Analiza sovražnega govora in sentimenta v parlamentarnem govoru v Sloveniji s pomočjo orodij za procesiranje naravnega jezika

CIRAR, NEJA (avtor), Robnik-Šikonja, Marko (mentor)

Diplomsko delo se ukvarja z analizo sovražnega govora in sentimenta v parlamentarnem govoru Državnega zbora RS. Sovražni govor postaja vedno bolj pereč družbeni problem, zato je namen diplomskega dela ugotoviti, v kolikšni meri se pojavlja med poslanci na parlamentarnih sejah ter kakšen sentiment je prisoten (pozitiven oz. negativen). Napisali smo programsko opremo v programskem jeziku python, pri čemer smo s pomočjo ogrodja HuggingFace izdelali več modelov strojnega učenja tipa BERT za prepoznavanje sovražnega govora in zaznavanje sentimenta v parlamentarnem govoru. Uporabili smo CroSloEngual BERT-model, ki je bil vnaprej naučen v slovenskem, hrvaškem in angleškem jeziku. Ugotovili smo, da je uporaba sovražnega govora v letih 2016–2020 v prvih štirih letih padala, nato pa je leta 2020 pričela strmo rasti. Pozitiven sentiment je bil v letih 2016 do 2018 pogosteje izražen kot negativen, v prihodnjih dveh letih, vse do junija 2020, pa je prevladoval negativen sentiment. Rezultati diplomskega dela dajejo vpogled v to, v kolikšni meri se med poslanci Državnega zbora RS pojavlja sovražni govor ter kakšen sentiment je prisoten v njihovem govoru, česar do sedaj ni ugotavljala še nobena študija. Diplomsko delo lahko služi kot izhodišče za nadaljnje študije na področju procesiranja parlamentarnega jezika ter pri analizi različnih aspektov parlamentarnega govora v Sloveniji.

• Jezik: Slovenski jezik
• Ključne besede: sentiment, sovražni govor, parlament, parlamentarni govor, procesiranje naravnega jezika
• Vrsta gradiva: Diplomsko delo/naloga
• Tipologija: 2.11 - Diplomsko delo
• Organizacija: FU - Fakulteta za upravo
• Kraj izida: Ljubljana
• Založnik: [N. Cirar]
• Leto izida: 2022
• Št. strani: IX, 53 str.
• PID: 20.500.12556/RUL-142886
• Datum objave v RUL: 30.11.2022

Sekundarni jezik

• Jezik: Angleški jezik
• Naslov: The analysis of hate speech and sentiment in Slovene parliamentary speech with natural language processing tools

Izvleček

The thesis deals with the analysis of hate speech and sentiment in the parliamentary speech of the National Assembly of the Republic of Slovenia. Hate speech is becoming an increasingly pressing social problem, so the purpose of the thesis is to determine to what extent it occurs among MPs at parliamentary sessions and what kind of sentiment is present (positive or negative). We used python programming language and the HuggingFace framework to build several BERT-type machine learning models for hate speech recognition and sentiment detection in parliamentary speech. We used the CroSloEngual BERT model, which was pretrained on Slovenian, Croatian and English language. We found that the use of hate speech between 2016 and 2020 declined in the first four years, but then began to rise sharply in 2020. In the years 2016 to 2018, positive sentiment was expressed more often than negative, and in the next two years, until June 2020, negative sentiment prevailed. The results provide a deeper insight into the extent to which hate speech occurs among the members of the National Assembly of the Republic of Slovenia and what kind of sentiment is present in their speech, which has not been determined by previous studies. The thesis can serve as a starting point for further studies in the field of parliamentary language processing and in the analysis of various aspects of parliamentary speech in Slovenia.

• Ključne besede: sentiment, hate speech, parliament, parliamentary debates, natural language processing