Analiza sentimenta za hrvatski jezik

master's thesis

• Title (english): Sentiment analysis for Croatian
• Author: Lucija Košković
• Mentor: Krešimir Šojat (mentor)
• Committee member: Marko Tadić (član povjerenstva)
• Committee member: Božo Bekavac (predsjednik povjerenstva)
• Granter: University of Zagreb; Faculty of Humanities and Social Sciences; (Department of Linguistics); Zagreb
• Defense date and country: 2023, Croatia
• Scientific / art field, discipline and subdiscipline: HUMANISTIC SCIENCES; Philology; General Linguistics

Abstract

S ciljem povećanja broja jezičnih resursa za hrvatski jezik, u sklopu ovog rada kreiran je leksikon mišljenja. Leksikon mišljenja sastoji se od ukupno 600 pridjeva, od čega ih je 300 pozitivnih te 300 negativnih. Leksikon je kreiran rječničkim pristupom izrade leksikona, tzv. bootstraping metodom proširivanja korištenjem hrvatskoga WordNeta (CroWN), hrvatskoga web korpusa (hrWaC) i Hrvatskoga jezičnog portala (HJP). Popis početnih riječi (seed words) kreiran je na temelju popisa najčešćih pridjeva u sklopu korpusa hrWaC. Na temelju skupa početnih riječi, popis pozitivnih i negativnih riječi proširivao se u nekoliko koraka. Najprije se programskim kodom pristupilo sinonimskim skupovima u sklopu CroWN-a. Potom, unutar hrWaC-a uz pomoć trezaurusa u sklopu SketchEnginea identificirani su dodatni sinonimi i slične riječi. Konačno, pretragom u sklopu HJP-a pronađeni su još neki od sinonima. Leksikon je testiran na setu filmskih recenzija, pri čemu je mjera točnosti iznosila 72.86%, mjera preciznosti 74.34%, mjera odziva 90.32% te F1-mjera 81.64%. Korištena metodologija omogućila je višu razinu provjere i točnosti kreiranog leksikona. Leksikoni mišljenja, kao i analiza sentimenta, korisni su alati s brojnim područjima primjene. Podizanje svijesti o manjku jezičnih resursa u hrvatskom jeziku ključno je za rast broja istraživanja u ovom području, kao i unaprjeđenje istih.

Abstract (english)

With the aim of augmenting linguistic resources for the Croatian language, this thesis introduces a sentiment lexicon. The sentiment lexicon comprises a total of 600 adjectives, consisting of 300 positive and 300 negative ones. The lexicon was created using a dictionary-based approach and a bootstrapping method by leveraging Croatian WordNet (CroWN), Croatian web corpus (hrWaC), and the Croatian Language Portal (HJP). The list of seed words was generated based on the most common adjectives in the hrWaC corpus. Using these seed words, the list of positive and negative words was expanded in several steps. Firstly, synonym sets within CroWN were accessed using Python script. Next, within hrWaC, additional synonyms and similar words were identified using a thesaurus in the SketchEngine. Finally, a search within HJP yielded more synonyms. The lexicon was tested on a set of movie reviews, with an accuracy measure of 72.86 %, precision measure of 74.34 %, recall measure of 90.32 %, and F1 score of 81.64 %. The employed methodology enabled a higher level of verification and accuracy in creating the lexicon. Sentiment lexicons, as well as sentiment analysis, are valuable tools with numerous applications. Raising awareness about the lack of linguistic resources in the Croatian language is crucial for the growth of research in this area and for its improvement.

• Keywords: analiza sentimenta; leksikon sentimenta; leksikon mišljenja; riječi mišljenja
• Keywords (english): sentiment analysis; sentiment lexicon; opinion lexicon; opinion words
• Language: croatian
• URN:NBN: urn:nbn:hr:131:493335
• Study programme: Title: Linguistics (double major); specializations in: General Linguistics, Comparative Linguistics, Computational Linguistics, Aplied Linguistics and Cognitive Linguistics, Cognitive; Course: Computational Linguistics; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra lingvistike (magistar/magistra lingvistike)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Access restricted to students and staff of home institution
• Created on: 2023-11-30 08:20:53