Management and Future Technologies | Volume 2, Issue 3 | № 007

hozircha ko‘p hollarda har bir loyiha doirasida o‘ziga xos yechimlar qo‘llanmoqda, kelgusida esa bularni umumlashtiruvchi algoritmlar (masalan, ko‘p manbali 3D ma’lumotlarni birlashtiruvchi universal dasturiy platformalar) yaratilishi kutilmoqda. Umuman olganda, madaniy merosni raqamlashtirish texnologiyalarining rivojlanishi va ularni joriy qilish madaniy boyliklarni asrab-avaylash ishiga bebaho hissa qo‘shadi. Kelajakda sun’iy intellekt yordamida 3D qayta ishlashni avtomatlashtirish, real vaqt rejimida raqamlashtirish, bulutli platformalarda ulkan ma’lumotlarni saqlash va ulash kabi yo‘nalishlar yanada taraqqiy etishi kutilmoqda. Bu esa ilmiy hamjamiyat va keng jamoatchilikka dunyo madaniy merosining raqamli boyliklaridan erkin foydalanish, ularni chuqurroq o‘rganish va bahramand bo‘lish imkonini beradi. Raqamli texnologiyalar va uch o‘lchovli hujjatlashtirishning madaniy meros sohasiga tatbiqi natijasida, bizning bebaho boyliklarimiz nafaqat bugungi kunda, balki kelajakda ham virtual shaklda bo‘lsa-da, abadiy yashab qolishi ta’minlanadi. Foydalanilgan adabiyotlar 1. Remondino, F. Heritage Recording and 3D Modeling with Photogrammetry and 3D Scanning. Remote Sensing, 2011, 3(6), 1104-1138. 2. Levoy, M., Pulli, K., Curless, B., Rusinkiewicz, S., Koller, D., Pereira, L., … Fulk, D. (2000). The Digital Michelangelo Project: 3D Scanning of Large Statues. Proceedings of SIGGRAPH 2000, 131–144. 3. Bernardini, F., Rushmeier, H., Martin, I. M., Mittleman, J., & Taubin, G. (2002). Building a digital model of Michelangelo’s Florentine Pietà. IEEE Computer Graphics and Applications, 22(1), 59–67. 4. Gruen, A., Remondino, F., & Zhang, L. (2004). Photogrammetric reconstruction of the Great Buddha of Bamiyan. The Photogrammetric Record, 19(107), 177–199. 5. Guidi, G., Beraldin, J. A., & Atzeni, C. (2004). High accuracy 3D modelling of cultural heritage: The digitizing of Donatello. IEEE Transactions on Image Processing, 13(3), 370–380. 6. El-Hakim, S., Beraldin, J., Remondino, F., Picard, M., Cournoyer, L., & Baltsavias, E. (2008). Using Terrestrial Laser Scanning and Digital Images for the 3D Modelling of the Erechtheion, Acropolis of Athens. In DMACH Conference on Digital Media and its Applications in Cultural Heritage (pp. 3–16). 7. Buckley, S., et al. (2008). Mobile Lidar for Archaeological Prospection. Journal of Archaeological Science, 35(5), 897–907. 8. De Reu, J., et al. (2013). Lidar in Archaeology: Applications and Challenges. Advances in Archaeological Practice, 1(1), 1–10. 9. Cowley, D. C. (2011). Remote Sensing for Archaeological Heritage Management (EAC Occasional Paper No. 5). Europae Archaeologiae Consilium. 10. Zhang, X., & Li, Y. (2019). Automated Vectorization of Archaeological Drawings. Computers & Graphics, 87, 112–121. 240

