Chatbot Informasi dan Kalkulator Pajak Menggunakan Google Vertex AI Berbasis Retrieval-Augmented Generation (tanyapajakai.com)
DOI:
https://doi.org/10.32493/joaiia.v7i3.61172Keywords:
Chatbot, Google Vertex AI, Retrieval-Augmented Generation, Kalkulator Pajak, Regulasi PerpajakanAbstract
Akses informasi regulasi perpajakan di Indonesia sering dinilai kaku dan sulit dipahami oleh Wajib Pajak. Penelitian ini mengimplementasikan Tanyapajakai.com, aplikasi web chatbot informasi dan kalkulator pajak yang menggunakan Retrieval-Augmented Generation (RAG), Vertex AI Search, dan Gemini 3.5 Flash atas dokumen resmi peraturan perpajakan Indonesia. Evaluasi menggabungkan validasi ahli pada sepuluh skenario, BERTScore dan ROUGE pada lima pasangan jawaban acuan, serta perbandingan observasional dengan tiga konfigurasi model bahasa serbaguna. Tiga ahli perpajakan menghasilkan rata-rata akurasi faktual 4,77 dari skala 5. Reliabilitas antarpenilai mencapai ICC(2,k) = 0,47, yang menunjukkan kesepakatan sedang dengan efek batas atas yang kuat. Evaluasi kuantitatif menghasilkan skor F1 BERTScore 84,06%, ROUGE-1 60,56%, ROUGE-2 42,78%, dan ROUGE-L 53,10%. Kasus berskor rendah terutama berkaitan dengan cakupan kalkulator yang belum lengkap dan logika pajak penghasilan bulanan yang belum mutakhir. Sistem mendukung informasi perpajakan yang kontekstual dan dilengkapi tautan sumber, tetapi jawabannya tetap bersifat informatif serta perlu diverifikasi melalui sumber resmi atau konsultasi dengan tenaga profesional.
References
[1] D. Khurana, A. Koli, K. Khatter, S. Singh, and M. Rachna, Natural Language Processing: State of the Art, Current Trends and Challenges, no. Figure 1. 2022.
[2] T. B. Brown et al., “Language Models are Few-Shot Learners,” no. NeurIPS, 2020.
[3] Y. Han, A. Ceross, and J. H. M. Bergmann, “The Use of Readability Metrics in Legal Text : A Systematic Literature Review,” pp. 1–22, 2024.
[4] R. Meita and H. Kurnia, “The effects of taxpayer knowledge and taxation socialization on taxpayer compliance : The role of taxpayer awareness in developing Indonesian economy,” vol. 6, pp. 89–96, 2020, doi: 10.5267/j.ac.2019.12.004.
[5] E. Adamopoulou and L. Moussiades, “Machine Learning with Applications Chatbots : History , technology , and applications,” Machine Learning with Applications, vol. 2, no. November, p. 100006, 2020, doi: 10.1016/j.mlwa.2020.100006.
[6] Z. Ji et al., “Survey of Hallucination in Natural Language Generation,” vol. 55, no. 12, 2023.
[7] P. Lewis et al., “Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks,” no. NeurIPS, 2020.
[8] Google, “Grounding with Agent Search | Gemini Enterprise Agent Platform | Google Cloud Documentation.” Accessed: Jun. 19, 2026. [Online]. Available: https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/grounding/grounding-with-vertex-ai-search
[9] “Gemini 3.5: frontier intelligence with action.” Accessed: Jun. 23, 2026. [Online]. Available: https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-5/#gemini-3-5-flash
[10] G. D. Albert and A. Voutama, “PENGEMBANGAN CHATBOT BERBASIS PDF MENGGUNAKAN LOCAL RETRIEVAL-AUGMENTED GENERATION (RAG) DAN OLLAMA,” Jurnal Informatika dan Teknik Elektro Terapan, vol. 13, no. 2, Apr. 2025, doi: 10.23960/jitet.v13i2.6361.
[11] I. D. Saputra, N. S. Harahap, S. Agustian, M. Fikry, and L. Oktavia, “Aplikasi Web Question Answering Menggunakan Langchain OpenAI Tentang Peraturan Perundang-undangan Bidang Pendidikan,” Journal of Computer System and Informatics (JoSYC), vol. 6, no. 1, pp. 293–304, Nov. 2024, doi: 10.47065/josyc.v6i1.6182.
[12] A. Faishol, “Implementasi Fitur Chatbot Pada Website Konsultasi Jasa Pajak Online berbasis Laravel dan React JS (Studi Kasus: AMZ Tax Consultant),” Implementasi Fitur Chatbot Pada Website Konsultasi Jasa Pajak Online berbasis Laravel dan React JS (Studi Kasus: AMZ Tax Consultant), vol. 10, 2026.
[13] C. Tech, T. Zhang, V. Kishore, F. Wu, K. Q. Weinberger, and Y. Artzi, “BERTScore: Evaluating Text Generation With BERT,” pp. 1–43, 2020.
[14] M. Barbella and G. Tortora, “Rouge Metric Evaluation for Text Summarization Techniques,” SSRN Electronic Journal, 2022, doi: 10.2139/ssrn.4120317.
[15] BPK RI, “Database Peraturan | JDIH BPK.” Accessed: Jun. 19, 2026. [Online]. Available: https://peraturan.bpk.go.id/
[16] Google, “Parse and chunk documents | Agent Search | Google Cloud Documentation,” 2025. Accessed: Jun. 19, 2026. [Online]. Available: https://docs.cloud.google.com/generative-ai-app-builder/docs/parse-chunk-documents
[17] Google, “Embeddings APIs overview | Gemini Enterprise Agent Platform | Google Cloud Documentation,” 2026. Accessed: Jun. 19, 2026. [Online]. Available: https://docs.cloud.google.com/gemini-enterprise-agent-platform/models/embeddings
[18] Vercel, “Next.js Docs | Next.js.” Accessed: Jun. 19, 2026. [Online]. Available: https://nextjs.org/docs
[19] Vercel, “tRPC | tRPC.” Accessed: Jun. 19, 2026. [Online]. Available: https://trpc.io/docs
[20] A. T. Jebb, V. Ng, and L. Tay, “A Review of Key Likert Scale Development Advances : 1995 – 2019 Two Views of Validity,” vol. 12, no. May, pp. 1–14, 2021, doi: 10.3389/fpsyg.2021.637547.
