FORENSICLLM: RAQAMLI KRIMINALISTIKA UCHUN MAHALLIY (LOKAL) KATTA TIL MODELI — ARXITEKTURA, METODOLOGIYA VA SAMARADORLIK TAHLILI
Keywords:
Keywords: ForensicLLM, raqamli kriminalistika, LLaMA, RAFT, RAG, fine-tuning, LoRA, Daubert mezoni, dalil qabul qilinishi, katta til modeli.Abstract
Ushbu maqolada raqamli kriminalistika sohasida ishlab chiqilgan mahalliy (lokal) katta til modeli — ForensicLLM tizimining arxitekturaviy, metodologik va empirik jihatlari chuqur ilmiy tahlil qilinadi. AQShning Luiziana shtat universiteti tadqiqotchilari tomonidan 2025-yilda taqdim etilgan ushbu model LLaMA-3.1-8B asosida Retrieval Augmented Fine-tuning (RAFT) usuli yordamida 1082 ta ilmiy maqola va 1390 ta kriminalistik artefakt ma'lumotlari asosida sozlangan. Maqolada modelning texnik asoslari — tokenizatsiya, joylashtirish (embedding), diqqat mexanizmi (attention), kvantizatsiya hamda LoRA/QLoRA moslashtirish usullari — shuningdek uning baholash natijalari (BERTScore, BGE-M3, G-Eval metrikalari bo'yicha bazaviy modeldan ustunligi va manba ko'rsatishdagi 86,6 foizlik aniqligi) tahlil qilinadi. Alohida e'tibor foydalanuvchi tadqiqoti natijalariga, model cheklovlariga hamda AQSh Daubert mezoni asosidagi dalil qabul qilinishi talablariga qaratiladi. Xalqaro (Yevropa, Rossiya, Singapur) va milliy (O'zbekiston) kontekst bilan solishtirilgan holda, ForensicLLM'ning DFIR sohasidagi ahamiyati va istiqbollari baholanadi.
References
I. Qonunchilik hujjatlari
[1] O'zbekiston Respublikasining "Kiberxavfsizlik to'g'risida"gi Qonuni, O'RQ-764-son, 2022-yil 15-aprel (2022-yil 17-iyuldan kuchga kirgan). Manba: O'zbekiston Respublikasi qonun hujjatlari milliy ma'lumotlar bazasi — lex.uz. Havola: https://lex.uz/uz/docs/-5960604
II. Ilmiy maqolalar
[2] Sharma B., Ghawaly J., McCleary K., Webb A.M., Baggili I. ForensicLLM: A local large language model for digital forensics // Forensic Science International: Digital Investigation. — 2025. — Vol. 52, Art. 301872. DOI: 10.1016/j.fsidi.2025.301872. URL: https://www.sciencedirect.com/science/article/pii/S2666281725000113
[3] Scanlon M., Breitinger F., Hargreaves C., Hilgert J.-N., Sheppard J. ChatGPT for digital forensic investigation: the good, the bad, and the unknown // Forensic Science International: Digital Investigation. — 2023. — Vol. 46, Art. 301609.
[4] Wickramasekara A., Breitinger F., Scanlon M. Exploring the Potential of Large Language Models for Improving Digital Forensic Investigation Efficiency // arXiv preprint. — 2024. arXiv:2402.19366. URL: https://arxiv.org/pdf/2402.19366
[5] Chaudhary N.K., Iyengar S.S., Saidakhrarovich G.S. Large Language Models for Digital Forensic Investigations // Information Security, Privacy and Digital Forensics. ICISPD 2024. Lecture Notes in Electrical Engineering, vol. 1456. — Singapore: Springer, 2026. DOI: 10.1007/978-981-95-2196-8_18
[6] Hu E.J., Shen Y., Wallis P., Allen-Zhu Z., Li Y., Wang S., Wang L., Chen W. LoRA: Low-Rank Adaptation of Large Language Models // arXiv preprint. — 2021. arXiv:2106.09685.
[7] Dettmers T., Pagnoni A., Holtzman A., Zettlemoyer L. QLoRA: Efficient Finetuning of Quantized LLMs // Advances in Neural Information Processing Systems. — 2024. — Vol. 36.
[8] Zhang T., Patil S.G., Jain N., Shen S., Zaharia M., Stoica I., Gonzalez J.E. RAFT: Adapting Language Model to Domain Specific RAG // arXiv preprint. — 2024. arXiv:2403.10131.
[9] Использование больших языковых моделей для поиска угроз кибербезопасности на основе методов глубокого обучения: анализ современных исследований // КиберЛенинка (электронная научная библиотека). URL: https://cyberleninka.ru/article/n/ispolzovanie-bolshih-yazykovyh-modeley-dlya-poiska-ugroz-kiberbezopasnosti-na-osnove-metodov-glubokogo-obucheniya-analiz
[10] Kompyuter va raqamli kriminalistika sohasida xalqaro hamkorlik va qonunchilikni modernizatsiya qilish masalalari // International Journal of Science and Technology (O'zbekiston). — 2026. — B. 3–8. Havola: https://science-technology.uz/index.php/journal/article/download/695/651/1256
III. Ilmiy tezislar va dissertatsiyalar
[11] Sharma B. ForensicLLM: A Local Large Language Model for Digital Forensics // LSU Master's Theses, No. 6059. — Louisiana State University, 2024. URL: https://repository.lsu.edu/gradschool_theses/6059/
[12] Grajeda C., Sanchez L., Baggili I., Clark D., Breitinger F. Experience constructing the Artifact Genome Project (AGP): managing the domain's knowledge one artifact at a time // Digital Investigation. — 2018. — Vol. 26, S47–S58.