RANCANG BANGUN ASISTEN VIRTUAL BERBASIS FINE-TUNING LARGE LANGUAGE MODEL DALAM MODEL PEMBELAJARAN OPEN INQUIRY LEARNING UNTUK MENINGKATKAN KEMAMPUAN CRITICAL THINKING

Wendi Kardian, - and Erna Piantari, - and Enjun Junaeti, - (2025) RANCANG BANGUN ASISTEN VIRTUAL BERBASIS FINE-TUNING LARGE LANGUAGE MODEL DALAM MODEL PEMBELAJARAN OPEN INQUIRY LEARNING UNTUK MENINGKATKAN KEMAMPUAN CRITICAL THINKING. S1 thesis, Universitas Pendidikan Indonesia.

Abstract

Pendidikan berkualitas dan kemampuan berpikir kritis menjadi fokus Indonesia Emas 2045, meskipun hasil PISA 2022 mengindikasikan kelemahan dalam critical thinking peserta didik. Penelitian ini bertujuan untuk mengembangkan sistem asisten virtual berbasis LLM yang terintegrasi dalam LMS untuk mendukung pembelajaran analisis data dan mengetahui pengaruhnya pada peningkatan critical thinking peserta didik, menggunakan metodologi CRISP-DM serta pendekatan R&D melalui model SLEEG (menggabungkan ADDIE dan PDCA). Sistem di-fine-tune dengan GPT-4o-mini menggunakan 1.500 data training dan 170 data validasi, sehingga mampu meningkatkan pemahaman konteks dan kemampuan berpikir kritis peserta dengan peningkatan N-gain 0.55 (kategori sedang), terutama pada aspek inferensi dan eksplanasi, meskipun evaluasi masih perlu ditingkatkan. Evaluasi performa model LLM menunjukkan akurasi rata-rata sekitar 83% untuk kelogisan AI dengan tingkat kesepakatan rater (QWK) sebesar 0.67, yang mencerminkan konsistensi dalam menilai kelogisan jawaban AI, interpretasi, dan analisis jawaban AI, meski aspek evaluasi, eksplanasi, dan regulasi diri masih belum optimal dalam mendukung kemampuan berpikir reflektif peserta. Selain itu, terdapat korelasi positif lemah (r = 0.32) antara proporsi dominasi model dan peningkatan kemampuan critical thinking, dengan pengaruh terkuat terlihat pada aspek interpretasi (proporsi 72% dan n-gain 0.53). Strategi pembelajaran yang diterapkan juga menunjukkan peningkatan kepuasan dan keterlibatan peserta didik, dengan skor CSAT 88% dan NPS 82%, meskipun tantangan seperti respons AI yang kurang relevan dan keterbatasan context window masih perlu penanganan lebih lanjut. Quality education and critical thinking are priorities for Indonesia Emas 2045, although the PISA 2022 results indicate weaknesses in students' critical thinking skills. The research aims to develop a virtual assistant system based on a Large Language Model (LLM) integrated into a Learning Management System (LMS) to support data analysis learning and to examine its impact on enhancing students' critical thinking skills, using the CRISP-DM methodology and a Research and Development (R&D) approach through the SLEEG model (which combines ADDIE and PDCA). The system was fine-tuned using GPT-4o-mini with 1,500 training data and 170 validation data, enabling improved context comprehension and enhanced critical thinking skills among students, with an N-gain increase of 0.55 (medium category), particularly in the aspects of inference and explanation, although evaluation still needs improvement. The performance evaluation of the LLM model showed an average accuracy of approximately 83% for logicallity of AI answer with a rater agreement level (QWK) of 0.67, reflecting consistency in assessing the logicality, interpretation, and analysis of AI-generated answers, although the evaluation, explanation, and self-regulation aspects remain suboptimal in supporting students' reflective thinking abilities. Additionally, there is a weak positive correlation (r = 0.32) between model proportion of domination and the improvement of critical thinking skills, with the strongest influence observed in the interpretation aspect (72% proportion and a 0.53 N-gain). The applied learning strategies also demonstrated an increase in student satisfaction and engagement, with CSAT scores of 88% and NPS scores of 82%, although challenges such as AI responses that are less relevant and limited context windows still need to be further addressed.

	Text S_KOM_2100016_Title.pdf Download (729kB)
	Text S_KOM_2100016_Chapter1.pdf Download (274kB)
	Text S_KOM_2100016_Chapter2.pdf Restricted to Staf Perpustakaan Download (1MB)
	Text S_KOM_2100016_Chapter3.pdf Download (1MB)
	Text S_KOM_2100016_Chapter4.pdf Restricted to Staf Perpustakaan Download (4MB)
	Text S_KOM_2100016_Chapter5.pdf Download (235kB)
	Text S_KOM_2100016_Appendix.pdf Restricted to Staf Perpustakaan Download (35MB)

Item Type:	Thesis (S1)
Additional Information:	https://scholar.google.com/citations?view_op=list_works&hl=en&hl=en&user=fTWCa88AAAAJ ID SINTA Dosen Pembimbing : Erna Piantari : 6143243 Enjun Junaeti : 5992648
Uncontrolled Keywords:	Analisis Data, Critical Thinking, Fine-tuning LLM, Model Open Inquiry Learning, SLEEG. Critical Thinking, Data analysis, Fine-tuning LLM, Open Inquiry, SLEEG
Subjects:	L Education > L Education (General) Q Science > QA Mathematics > QA76 Computer software
Divisions:	Fakultas Pendidikan Matematika dan Ilmu Pengetahuan Alam > Program Studi Pendidikan Ilmu Komputer
Depositing User:	Wendi Kardian
Date Deposited:	18 Sep 2025 02:30
Last Modified:	18 Sep 2025 02:30
URI:	http://repository.upi.edu/id/eprint/139780

Actions (login required)

View Item