Duygusal akışın ikili etkileşimlerde keşfi: türk çift diyaloglarında bağlamsal ve bağlamsal olmayan duygu modelleri üzerine bir çalışma

This thesis introduces the "Couple Dialogue" dataset, tailored for analyzing conversational sentiment in Turkish. The dataset includes 14,294 utterances from 118 couple conversations, each meticulously annotated to reflect sentiment shifts and interpersonal interactions. The study contrasts two primary modeling approaches: the Non-Contextual Model and the Contextual Model. The Non-Contextual Model treats utterances in isolation, neglecting the intricacies of conversational dynamics and sentiment progression. This model incorporates an in-depth morphological analysis to account for the agglutinative and morphologically complex nature of the Turkish language, including diverse word forms and negation morphemes essential for sentiment analysis. Conversely, the Contextual Model leverages cutting-edge Large Language Models (LLMs) such as BERT, GPT-3.5 Turbo, Llama 2, and GPT-4, in addition to structures like DialogueRNN. This approach considers the sequential and relational dimensions of dialogues through prompt-based, fine-tuned, and embedding-based methods, with a focus on advanced embedding techniques and fine-tuning, particularly using pre-trained and fine-tuned Turkish BERT. The Contextual Model significantly outperforms the Non-Contextual Model, with a 9.98\% enhancement in Weighted F1 scores, as validated by statistical tests. This research not only pioneers the application of LLMs in Turkish conversational sentiment analysis but also emphasizes the importance of contextual comprehension in discerning complex emotional nuances in couple dialogues. It establishes a strong foundation for future research in sentiment analysis within linguistically diverse contexts.