Description
I was testing the improvements made in #250 today using GPS spoofing. I was quite surprised how many items have question marks as icons. But the problem seems to be the data we use:
https://tools.wmflabs.org/wiki-needs-pictures/data/data.csv
The list classifies 3600 things as "adm3rd", which is probably a artefact of some vandalism? 150 000 items are undefined. That is 75 % of the 200 000 items in the list.
Another issue is that the list gives a string for the target-item of the p31 statement. I think it would be better to use the Q-ID both in the data and in the program code (Don't compare to the English label which could change over time).
Probably the easiest way of solving this would be to switch to a query.wikidata.org request. That returns the p31 statement along with some other information (e.g. the locale of the user to fetch the right labels). With the p31 target we can then assign the icons like this:
switch(place.qid) {
case "Q4022":
icon.setImageResource(R.drawable.icon_river;
break;
case "Q355304":
icon.setImageResource(R.drawable.icon_river);
break;
Items:
Assigning the symbols is not trivial, but for a start we can just collect the most used items in p31 statements.