期刊名称：International Journal of Computer and Information Technology
印刷版ISSN：2279-0764
出版年度：2016
卷号：5
期号：6
页码：491-499
出版社：International Journal of Computer and Information Technology
摘要：In this study, we have developed a Turkish
semantic relation extraction tool. The tool takes an unparsed
corpus as input and gives hyponym, meronym and antonym
words with their reliability scores as output for given target
words. Corpus is parsed by Turkish morphological parser called
Zemberek and word vectors are created by Word2Vec for each
unique word in corpus. To extract relation patterns, hyponymy,
holonymy, antonymy pairs called initial seeds are prepared then,
all possible relation patterns are extracted using initial seeds.
Reliability of patterns are calculated using corpus statistics and
various association metrics. Reliable patterns are selected to
extract new semantic pairs from parsed corpus. To determine
correctness of extracted pairs, total pattern frequency, different
pattern frequency and Word2Vec vector cosine similarity have
been used. After experiments, we have obtained 83%, 63%-86%,
and 85% average precisions for hyponymy, holonymy and
antonymy relations, respectively.
