ARTS week 2
Evaluate good search part I - measure it
Chose this article because I'm actively working on improving search experience for our platform.
This article brings general guidelines on what / how to evaluate search quality.
What to measure?
Precision, among the recalled results, relevant results / recalled results
Recall, among all relevant results, recalled results / all relevant results
Indeed these two are very well-acknowledged search metrics by today's industry.
How to measure?
When it comes to the question of
How, the short answer is always (at least 90%)
it depends. We all want perfect solutions but mostly ended up with some feasible solution that is highly adapted to our situation. But this isn't saying there is nothing we could try first.
One good thing to try IMO is :
Measure precision by CTR (click-through-rate), that is, # of clicks / # of views
Do above for only the top ranked items, because only a small amount of users would go through the entire list of search results.
本周遇到一个mysql slow query问题，研究了一下和index merge有关.
Short term fix
使用hint让表B强制使用PRIMARY index :
Long term fix
一般来说出现了intersect index merge，表示index有可以优化的空间，比如建立一个PRIMARY + status的联合索引.