RecentnessRanker and data_meta_field ... a fuller example?

[demongolem-biz2]

I am going through each of the 4 rankers that do not require an API, and I am currently working on a test case for what the RecentnessRanker brings me and how to use it in practice for semantic search application of Haystack.

I have a FAISSDocumentStore where my documents go, but I don't have this data_meta_field in my documents to being with. And when I run the query pipeline, Haystack reminds me that I am missing this key, whatever value I have set it to. So first, how do I initialize these dates (to what value and how with the Document objects in the FAISSDocumentStore if they are introduced via an indexing pipeline)? Then secondly, how would this data_meta_field be updated in practice as I suppose users query which should trigger a new value in this data_meta_field?

I just feel like so much of the story is not told about what is going on here and there seems there should be a tutorial for it unless if it is going away in the near future?

[demongolem-biz2]

I think I had the wrong idea of this ranker. What I gather now is this date is the creation date of the document, which will endure forever and therefore does not need updating.

Given that I am doing semantic search, what I was thinking of is a ranker based upon how often the data has been accessed in the past. So documents which had been retrieved "recently" would gain a little boost to those that have not been retrieved "recently". This would be the how popular has a document been in past queries, is it a document which people find particularly useful.

[julian-risch]

@demongolem-biz2 You're right. The date in the document is usually the creation date of the document, not the last time it was accessed.