Hasil Pencarian  ::  Simpan CSV :: Kembali

"Dalam lanskap pendidikan modern yang semakin terintegrasi dengan teknologi AI, kebutuhan akan interaksi yang lebih dinamis dan adaptif antara mahasiswa dan sistem AI menjadi krusial, misalnya dalam simulasi ujian lisan. Oleh kare- na itu, penelitian ini memanfaatkan berbagai skema prompting seperti Proactive Chain-of-Thought (PCoT), Standard Prompting, Demonstration-based Prompting, Instruction-based Prompting, serta pendekatan Zero-Shot dan Few-Shot untuk me- ningkatkan proaktivitas LLM. Hasilnya, ProCoT Zero-Shot terbukti efektif dalam identi kasi kebutuhan Clari cation Need Prediction (CNP) dengan F1 Score 0.99. Untuk Clari cation Question Generation (CQG), ProCoT Few-Shot menunjukkan keunggulan signi kan dengan BLEU-1 41.8. Sementara itu, dalam target-guided dialogue, metode Proactive (baik Zero-Shot maupun Few-Shot) menunjukkan ki- nerja superior dalam kualitas respons generatif untuk pemanduan dialog menuju target dialog. Penelitian ini menyimpulkan bahwa efektivitas prompting sangat spe- si k terhadap tugas dan memerlukan pemilihan strategi untuk mencapai interaksi AI yang lebih efektif dan adaptif.

---

In the rapidly evolving landscape of modern education, increasingly integrated with AI technology, the need for more dynamic and adaptive interactions between stu- dents and AI systems has become crucial. This research analyzes and optimizes various prompting techniques to enhance the proactivity of Large Language Mo- dels (LLMs), speci cally in clari cation scenarios (Clari cation Need Predictio- n/CNP and Clari cation Question Generation/CQG) and target-guided dialogues. Various prompting schemes, including Proactive Chain-of-Thought (PCoT), Stan- dard Prompting, Demonstration-based Prompting, Instruction-based Prompting, as well as Zero-Shot and Few-Shot approaches, were utilized. The results show that ProCoT Zero-Shot proved highly effective in Clari cation Need Prediction (CNP), achieving an F1 Score of 0.99. For Clari cation Question Generation (CQG), Pro- CoT Few-Shot demonstrated signi cant superiority with a BLEU-1 score of 41.8. Meanwhile, in target-guided dialogues, the Proactive method (both Zero-Shot and Few-Shot) showed superior performance in generative response quality for guiding dialogues towards a target. This research concludes that prompting effectiveness is highly task-speci c and necessitates a nuanced strategy selection to achieve more effective and adaptive AI interactions."