PLSI Utilization for Automatic Thesaurus Construction

Abstract

Abstract. When acquiring synonyms from large corpora, it is important to deal not only with such surface information as the context of the words but also their latent semantics. This paper describes how to utilize a latent semantic model PLSI to acquire synonyms automatically from large corpora. PLSI has been shown to achieve a better performance than conventional methods such as tf·idf and LSI, making it applicable to automatic thesaurus construction. Also, various PLSI techniques have been shown to be effective including: (1) use of Skew Divergence as a distance/similarity measure; (2) removal of words with low frequencies, and (3) multiple executions of PLSI and integration of the results.

Co-clustering with Recursive Elimination for Verb Synonym Extraction from Large Text Corpus

“…The data are available athttp://vsearch.cl.cs.okayama-u.ac.jp/ index.php † † The detailed setting is described in Hagiwara et al [5]. † † † The number of output clusters will be the same as the number of words, i.e., tokens.…”

“…The major word clustering schemes are word distributionbased methods [1], [2] and decomposition-based methods [3]- [5]. The decompositional methods cluster words on the basis of the orthogonal vectors of the assumed latent semantic clusters between words.…”

“…We employed Hagiwara's synonym extraction approach [5] that is based on the probabilistic latent semantic indexing (PLSI) [14] method. Hagiwara's approach collects similar words by evaluating the similarity of two words on the basis of PLSI-based probability distribution.…”

Extraction of Verb Synonyms using Co-clustering Approach

“…As a vector-based approach of collecting verb synonyms, we employ Hagiwara's approach [5] which applies PLSI to decreasing unrelated vector space. The Hagiwara's approach extracting similar verbs to a key verb on the basis of similarity of vectors between verbs.…”

“…The major methods of word clustering are word distribution-based methods [6,11] and decompositionbased methods [12,10,5]. The decompositional methods cluster words on the basis of orthogonal vectors of assumed latent semantic clusters between words.aim of the methods is to make a compact orthogonal vectors based on global optimization such as EM algorithm.…”