Automatic synset detection from Turkish dictinary using confidence indexing

---

Tezin Türü: Doktora

Tezin Yürütüldüğü Kurum: Çukurova Üniversitesi, Fen Bilimleri Enstitüsü, Bilgisayar Mühendisliği, Türkiye

Tezin Onay Tarihi: 2020

Tezin Dili: İngilizce

Öğrenci: ERHAN TURAN

Danışman: Umut Orhan

Açık Arşiv Koleksiyonu: AVESİS Açık Erişim Koleksiyonu

Özet:

In this study, a Turkish semantic network is designed from a non-machine-readable monolingual dictionary. Dictionary lemmas and definitions are extracted and processed into a Lemma-Sense weighted bipartite graph model and analyzed for semantic relations. Primary semantic relations of a general semantic network as hypernym, synonym and antonym analyzed based on Lemma-Sense dictionary and added to the semantic network at sense level. Synonym relations are tagged with a confidence level for an improved synset detection. Also, morpho-semantic relations added between the lemmas and their derived and compound lemmas. N-Gram analysis is used to find patterns of any additional semantic relation. These additional semantic relations are supplemented to the semantic network. Finally, synonyms are clustered to form the synsets with a spanning-tree based synset detection algorithm. Synset results are compared with an up-to-date and notable Turkish wordnet.