OpenBiodiv: A Knowledge Graph for Literature-Extracted Linked Open Data in Biodiversity Science

Round 1

Reviewer 1 Report

I've created a table of suggestions (uploaded). Most are small text suggestions. There are also a few I have flagged as Minor, worth addressing if the document is revised, and an issue (with references in a couple of places) that is Major.

The Major issue is that the contrast between LOD and relational platforms is asserted more than argued and fails to consider whether some of the listed relational platforms may have goals that are at least as well supported by a more relational model. I believe the paper would be stronger if it explored these actual contrasts more effectively.

I think the paper is important and worth publishing. Although I've declared this issue to be Major, I prefer to recommend "Accept after minor revision" since I think the changes are not too massive and I do not wish to recommend "Reconsider".

Comments for author File: Comments.docx

Author Response

Dear Reviewer,

Thank you for the thorough comments and suggestions for our manuscript! We agree with the points you have made and, as a consequence, have revised the manuscript accordingly. Please, find our responses to your comments below.

Sincerely,

the authors

Author Response File: Author Response.docx

Reviewer 2 Report

This MS describes the OpenBiodiv knowledge graph. It is a very interesting project, closely aligned with my own interests (and hence I should declare that I'm the author of the Ozymandias knowledge graph that they discuss). I have some comments on the comparison with Ozymandias, some queries about OpenBiodiv, and some small issues with the manuscript.

Issues with comparison with Ozymandias

The comparison with my own Ozymandias knowledge graph (Table 4) is a little misleading:

 "Ontology used" Ozymandias primarily uses schema.org, but also taxref-ld and TDWG LSID (see ref [26] for details).

"Number of statements" What do the 729 million statements represent? Given that OpenBiodiv is very detailed, it also has a lot of redundancy (e.g., the same entity can have multiple types, connections between entities are often expressed in multiple ways, e.g. using multiple terms for the same relationship, and including both a property and its inverse), so perhaps a better comparison is how many entities are in the two knowledge graphs. As stated on the Ozymandias home page https://ozymandias-demo.herokuapp.com "[t]he knowledge graph contains 203,652 taxa, 444,222 taxonomic names, 6,210 journals, 32,548 people, 68,217 articles". What are the equivalent counts for OpenBiodiv?

"Publications" For Ozymandias the primary source is indeed "Australian Faunal Directory" but there are many journals within that database, including several listed for OpenBiodiv

I think a reader trying to compare these two knowledge graphs might benefit from a figure showing the different ways the two knowledge graphs model taxonomic knowledge and publications.

Issues with the knowledge graph

Fig 2 shows the connections between Pensoft, Plazi, and the Biodiversity Literature Repository, which all contribute to OpenBiodiv. I find it frustrating that these three resources are still fundamentally silos. They use different identifiers for the same things, and these identifiers are rarely shared across the projects, nor are they cross-linked.

Lack of connections between OpenBioDiv and BLR.

The paper "Australian Assassins, Part I: A review of the Assassin Spiders (Araneae, Archaeidae) of mid-eastern Australia" https://doi.org/10.3897/zookeys.123.1448 is in OpenBioDiv http://openbiodiv.net/8473acfe-21cd-404a-8c44-ebcf643c1076

Figure 1 from that paper is http://graph.openbiodiv.net/resource?uri=http%3A%2F%2Fopenbiodiv.net%2F8e2019e4-9940-48d9-b29f-56b5d360b89c which makes no reference to the figure identifier used in the journal (http://doi.org/10.3897/zookeys.123.1448.figure1), nor the corresponding BLR record in Zenodo (https://zenodo.org/record/930589)

Lack of connections between OpenBioDiv and Plaza.

OpenBioDiv and Plazi share some identifiers in the sense of reusing UUIDs, such as

http://treatment.plazi.org/id/1B5C87BFFF87FFB6FDAF1697FABDF87D and http://openbiodiv.net/1B5C87BFFF87FFB6FDAF1697FABDF87D

The OpenBioDiv record http://openbiodiv.net/1B5C87BFFF87FFB6FDAF1697FABDF87D is contained within http://openbiodiv.net/817b1e6f-9eb6-4ffa-a37f-36c162db2053 which is the article with the prism:doi property "10.5281/zenodo.1162066".

The corresponding Plazi record is linked to the article via the property trt:publishedIn which link the article to "http://dx.doi.org/10.5281/zenodo.1162066". Hence the same article is referred to by two different properties with two different values. This is not in the spirit of linked data.

Furthermore, within the records the sections aren't strictly comparable: http://treatment.plazi.org/id/1B5C87BFFF87FFB6FDAF1697FABDF87D has four subsections identified by fragment identifiers, e.g. http://treatment.plazi.org/id/1B5C87BFFF87FFB6FDAF1697FABDF87D#section_1 whereas in OpenBioDiv there are three sections (found using the SPARQL below)

PREFIX c4o: <http://purl.org/spar/c4o/>

PREFIX po: <http://www.essepuntato.it/2008/12/pattern#>

SELECT *

WHERE

{

?subject po:isContainedBy <http://openbiodiv.net/1B5C87BFFF87FFB6FDAF1697FABDF87D> .

?subject c4o:hasContent ?content .

}

So we have inconsistent structuring of the documents, and different identifiers. It would be nice if these inconsistencies were resolved.

Other graph issues

Not all URIs in the database resolve in OpenBioDiv, e.g. http://openbiodiv.net/1B5C87BFFF87FFB6FDAF1697FABDF87D versus http://graph.openbiodiv.net/resource?uri=http:%2F%2Fopenbiodiv.net%2F1B5C87BFFF87FFB6FDAF1697FABDF87D Is this intentional?

Minor issues with MS

Li 24 “A lot more” is vague

Li 113 Roderic not Roderick

Li 145 It’s not obvious from the citation “ OpenBiodiv; Pensoft Publishers, 2018;”  what the “base package” is, or how I obtain it.

Li 177 Table 1, http://purl.org/dc/elements/1.1/ is in a bigger font

Li 340 “How many articles about Harmonia manillana were published every year between 2000 and 2019? Why not provide the user with a SPARQL query to answer this question?

Li 390 Can the authors be more specific about how OpenBiodiv relates to Bouchot, for example, could it be used to measure compliance with the goals of that declaration?

Li 499 (Ref. 41) What is the actual repo in http://graph.openbiodiv.net that we should use? There are multiple repositories that can be chosen, is it "depl2018-lite" ? Is there a way to have a direct link to query interface with the correct repository selected?

Author Response

Dear Reviewer,

Thank you for the thorough comments and suggestions for our manuscript! We agree with most of the points you have made and, as a consequence, have revised the manuscript accordingly. Please, find our responses to your comments in the attached document.

Sincerely,

the authors

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

Nice to see the example query in the appendix, this gives the reader a way to dive into the knowledge graph. My comments have been addressed as much as can be within the scope of the article. There is clearly a lot to do to reconcile the different approaches Plazi, BLR, and Pensoft take to representing the same information.