Project Update

[Michael Mayer]

Dear contributors,

We're back in business! I've recently improved image classification (now based on NASNet Mobile) and added auth to the built-in TiDB server. Color extraction is also more sophisticated. Theresa is doing additional fine tuning right now. New demo will be released soon.

Anyone knows a compact database of similar nouns and locations for implementing our search? If you search for fish, you should also find sharks. Same for NYC and New York.

Note that we have a new chat: https://gitter.im/browseyourlife/community 

It's easier to join and consumes less browser memory than Slack. We'll keep the Slack channel open for questions though.

Have a great labour day,

Michael

[Andrea Ceroni]

Hi Michael,

For suggesting related query terms you could have a look at public resources like WordNet and its synsets, Yago2 and DBpedia. They are quite big though and probably there will be some more work to be done on top of them to get what you want.

Another possibility could be suggesting related words based on a db of word embeddings. But this would require even more additional work I guess.

Best,

Andrea

--
You received this message because you are subscribed to the Google Groups "PhotoPrism Developers" group.
To unsubscribe from this group and stop receiving emails from it, send an email to developers+...@photoprism.org.
To post to this group, send email to devel...@photoprism.org.
To view this discussion on the web visit https://groups.google.com/a/photoprism.org/d/msgid/developers/CADra6coc1WnGSGGSK9pib_rsCRzsCYpxunYk1fHaWGEJUFLTYw%40mail.gmail.com.

[Michael Mayer]

Hello Andrea,

thanks for providing additional hints! I already had a look at Yago,
DBpedia and Wikidata (successor of Freebase):

https://www.wikidata.org/wiki/Wikidata:Database_download

You are right, they are all very big and would require significant
additional work. Wikidata for example uses latin names for animals and
their categories:

https://www.wikidata.org/wiki/Q48410245

So the "Great white shark" is listed as "Squalus carcharias". Not even
a reference to fish on this page.

To me this looks like a standard use-case somebody out there already
solved a million times...

Michael

> To view this discussion on the web visit https://groups.google.com/a/photoprism.org/d/msgid/developers/CAMq4aV%2BNdVU6esGWY%3DHujhvHaK%2BGw%3D0W28XBEGx9H_GRxhRTRA%40mail.gmail.com.

[Andrea Ceroni]

I think you should get what you need by traversing the ontology up via, for instance, is-a relations. At some point you would find that a shark is a fish. I have used Yago2 long time ago for similar tasks, but again the size might be an issue. Maybe you could isolate part of the ontology with only the relations you need?

Andrea

To view this discussion on the web visit https://groups.google.com/a/photoprism.org/d/msgid/developers/CADra6cqMCgKC_SBdcLJhMMmrP%2BXCuMnksHfyL9aw2so98BvjVw%40mail.gmail.com.

[Michael Mayer]

Yes, I find that it's a fish, but in latin. So I also need to consider
the various translations and alternative terms and descriptions which
is a mess. Users also don't search for felidae when they want to see
cats. It's pretty scientific and not good for general use without lots
of clean-up.

Extracting the ontology from Yago2 might be easier, although also a
lot of work. Maybe someone can help with that.

Michael

> To view this discussion on the web visit https://groups.google.com/a/photoprism.org/d/msgid/developers/CAMq4aVLtyiu%3DRrQSdzd3%3DMbiB-53UsmiHrqFt%3DGi-qCY%3D0rviw%40mail.gmail.com.

[Michael Mayer]

What "YAGO theme" should we download for building our ontology or do
we need everything?

TAXONOMY: All types of entitites, and the class structure of YAGO.
Moreover, it has formal definitions of YAGO relations.
SIMPLETAX: An alternative, simpler taxonomy of YAGO.
CORE: Core facts of YAGO, such as the facts between entities, the
facts containing literals,i.e., numbers, dates, strings, etc.
GEONAMES: Geographical entities, classes taken from GeoNames.
META: Temporally and spatially scoped facts together with statistics
and extraction sources about the facts.
MULTILINGUAL: The multilingual names for entities.
LINK: The connection of YAGO to Wordnet, DBPedia, etc.
WIKIPEDIA: Multilingual infobox attributes, templates, sources, etc.
for Wikipedia infoboxes.
OTHER: Miscellaneous features of YAGO, such as Wikipedia in-outlinks,
GeoNames data etc.

See https://www.mpi-inf.mpg.de/departments/databases-and-information-systems/research/yago-naga/yago/downloads/