## Metadata - Author: [[doug-turnbull|Doug Turnbull]] - Full Title:: What AI Engineers Should Know About Search - Category:: #🗞️Articles - URL:: https://softwaredoug.com/blog/2024/06/25/what-ai-engineers-need-to-know-search?utm_source=substack&utm_medium=email - Finished date:: [[2024-06-30]] ## Highlights > Search comes with a [precision / recall tradeoff](https://link.springer.com/referenceworkentry/10.1007/978-0-387-30164-8_652#:~:text=Precision%20%3D%20Total%20number%20of%20documents,to%20define%20these%20two%20measures.). If you cast a wide net, you’ll get more relevant results in the mix, but also likely be showing the users lots of irrelevant ones too! ([View Highlight](https://read.readwise.io/read/01j1mmpza767z5pg2n39579jz3)) ## New highlights added [[2025-05-07]] > • The fanciest solutions don’t matter as much as getting a good evaluation framework setup to evaluate the quality of search results ([View Highlight](https://read.readwise.io/read/01jtmjv7y5fsh1hk7ngs4533ev)) > A lot of metrics exist for measuring the quality of a query - if query “zoolander” returns some search results, we can reference the judgments, to see whether or not we gave relevant results. Statistics like (n)DCG, ERR, MAP, Precision, Recall, F-Score are well understood in the search industry ([View Highlight](https://read.readwise.io/read/01jtmkaetkdy9hns8fazbvtrws)) > There are tools to compute the "true" specificity of a term beyond direct IDF, such as [blending IDF across fields](https://www.elastic.co/guide/en/elasticsearch/reference/current/query-dsl-multi-match-query.html), or merging all the text into one big field. ([View Highlight](https://read.readwise.io/read/01jtmkccsc3qmm45fyp8w6hkjc))