Currently, each time you re-submit the form, we completely throw away the entities data. Instead, we should keep it. We need to figure out some other way to find out which entities to process in the next iteration (currently somewhere in getTriples and getNewEntityIds).

Right now the script will probably blow up when encountering somevalue or novalue. I don’t think we need to display those in the graph, but it shouldn’t stop traversal of other statements and entities.

When using properties that are applied at the lexeme level (e.g. P5191 "derived from"), the label of the graph nodes is the lexeme's lemma; but when a property that connects senses is used instead, what is shown is currently just the sense's gloss. I'd expect something like <lemma>: <gloss> (or even just the lemma) to be a more intuitive label for the graph nodes.

This was originally mentioned in the commit message of 00f3153. Adding an issue for visibility and tracking purposes :)

When I look at https://lucaswerkmeister.github.io/wikidata-lexeme-graph-builder/?subjects=L8160&predicates=P5191 I've got only 11 links from the L8160 but there is more (18 according to https://www.wikidata.org/w/index.php?title=Special%3AWhatLinksHere&target=Lexeme%3AL8160&namespace=146 )

Same if I change the Qid as entrey, z.B. https://lucaswerkmeister.github.io/wikidata-lexeme-graph-builder/?subjects=L8793&predicates=P5191

I guess it's to avoid too big graph but maybe we could have the limit as an input? (maybe also add the iteration as an input, like on https://angryloki.github.io/wikidata-graph-builder/ ;) )

If someone accidentally uses the wrong property (data type quantity, or string, or etc.), the website probably blows up. I suppose it shouldn’t.

Perhaps in small, grey text under the input boxes. Not everyone knows the L- and P-codes by heart, so showing them would help assuring them that they're using the right ones :)

If we request more than 50 entities at once, we get an API error. Split them up instead.

Make the website generally easier to use.

Just like we support starting from a set of entities, allow following a set of properties. Shouldn’t be too hard to implement.

Suggested by VIGNERON on-wiki.

We add duplicate label/lemma texts on each restart, which makes the texts look ugly in the SVG. Fix that. (Then we can also label entities that we haven’t loaded yet with their entity ID instead of the empty string.)

Maybe it could be half-transparent if visibility of the edges underneath is compromised.