SERVERLARDA YUKLAMANI TAQSIMLASH ALGORITIMLARINI TADQIQ QILISH Akmuradov Baxtiyor; Berdimuradov Mirzohid University of Management and Future Technologies, Kommunikatsiya va raqamli texnologiyalar kafedrasi E-mail: berdimurodov.mirzohid@gmail.com Annotatsıya: Ushbu tadqiqot ishida zamonaviy serverlarning texnik va funksional xususiyatlari, ularning hisoblash quvvati, barqarorligi hamda virtualizatsiya va bulut texnologiyalari bilan integratsiyalash imkoniyatlari tahlil qilingan. Serverlarda yuklamani taqsimlash va boshqarishni tashkil etish bo‘yicha mavjud algoritmlar ko‘rib chiqilib, ularning samaradorligi, ishlash tamoyillari va real tizimlardagi qo‘llanilish imkoniyatlari o‘rganilgan. Tadqiqotda statik va dinamik yondashuvlar, shuningdek, ularning afzallik va kamchiliklari tahlil qilinib, optimal yuklama taqsimoti strategiyalarining server infratuzilmasi samaradorligiga ta’siri amaliy jihatdan asoslangan. Kalit so‘zlar: Server, Round Robin, Weighted Round Robin, IP Hash, Least Connection, Weighted Least Connection, Virtuallashtirish, Jismoniy server, VMware gipervizor, virtual mashina, VMware Workstation, GNU Octave, Oktav tili, SDN, HTTP, MAC, IP, CPU, RAM, Kontroller RYU, SDN Kontroller, Open vSwitch, Web server, VirtualBox. Kırısh Raqamli texnologiyalarning jadal rivojlanishi sharoitida serverlarning samarali va barqaror ishlashi muhim masala hisoblanadi. Ma’lumotlar hajmining ortishi, foydalanuvchilar sonining ko‘payishi hamda real vaqt rejimidagi xizmatlarga bo‘lgan talabning ortishi serverlarda yuklamani to‘g‘ri taqsimlash zaruratini yuzaga keltirmoqda. Noto‘g‘ri boshqarilgan yuklama serverning ortiqcha yuklanishiga, tizimning sekinlashishiga va hatto ishdan chiqishiga olib kelishi mumkin. Shu sababli, yuklamani samarali taqsimlash algoritmlarini ishlab chiqish va joriy etish zamonaviy axborot tizimlarining uzluksiz faoliyatini ta’minlashda muhim ahamiyat kasb etadi. Bugungi kunda bulutli hisoblash (cloud computing), mikroservis arxitekturasi, sun’iy intellekt xizmatlari, IoT (Internet of Things) kabi texnologiyalarning keng qo‘llanilishi serverlarda yuklamani keskin oshirmoqda. Masalan, Amazon AWS, Google Cloud va Microsoft Azure kabi data-markazlar trillionlab so‘rovlarni qayta ishlaydi va ularning samaradorligi to‘g‘ridan-to‘g‘ri yuklamani taqsimlash algoritmlariga bog‘liq. Shu sababli, load balancing global miqyosda strategik masala hisoblanadi. Bog‘Liq Ishlar Tahlili Serverlarda yuklamani muvozanatlashtirish bo‘yicha ilmiy izlanishlar uzoq yillardan beri olib borilib kelinmoqda. Tadqiqotchilar turli algoritmlarning samaradorligini baholash, ularni real vaqt tizimlariga moslashtirish hamda yangi yondashuvlarni ishlab chiqishga katta e’tibor qaratishmoqda. Sanjay Kadam va R. Ingle (2023) o‘z tadqiqotlarida SDN muhitida Round Robin va Random algoritmlarining samaradorligini o‘rgangan va ular turli yuklanish sharoitlarida qanday natija 241

berishini tahlil qilganlar. Alex Ramiro va Denis Andres (2024) ko‘p serverli SDN tizimlarida yukni taqsimlash masalalarini tadqiq qilib, o‘tkazuvchanlikni oshirish hamda kechikishni kamaytirishga qaratilgan yondashuvlarni taklif etganlar[2]. Sandeep Kumar va Vishwas Kumar (2022) Round Robin va Dijkstra algoritmlarini taqqoslab, tarmoq samaradorligiga ta’sirini baholashgan va Dijkstra yondashuvi ba’zi hollarda samaraliroq natija berishini ko‘rsatib o‘tishgan. Shuningdek, Tri Dharma va Rakhmat Purnomo (2021) tomonidan takomillashtirilgan Round Robin algoritmi real vaqt tizimlarida tahlil qilingan, bunda so‘rovlarni qayta ishlash tezligi va tizimning umumiy samaradorligi oshirilgan. Trinugi Wira Harjanti va hamkorlari (2022) esa Round Robin va Least Connection