https://bugzilla.wikimedia.org/show_bug.cgi?id=46641
Web browser: ---
Bug ID: 46641
Summary: Storage for analysis results, term props
Product: MediaWiki extensions
Version: unspecified
Hardware: All
OS: All
Status: NEW
Severity: normal
Priority: Unprioritized
Component: WikidataRepo
Assignee: [email protected]
Reporter: [email protected]
CC: [email protected],
[email protected]
Classification: Unclassified
Mobile Platform: ---
In several places we need some kind of storage place for results gathered from
heavy analysis. Two (three) of three (four) known places have somewhat working
solutions. Those are Items and Properties, and to some extent Queries. For
those we can use page props. The one with no solution is terms used in Claims,
but it seems like we can make some kind of "term props" that do the same as
page props.
It is one thing that is somewhat annoying about this way of doing this. Page
props is said to be easy (lightweight) to regenerate but the analysis results
are not in general very lightweight, they are actually quite heavy to generate
in some cases.
The analysis results for terms are in general terms from one property that are
similar to terms from one or more other properties. The results should then be
stored in something like a "term props" table. The results are typically
employed as ranking during sorting of terms to be used in claims.
The calculation can be done by using a multivariate Bayes and sorting on the
resulting a posteriori probability. This is a two-level map-reduce problem
where the first part is the Bayes and the second is some kind of dimensionality
reduction.
The analysis results for properties are in general properties from one item
that are similar to properties from one or more other items. The results in
this case can be saved as ordinary page props. Usually the results from the
analysis of properties are employed as ranking of properties during creation of
claims, if a property is commonly used together with some other property it
will be ranked up in the list. Otherwise it is similar to the terms analysis.
--
You are receiving this mail because:
You are on the CC list for the bug.
You are watching all bug changes.
_______________________________________________
Wikibugs-l mailing list
[email protected]
https://lists.wikimedia.org/mailman/listinfo/wikibugs-l
