Is it possible to sample only a subset of works when using oa_snowball ?

[adrientaudiere]

Hi all,
Thanks for your very usefull package.

I wonder if it is possible to add an equivalent of options = list(sample = 10, seed = 1)) to each snowball query ? For the moment if I add options = list(sample = 10, seed = 1), it only apply to the first oa_fetch query. And the option citing_params and cited_by_params add a filter in the api request, so the api parameter sample is not usable.

Best,

[trangdata]

Hi @adrientaudiere could you elaborate on the motivation to perform such an analysis? Why would you want a subset of the cited/citing work?

[adrientaudiere]

Hi @trangdata,
I run some "big" analysis, and sometimes I only want to perform the analysis on a small subset of random work to test the entire pipeline with a few works. I know that I can filter the dataset after snowball, but when I saw the option sample in oa_fetch, I thank that it would be useful to be able to add sample in oa_snowball. It's only a small feature request, I understand that it is not very important (-;.

[yjunechoe]

The idea is not completely out of scope, but I'd also like to hear more before considering implementing this (though it seems like sampling, specifically, is not the point?)

If the point is to test a pipeline with a small oa_snowball() output, I'd probably prefer using the to/from_publication_date filter in the citing/cited_by params - this is how we also write our own tests for oa_snowball()! That will also give you a stable subset to work with (unless you're specifically looking to test against output variability?).

library(openalexR)

x <- oa_fetch(doi = "10.1038/s41586-022-05258-z")

# >200 connections
x$cited_by_count + length(x$referenced_works[[1]])
#> [1] 221

# Search the range of 1 year forward, 10 years back
pub_date <- as.Date(x$publication_date)
snowball <- oa_snowball(
  identifier = x$id,
  citing_params = list(to_publication_date = pub_date + 365),
  cited_by_params = list(from_publication_date = pub_date - 3650)
)

# Returns a subset
nrow(snowball$nodes)
#> [1] 112

hist(as.Date(snowball$nodes$publication_date), "years", freq = TRUE)
abline(v = pub_date, col = "red", lwd = 3)

[adrientaudiere]

Oh yes, it's a perfect trick. Maybe it deserves to be present in the example of the function. Thank you.

[yjunechoe]

Oh yes, it's a perfect trick. Maybe it deserves to be present in the example of the function. Thank you.

We had actually planned for a full snowballing vignette but haven't got around to it. We'll keep this example in mind!

[yjunechoe]

Update: no longer necessary to do date conversion of publication_date in dev

library(openalexR)

x <- oa_fetch(doi = "10.1038/s41586-022-05258-z")

# >200 connections
x$cited_by_count + length(x$referenced_works[[1]])
#> [1] 223

# Search the range of 1 year forward, 10 years back
snowball <- oa_snowball(
  identifier = x$id,
  citing_params = list(to_publication_date = x$publication_date + 365),
  cited_by_params = list(from_publication_date = x$publication_date - 3650)
)

# Returns a subset
nrow(snowball$nodes)
#> [1] 114

hist(snowball$nodes$publication_date, "years", freq = TRUE)
abline(v = x$publication_date, col = "red", lwd = 3)

Closing to track vignette discussion in #284