PoliticalNLP

Taking politicians words in political duels under data scrutiny, analyzing their most frequent narratives and used words in Slovak language.

Berme politikov za slovo analyzovaním ich vyjadrení v politických dueloch, v slovenskom jazyku.

• Whisper for speech-to-text
• ChatGPT for smoothing out phonetically incorrect text
• Basic NLP tokenization and stemming to acquire most used narratives and words

Case study

TA3, V Politike: Fico vs. Kollár

• sourced from Televizia TA3 YouTube channel
• reveals that the focal point of discussion was the current president Zuzana Čaputová and former police president Stefan Hamran, without the need to watch the political duel, which may prove detrimntal to mental health and to possible engagement of public in activism
• paves the way to counter misinformation by showing each politicians "background agenda" based on their rhetorics, purely data-based

Fig. 1 | Histogram showing most common words in the political duel (skipping high-frequency words such as "ktorý/ktorá" etc.)

Future work

• classification of different speakers using frequency analysis (Wavelet, Quefrency or simple Fourier transform) so words can be correctly assigned to speakers
• sentiment analysis
• most common statement outlineing
• researching the use of a homeland MLM SlovakBERT
• streamlining with GPT4 instead of GPT3.5