數位世代習慣於從網路獲取資訊，然而網路內容往往相當龐雜且未必正確，如何將非結構化的文字內容擷取表示成知識是一大挑戰難題，尤其是涉及領域特有知識的醫療健康照護領域。目前尚未有中文健康照護知識庫及其應用，仍有三個待解決的問題。第一、探索領域專有詞彙的辨識方法；第二、缺乏領域詞彙彼此間的關係擷取方式；第三、基於知識表達與推論的智慧醫療應用尚不多見。有鑑於此，本計劃擬以三年時間建置一個中文醫療健康照護知識庫以及基於知識庫之問答系統。第一年提出整合深度類神經網路和序列標記模型的方法，用以辨識領域命名實體；第二年提出基於深度類神經網路的個體關係擷取模型，將第一年辨識的領域詞彙作為候選個體，探索領域詞彙間彼此的依存關係，整合成一個中文醫療健康照護領域知識庫；第三年先將第二年產出的知識庫，連結到BabelNet擴充成多國語言詞彙語意網路，然後視覺化成知識圖譜，接著在這個知識圖譜上做邏輯推論，用以開發一個基於知識庫之醫療健康照護問答系統。 ;The digital generation is used to accessing the information from the Web, however, it usually contains a large amount of content that may possibly incorrect. Therefore, how to extract and organize the unstructured text content into the represented knowledge is a very challenging difficulty, especially the medical healthcare domain always involves domain-specific knowledge. So far no Chinese knowledge base for medical healthcare domain exists. There are still three issues being addressed. First, it is worth exploring the methods to identify domain-specific named entities. Second, no research focuses on domain-specific entity relation extraction. Third, few intelligent medicine applications based on knowledge representation and reasoning.A three-year project that focuses on Chinese knowledge base construction for medical healthcare domain and its application of question answering are proposed to address the above three issues. In the first year, we integrate deep neural networks with sequential labeling models to recognize domain-specific named entities. In the second year, we propose a deep neural network model to extract relations based on the recognized entities in the first year, and then integrate all extracted entity-relationships into a knowledge base. In the third year, we link our constructed knowledge base to BabelNet for enriching as a multilingual lexical-semantic network, visualize it as knowledge graphs, and build a question answering system for medical healthcare domain through reasoning on our Chinese knowledge base and graphs.
