semantize awesome lists (build awesomes KG)

[VladimirAlexiev]

((As a sort of "opposite" of #16, here's something I posted April 2, 2021 at Ontotext (https://ontotext.atlassian.net/issues/DATA-110) but never got around to do.
It has absolutely nothing to do with Shapes, but the community here (who like semantics and awesome lists) may provide interesting ideas.))

"awesome" is not some stupid marketing slogan, it's a movement to provide excellent lists of resources and surveys on a certain topic.

https://awesomelists.top/ has collected an amazing wealth of such lists.

• It's driven by https://github.com/sindresorhus/awesome and strict content format rules that allow that app to parse the lists: pull_request_template.md, contributing.md, create-list.md, awesome-lint, generator-awesome-list to generate a skeleton for a list

I think it's an excellent internship project to turn this "awesome" github markdown info to LOD.

• parse markdown of https://github.com/sindresorhus/awesome
• parse markdown of all linked lists
  • IMPORTANT: the above steps are probably already done by https://github.com/lockys/Awesome.json , which probably drives https://github.com/lockys/AwesomeSearch and https://github.com/lockys/AwesomeSearchReact , which drive https://awesomelists.top/
• optionally, parse Github info of those resources that are Github projects. JSI have a Github dump as part of enrichMyData/InnoGraph, and have labels, READMEs, and parsed out AI/ML related info
• design a semantic model
  • ideally, interpret Awesome.json as JSONLD by adding a context
  • should start from SKOS (topics = a tree of concepts)
  • then add resources under them
  • convert the info to triples
  • optionally, make/generate Semantic Object Model so we can have GraphQL
• provide FTS
• enrich with external info from Wikidata. Many resources should be matchable by website or name+topic
  • eg "Linkedin Gobblin" ~ https://gobblin.apache.org/ ~ https://projects.apache.org/project.html?gobblin ~ https://www.wikidata.org/wiki/Q28915176 by website or Apache Project ID
• try to determine resource type (is it a piece of software, dataset, survey, individual research paper, or what?)
• provide some topic overviews or other ways of visualizing the wealth of info
• explore whether the same resource occurs in several lists/topics, to compute lateral links between topics ("skos:related" or "rdfs:seeAlso")

Note: lockys/AwesomeSearch#121 would provide better linkability to "awesome" topics (subheadings) on that website

[VladimirAlexiev]

@simonstey , @tpluscode , @amivanoff
What do you think of this? Is it interesting to semantize ALL awesome lists and provide a better search over them, and maybe linking out to other resources?

https://awesomelists.top is a great aggregation, but the search is very bad:

• Eg "zeppelin" doesn't find https://awesomelists.top/#/awesome-spark/awesome-spark
• Eg "gobblin" finds weird strings but not https://awesomelists.top/#/igorbarinov/awesome-data-engineering
  

[VladimirAlexiev]

In contrast, https://www.trackawesomelist.com/search/ has a great search.
Found it at #6, thanks @amivanoff !
Example:

https://github.com/trackawesomelist/trackawesomelist-source says "json files as database" so we could that as source for RDFization.

[simonstey]

@VladimirAlexiev interesting.. I only found https://github.com/trackawesomelist/trackawesomelist-source/blob/main/migrations/old-meta.json and https://github.com/trackawesomelist/trackawesomelist-source/blob/main/config.yml where it seems to be dynamically parsing the github repos listed in the config.yml here https://github.com/trackawesomelist/trackawesomelist-source/blob/main/adapters/github.ts

[amivanoff]

From the description, this task summary sounds more like "Build Awesomes Knowledge Graph". From awesome lists parsing to KG in a triplestore.

Similar to the recent MLSea-KG attempt to build the machine learning KG with extended ML Schema and a search engine, by @andimou and colleagues.

Awesomes semantization sound to me more like "adding RDFa-like Schema.org-like markup to all awesome lists". An "impossible task" considering the amount of interaction with maintainers.

Maybe it is better to refine task title to something about knowledge graph construction to avoid this type of confusions?

[VladimirAlexiev]

• @amivanoff I changed the task name as you suggested.
• The idea is not to add RDFa to awesome lists, but only to parse awesome lists and try to connect them to other LOD
• MLSea is amazing, thanks! Posted merge MLSea and LinkedPapersWithCode dtai-kg/MLSea-KGC#2