-
|
Hi, |
Beta Was this translation helpful? Give feedback.
Replies: 2 comments
-
|
Yes, it is well known that query latency increases as k in top-k retrieval increases. Here's a relatively recent survey that provides lots of details: https://www.nowpublishers.com/article/Details/INR-057 Insight is that top-k docs are kept in the heap during query evaluation - the larger the k, the more "work" the algorithm needs to do. Modern algorithms are fast because they are able to (heuristically) answer this question: can this document possibly be in the top k? If no, I can skip it. |
Beta Was this translation helpful? Give feedback.
-
|
Hi @lintool , |
Beta Was this translation helpful? Give feedback.
Yes, it is well known that query latency increases as k in top-k retrieval increases. Here's a relatively recent survey that provides lots of details: https://www.nowpublishers.com/article/Details/INR-057
Insight is that top-k docs are kept in the heap during query evaluation - the larger the k, the more "work" the algorithm needs to do. Modern algorithms are fast because they are able to (heuristically) answer this question: can this document possibly be in the top k? If no, I can skip it.