NNexus::Classification - Dismabiguation logic for NNexus concept harvests
use NNexus::Classification qw(disambiguate msc_similarity); $concepts_refined = disambiguate($concept_harvest,%options); $similarity_score = msc_similarity($category1,$category2);
NNexus::Classification contains disambiguation and clustering algorithms for determining a subset of "relevant" concept candidates from a given concept harvest. Relevance is determined heuristically.
The current algorithm considers two facets of "relevance":
1. Relevant candidates come from empirically similar domains of knowledge. To this extent, a similarity metric has been extracted from 3+ million mathematical reviews in Zentrallblatt Math, each annotated with categories from the Math Subject Classification. 2. Technical terms are more likely to be relevant. Consequently: - The more words in a candidate, the more likely that it is a term - The more characters in a candidate, the more likely that it is a term
$concepts_refined = disambiguate($concept_harvest,%options);
Disambiguates a concept harvest, as returned by NNexus::Discover, following the algorithm in the description.
Currently the only accepted option is a boolean value for "verbosity".
$similarity_score = msc_similarity($category1,$category2);
Retrieves the ZBL similarity score of two MSC categories given via the standard MSC naming scheme (e.g. 00-XX, 15Axx, 15B33)
Note that currently the similarity metric only covers the top-level MSC categories.
Deyan Ginev <email@example.com>
Research software, produced as part of work done by the KWARC group at Jacobs University Bremen. Released under the MIT License (MIT)