- 學位論文
基於領域知識圖譜建構知識問答—以W公司軋鋼刮傷專案為例
Constructing Knowledge Question Answering Based on Domain Knowledge Graph – A Case Study of W Company’s Steel Rolling Scratch Project
摘要
為提高鹽水不銹鋼廠營運效益，公司於 2020年推動核心能力中心專案，由使用單 位自行提出題目，各項選題均是長年來公司面臨的技術瓶頸，也都是高層極為重視之 事項，雖然目標清楚，但解題困難度高，需投入大量的技術人力與資源，在短時間要 獲得大突破是不容易的事情，因此核心能力中心專案在時程規劃上通常需要一年以上 的時間，如果中途參與專案之同仁因其他外在因素離開專案，可能造成專業領域知識 流失、或是如何讓新參與專案員工能快速上手 本文主要參與的題 目為 熱軋 刮傷預測 預防能力 ，由於 線材刮傷的改善難度高，同業競爭對手大部分不具備此技術，建立此 技術能力可大幅提升高 熱軋生產效能與確保交期， 本文 研究主要著重在如何系統化、 科學化將 熱軋刮傷 預測 有關知識儲存起來 因此本文提出了一種領域知識圖譜的構建 方案，以及在此圖譜的基礎上對話框問答系統的實現方案。主要貢獻有 四項 ：：(1)提出 公司廠區領域知識圖譜的構建方案、總體框架 ；；(2) 通過建立領域知識圖譜的本體，詳 細描述了圖譜中各實體的屬性以及實體間的關係 ；；(3)通過編寫網絡爬蟲，從開放網站 抓取關於廠區遇到問題的相關訊息；通過實體對齊解決異構 資料源 的所導致的數據重 疊、歧義；通過知識抽取得到 RDF三元組。最後將所得 RDF數據存儲在圖形資料庫 中並完成領域知識圖譜的構建 ；；(4)在構建完成的領域知識圖譜的基礎上，通過自然語 言處理、 對話框 技術構建 對話框 問答系統。基於自然語言處理技術實現問題與圖譜查 詢語言間的轉換，基於 對話框 技術實現圖譜查詢結果集的圖形化展示。
並列摘要
In order to improve the operational efficiency of the Yanshui stainless steel plant, the company will promote the core competence center project in 2020, and the user will propose the topic by itself. Each topic selection is a technical bottleneck faced by the company for many years, and it is also a matter that the top management attaches great importance to. Although the target Clear, but it is difficult to solve the problem, and requires a lot of technical manpower and resources. It is not easy to achieve a big breakthrough in a short period of time. Therefore, the core competence center project usually takes more than one year in time schedule planning. Colleagues participating in the project leave the project due to other external factors, which may lead to the loss of knowledge in the professional field, or how to make new employees who participate in the project get started quickly; the main topic of this article is the ability to predict and prevent hot rolling scratches. It is difficult to improve. Most of the competitors in the same industry do not have this technology. The establishment of this technical capability can greatly improve the production efficiency of high hot rolling and ensure the delivery time. The research in this paper mainly focuses on how to systematically and scientifically predict the knowledge of hot rolling scratches. Therefore, this paper proposes a construction scheme of a domain knowledge graph, and an implementation scheme of a dialog question answering system based on this graph. There are four main contributions: (1) Propose the construction plan and overall framework of the company's factory domain knowledge graph; (2) By establishing the ontology of the domain knowledge graph, the attributes of each entity in the graph and the relationship between entities are described in detail; (3) ) By writing web crawlers, we can grab relevant information about problems encountered in the factory area from open websites; solve data overlap and ambiguity caused by heterogeneous data sources through entity alignment; and obtain RDF triples through knowledge extraction. Finally, the obtained RDF data is stored in the graph database and the construction of the domain knowledge graph is completed; (4) On the basis of the completed domain knowledge graph, a dialog question answering system is constructed through natural language processing and dialog technology. The conversion between questions and graph query language is realized based on natural language processing technology, and the graphical display of graph query result sets is realized based on dialog box technology.