Link analysis methods like the popular PageRank are increasingly being applied to lexical knowledge bases to deal with a number of natural language processing problems, including unsupervised word sense ranking and disambiguation. Compared to plain-text, the topic of sense ranking in semistructured data has been however studied marginally. This paper aims to bridge PageRank-based word sense ranking and tree-structured data. We propose PageRank-style methods for the structural sense ranking problem, which take into account tree structural relations as well as semantic relatedness in the constituents of tree data. The proposed methods are comparatively evaluated with existing PageRank methods for word sense disambiguation. Effectiveness and efficiency of PageRank methods have been assessed on various data with different domain vocabularies.
Evaluating PageRank methods for structural sense ranking in labeled tree data
TAGARELLI, Andrea;
2012-01-01
Abstract
Link analysis methods like the popular PageRank are increasingly being applied to lexical knowledge bases to deal with a number of natural language processing problems, including unsupervised word sense ranking and disambiguation. Compared to plain-text, the topic of sense ranking in semistructured data has been however studied marginally. This paper aims to bridge PageRank-based word sense ranking and tree-structured data. We propose PageRank-style methods for the structural sense ranking problem, which take into account tree structural relations as well as semantic relatedness in the constituents of tree data. The proposed methods are comparatively evaluated with existing PageRank methods for word sense disambiguation. Effectiveness and efficiency of PageRank methods have been assessed on various data with different domain vocabularies.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.