in the SMM4H 2017 medical concept normalization task. In: Proceedings of the Second Workshop on Social Media Mining for Health Research and Applications Workshop Co-located with the American Medical Informatics Association Annual Symposium (AMIA 2017); 2017: 54–58. http://ceur-ws.org/Vol-1996/paper10.pdf. Accessed May 8, 2018.
[42] Han S , Tran T, Rios A, Kavuluru R. Team UKNLP: detecting ADR Mentions on Twitter. In: Proceedings of the Second Workshop on Social Media Mining for Health Research and Applications Workshop Co-located with the American Medical Informatics Association Annual Symposium (AMIA 2017); 2017: 49–53. http://ceur-ws.org/Vol-1996/paper9.pdf. Accessed April 30, 2018.
[43] Chia-Hsuan Chang, Fang-Yu Chang, San-Yih Hwang, Christopher C. Yang “Prompting for Few-shot Adverse Drug Reaction Recognition from Online Reviews” 2023 IEEE 11th International Conference on Healthcare Informatics (ICHI), Houston, Texas, USA, 2023.