camerontt2000
commented
11 years ago Sorry, "hyponym" should of course be replaced with "Narrower" terms.
Narrower Terms: Our idea section adopts the semantics of SKOS (link to docs), a W3C standard for describing taxonomies governed by semantic broader and narrower relationships. Our methods deem a term X to be a candidate for being classified under Y in our idea taxonomy when statistical analysis has deemed X highly related to and more specific than Y.
colinallen
Not so much a bug as a design idea:
I constantly find myself having to explain the various lists that appear on an idea page (hyponyms, instances, occurrences, etc.) to people I am showing the site. How about instead of having just a "list generated by statistical measures" hover text for the gear icon, we could have a brief explanation of the relation born between items on that list and the target entity. We should also then introduce some kind of icon for "instances" and "links" with similar hover-over text indicating that they are inferred in the taxonomy.
Examples
Instances: X is an instance of Y when human feedback has deemed X more specific than and highly related to Y, and our logical reasoner (which also settles inconsistencies in global feedback) has categorized X under Y's portion of the InPhO semantic taxonomy.
Links: Semantic links allow bridges in our taxonomy between terms that are highly related by cannot be linked by taxonomic subsumption relationships. X is linked to Y when human feedback has determined that they are highly related, but due to inconsistency or taxonomic cycles, the reasoner cannot classify X under Y.
Related terms: Our statistical methods deem a term X to be related to term Y when an occurrence of term X contains a significant amount of information about an occurrence of a term Y. In other words, if our a priori and a posteriori expectations about Y's distribution are very different given X, then X is probably highly related to Y.
(repeat similar for "related thinkers")
Hyponyms: A hyponym is a term whose semantic field is included within that of another term. In our case, X is considered a candidate hyponym for Y when statistical analysis has deemed X highly related to and more specific than Y.
A deeper worry: we might want to stop using the word "hyponym". Technically X is a hyponym of Y only if X is a subtype of Y, and this just isn't true for most of the instances and hyponym candidates we use (it's true if we use the "idea_about" prefix, but we really don't have to fight this fight anymore now that SKOS is available). We're really just not doing the right kind of NLP to have a very diagnostic search for context-insensitive hyponyms. If we move to SKOS as an underlying model, then we don't have to continue conflating "isa" and the more permissive "broader" in our explanations. So, probably "hyponyms" should be replaced with
Broader Terms: Our idea section adopts the semantics of SKOS (link to docs), a W3C standard for describing taxonomies governed by semantic broader and narrower relationships. Our methods deem a term X to be a candidate for being classified under Y when statistical analysis has deemed X highly related to and more specific than Y.