| 摘要: | 近年來，以3D建築資訊模型為重心基礎的BIM（Building Information Modeling）技術在營建工程產業的應用發展上大幅進步，因而造成越來越多人對於BIM領域開始產生好奇，在網路上、在圖書館內查詢相關文章、資訊來認識BIM，並增加此領域之專業知識。
本研究目的為研究出用於BIM訊息(專業知識、最新消息)和以BIM文件為主要搜尋範圍的智慧型搜尋功能的中文問答系統，並開發手機聊天機器人的應用程式作為問題輸入的使用者介面，供國內應用BIM使用者所需資源。本研究提出以BERT(Bidirectional Encoder Representations from Transformers)語言模型為基礎核心，依據專業使用者及一般使用者兩類不同用戶，將問答文本劃分為專業使用者較常查詢之BIM文件、資訊，和一般使用者欲了解的BIM領域知識或是最新消息，並透過大量問答文本資料進行深度學習訓練，並以伺服器和Android系統之手機進行Socket間的通訊，以達成傳送手機端問題及服務器答案之目的。並且，在不同的問答測試下，將其結果與不同瀏覽器的搜尋結果做比較。最終，本論文可獲得，此系統確實可以正確地給予兩句不同字但同義之句子同樣之回覆，且透過此運用於聊天機器人app的問答系統，的確可以有效地將搜尋結果範圍縮小、搜尋速度加快。然而，在較短的句子中，兩句不同語義但部分文字之位置、字相同的句子，可能會被系統誤認為相同問題，並給予一樣之答案，並造成誤答，未來可能透過向BIM領域專業人士發送問卷的方式，來提升文本的專業性，由修改語句架構的方式，增加句子間的差異性，提升文本準確性，並且增進語言模型自然語言之判斷能力，提升系統之正確性。
;In recent years, BIM (Building Information Modeling) technology based on 3D building information modeling has made great progress in the application and development of the construction engineering industry. This has caused more and more people to become curious about the BIM field. Search related articles and information in the library to learn about BIM and increase professional knowledge in this field.
The purpose of this research is to develop a Chinese question answering system for BIM messages (professional knowledge, latest news) and smart search functions with BIM documents as the main search scope, and to develop a mobile chat robot application as a user interface for question input , For the resources needed by domestic BIM users. This research proposes to use the BERT (Bidirectional Encoder Representations from Transformers) language model as the basic core. According to two different types of users, professional users and general users, the question and answer text is divided into BIM documents, information, and general that professional users frequently query. The user wants to know the BIM domain knowledge or the latest news, and conducts deep learning training through a large amount of question and answer text data, and communicates between the server and the mobile phone of the Android system through Socket to achieve the purpose of sending mobile phone questions and server answers . And, under different Q&A tests, compare the results with the search results of different browsers. In the end, this paper can obtain that this system can indeed give two different words but synonymous sentences with the same reply correctly, and through this question and answer system applied to the chatbot app, it can indeed effectively narrow the search results and search speed. accelerate. However, in a shorter sentence, two sentences with different meanings but with the same position and word in part of the text may be mistaken for the same question by the system, and the same answer will be given, resulting in a wrong answer. In the future, it may be transferred to the BIM field. Professionals send questionnaires to enhance the professionalism of the text. By modifying the sentence structure, the difference between sentences is increased, the accuracy of the text is improved, and the judgment ability of the natural language of the language model is improved, and the correctness of the system is improved. |
