引導式測驗 · 中等 等級
重新排序模型測驗
測試檢索器和重新排序器的角色,比較交叉編碼器與雙編碼器,並檢查候選集的清單排序。
問題 1 的 8
在典型的兩階段檢索管道中,重新排序器的工作是什麼?
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。
- 在典型的兩階段檢索管道中,重新排序器的工作是什麼?
- 候選人候選名單的相關性重新評分和重新排序
- 從數百萬份文件中提取最初的候選者
- 將文件儲存在資料庫中
- 產生最終的自然語言答案
- Besides MS MARCO, on which dataset did Nogueira and Cho report strong results for their BERT passage reranker?
- 影像網
- TREC-CAR
- 圖書館演講
- SQuAD only
- 為什麼不能使用跨編碼器重排序器直接搜尋數百萬個文件?
- 它僅適用於圖像
- 它無法讀取文本
- 必須為每個查詢文檔對重新計算其分數,這在規模上太慢了
- 根本不需要計算
- 雙編碼器相對於交叉編碼器的主要優點是什麼?
- 細粒度相關性更準確
- 它總是完美地重新排序結果
- 它不需要訓練數據
- 它的文檔嵌入可以預先計算,使檢索快速且可擴展
- How can a reranker affect context quality in a retrieval-augmented generation pipeline?
- It can move more relevant retrieved candidates higher in the context shortlist, but does not guarantee correctness
- It ensures every retrieved passage is factually correct
- It creates missing documents in the corpus
- It removes the need to evaluate retrieval
- In listwise reranking, what does the model consider together?
- Only one document with no query
- A query and a set of candidate documents when deciding their ordering
- Each document embedding before any candidate retrieval
- A single relevance label for the full training corpus
- What should a team verify before interpreting reranker scores as probabilities?
- Whether every score is above zero
- Whether the first-stage retriever was removed
- Whether the scores have been calibrated and evaluated for that use
- Whether the model has more parameters than the retriever
- What limitation does a reranker inherit from the first-stage retrieval system?
- It always returns a calibrated probability
- 它消除了對檢索的需求
- 它保證產生正確的答案
- 它無法推廣候選集合中未包含的相關文件