引導式測驗 · 中等 等級
從文字測驗中擷取關係
測試關係提取如何將文本轉化為結構化事實,以及其常見管道和監督方法如何失敗。
問題 1 的 8
關係提取系統的典型輸出是什麼?
預覽每個問題8 問題
當您參加測驗時,就會出現答案和解釋。
- 關係提取系統的典型輸出是什麼?
- 透過關係連結兩個實體的結構化三元組
- 句子的文法分析樹
- 情緒評分從消極到積極
- 將句子翻譯成另一種語言
- 在關係提取的通用管道中,什麼可以在對分類之前識別實體提及?
- 文字轉語音合成
- 命名實體識別
- 影像分割
- Audio source separation
- 為什麼遠端監督會產生吵雜的關係提取標籤?
- Knowledge bases contain no structured facts
- It requires every sentence to be hand-labeled
- A sentence may mention two linked entities without stating their knowledge-base relation
- It removes entity mentions before training
- 許多基於 Transformer 的 RE 模型如何指示要分類的實體對?
- 首先翻譯句子
- 透過刪除實體之一
- 透過將文字轉換為音訊
- 透過在兩個實體周圍插入特殊標記令牌
- 「開放」關係提取與標準 RE 有何不同?
- 它直接從文本中提取關係短語,無需預先定義一組關係類型
- 它只適用於數字
- 它比任何其他方法需要更多的標記數據
- 每個文檔只能找到一個關係
- Why include a no-relation label when classifying candidate entity pairs?
- To translate the sentence into another language
- To represent pairs for which the text does not assert a target relation
- To replace named entity recognition
- To guarantee that every pair has a relation
- What does document-level relation extraction allow a system to use?
- Only the spelling of a single entity
- Audio features from a spoken recording
- Evidence that may be spread across multiple sentences
- A fixed relation without reading text
- How does open relation extraction differ from a closed relation schema?
- It cannot use text as input
- It always requires a knowledge base to label every sentence
- It only identifies entity boundaries
- It can express relation phrases without choosing only from a predefined label inventory