醫院藥局的藥師是作為專業醫師與民眾之間的重要角色，而用藥的衛教也 是醫院藥師專業服務的重要業務。近年來預防醫學的提倡，許多檢查與衛教的 需求量日漸提升，尤其是以大腸鏡為大宗，大腸鏡檢查亦是作為早期檢測大腸 癌的方法，早期發現與治療可以降低死亡率。
根據醫院藥師的統計，對於每一位民眾所作的大腸鏡檢查，需考量每個人 的個別因素，因此需要花費數十分鐘的解釋。如何應用人工智慧技術來提供多 元化的服務、有效率的解決需求，將是對社會非常重要的貢獻。
為了解決使用者的大腸鏡衛生教育提問，本篇論文提供了一種分析 Intent 的方法，藉由抽取問句中的 Q-V-F-M-O-T 結構來分析大腸鏡衛教之相關問題， 且透過 Knowledge Graph 和 Template-based Answers 作為回答的知識庫，並 建立一個 iOSApp 讓使用者可以即時的詢問，並給予回答。
最後，經測試隨機收集的 220 句問句，有 83% 的問題是可以被準確回答。 此外，我們隨機挑選了二十位使用者，進行論文系統的場域測試。於結果顯示， 使用者評分對於系統資訊的充足度平均為 3.8 分，回答的答案可讀性平均為 4.2 分，信任度平均為 4 分，使用體驗平均為 3.85 分，其中滿分為 5 分。

Hospital pharmacists play an important role and education for medication is also an important task. Nowadays, preventive healthcare has been promoted, and the demand for health examinations and patient education have been increasing, especially for colonoscopy which is a method for early detection of colorectal cancer. Early detection and treatment can reduce mortality.
According to the statistics, pharmacist needs to consider each person's individual factors during the patient education of colonoscopy, so it usually takes ten minutes for a person. How to apply artificial intelligence technology to provide services and solve user needs is very important.
In order to solve questions about colonoscopy from users, this paper provides a method to analyze intents by extracting Q-V-F-M-O-T structures in question sentences, using Knowledge Graph and Template-based Answers as a knowledge base for answers and we built an iOS application for user to ask questions and get corresponding answers.
At the last, the system achieved a recall rate about 83% with random 220 questions. Furthermore, we randomly selected twenty users to do an actual testing and the results show that we scored an average of 3.8 points for the sufficiency of information, 4.2 points for the answer readability, 4 points for trust level, and 3.85 points for user experience which the full score is 5 points.

[1] Patient Education Sheets for Colonoscopy from JEN-AI Hospital, Taiwan. Available:
https://health.jah.org.tw/content/index.asp?m=1&m1=4&m2=107&gp=105
[2] P.-Y. Chiu, A Health-Advisory Chatbot Based on Relation Path Searching of Healthcare Knowledge Base, 2020.
[3] J.C. Cai, Summary Generation for Chinese Patient Complaint based on Medical Entity Recognition and Medical Terminology Mapping, 2019.
[4] HanLP. Available: https://www.hanlp.com/
[5]台灣食品成分資料庫 2020 版(UPDATE2), Taiwan Food and Drug Administration, [Online]. Available: https://consumer.fda.gov.tw/Food/TFND.aspx?nodeID=178
[6] 西藥、醫療器材及化粧品許可證查詢, Taiwan Food and Drug Administration, [Online]. Available: https://info.fda.gov.tw/mlms/H0001.aspx
[7] UpToDate, [Online]. Available:
https://www.uptodate.com/contents/perioperative-management-of-patients-receiv ing-anticoagulants/print
[8] ATC Code, Wikipedia, [Online]. Available:
https://en.wikipedia.org/wiki/Anatomical_Therapeutic_Chemical_Classification _System
[9] Neo4j, Neo4j, Inc., [Online]. Available: https://neo4j.com 29
[10] Cypher, Neo4j, Inc., [Online]. Available:
https://neo4j.com/developer/cypher/
[11] 詞法分析之 Bi-LSTM-CRF 框架, [Online]. Available: https://www.twblogs.net/a/5b7c79642b71770a43db24a5
[12] 如何辨別機器學習模型的好壞?秒懂, Confusion Matrix, [Online]. Available: https://www.ycc.idv.tw/confusion-matrix.html