Auto-suggest related resources through named entity recognition (NER)

Driebot commented 8 years ago

In gitlab by david on Mar 21, 2016, 17:23

In admin, parse resource body text and recognise keywords in Ginger. Then based on those resources, automatically suggest related resources. Content editors can then accept/reject the suggestions. When they accept, create an outgoing edge to the selected resource(s).

For parsing the text:

Apache Stanbol
DBpedia Spotlight can also be used with other datasets, your own text, in custom domains, etc. This will require that you rebuild the models using our Data Generation https://github.com/dbpedia-spotlight/dbpedia-spotlight/wiki/Data-generation-manual pipeline.

Also needs some UI work where editors can accept/reject suggestions.

/cc @emine @fred

ddeboer commented 8 years ago

Can https://research.googleblog.com/2016/05/announcing-syntaxnet-worlds-most.html help? Seems to be about language understanding (syntax/grammar) rather than simply extracting entities.

Dirklectisch commented 8 years ago

You are certainly right that it is more advanced than just ripping out the entities. But it can also just filter nouns, verbs or whatever you want (if I understand that blog post correctly).

fredpook commented 8 years ago

DBpedia spotlight already filters "noise" since the entities come from crowd-sourced wikipedia pages.

ddeboer commented 6 years ago

We now have a DBpedia Spotlight module for extracting DBpedia resources from text.

driebit / ginger

Auto-suggest related resources through named entity recognition (NER) #46