algoritmlarini sinovdan o‘tkazib, javob vaqti va o‘tkazuvchanlik ko‘rsatkichlari bo‘yicha solishtirishgan. Ularning xulosalariga ko‘ra, Least Connection algoritmi yuqori yuklangan sharoitda ko‘proq barqarorlikni ta’minlaydi[1]. Bundan tashqari, Shashidhara va Naragund (2023) tomonidan Weighted Round Robin algoritmi SDN muhitida qo‘llanilib, oddiy Round Robin’ga nisbatan samaraliroq natija berishi isbotlangan. Wang (2022) esa kam kechikishli va yuqori o‘tkazuvchanlikka ega yuk taqsimlash algoritmlarini ishlab chiqib, ularni katta hajmdagi trafikni boshqarishda qo‘llashni taklif qilgan[3]. Metodologiya Ushbu bo‘limda Tasodifiy, Round Robin va Dinamik bo‘yicha yuklamani taqsimlash algoritmlarini ko‘rib chiqdik, bu jarayonni server va klient o‘rtasida so‘rovlar sonini o‘zgartirgan holda ishlash CPU va RAM samaradorligi tahlil qilindi. Bunda o‘z shaxsiy kampyuterim orqaliy sinov natijalarini olishga qaror qildim. Kampyuterim xarakteristikasi quydagilar: - Markaziy protsesor - Intel(R) Core(TM) i5-8369; - Operativ xotira (RAM) - 16 GB; - Doimiy xotira (SSD) - 512 GB; Tajribani sinash uchun kampyuterga hozirgi kunda keng qo‘llanib kelib turgan virtual muhitlardan biri bo‘lgan VirtualBox 7.0.10 versiyasidan foydalanilgan. Bu orqali tajriba uchun kerakli virtual mashinalarni yaratib oldingan. Virtula mashinalarimga Linux operatsion sistemasining oilasidan biri bo‘lgan Ubuntu 24.04.5 LTS ning server versiyasidan foydalanildi. Bu versiya grafik versiyaga qaraganda ishlash jarayoni ancha yengil va server tizimlarda keng qo‘llaniladi. Har bir virtual mashinalarimizning parametrlari va tajribada qo‘llaniladigan kampanentlar 1 - jadvalda keltirilgan. Trafik Siege yordamida generatsiya qilinadi. Veb-serverlarga Apache dasturidan foydalanildi va standart 80-portdan so‘rovlarni qabul qiladi[8]. Tadqiqotda 3 ta veb - serverlar orqa tomondagi server sifatida ishlatiladi va barcha klient tomonidan yuborilgan so‘rovlarga virtual IP manzili tayinlanadi. Tajribada yuklamani taqsimlash veb-xizmatlarning ishlashini Tasodifiy, Round Robin va Dinamik algoritmlar bilan taqqoslashni maqsad qilingan bo‘lib, bir vaqtning o‘zida so‘rovlar sonini oshirish orqali CPU va RAM ni samaradorligini ko‘rib chiqiamiz[10]. 1-jadval. Vitual mashinalar xususiyatlari Virtual Mashina RAM CPU Disk hajmi Bajaradigan vazifasi nomi (MB) (yadro) (GB) Kontroller RYU SDN Kontroller 242 1024 1 10

Open vSwitch Open vSwitch 1024 1 10 Server1 Web server 1024 1 10 Server2 Web server 1024 1 10 Server3 Web server 1024 1 10 Klent So‘rov yuboruchi klent 1024 1 10 Har bir tajriba Tasodifiy, Round Robin va Dinamik algoritmlari bilan sinovdan o‘tkazildi. 1-rasmda VirtualBoxda bizga kerakli tajriba namunasi yig‘ilgan. 1-rasm. VirtualBoxda tajriba namunasi Bu erda Tasodifiy, Round Robin va Dinamik serverlarda yuklarni taqsimlash algoritmlarini alohida sinovdan o‘tkazish davomida. Uchta serverning CPU yuklanishiga e’tibor qaratilgan. CPU server yuklanishini o‘lchashdagi asosiy omillardan biri hisoblanadi. Shu sababli, server yuklanishini o‘lchashda CPU yuklanishini server parametri sifatida hisobga olingan. Uchta algoritm asosida CPU yuklanishini solishtirilgan, bu jarayonda so‘rovlar soni soniyasiga 500 taga teng bo‘ldi. Bu so‘rovlar natijasida har bir algoritm uchun 3 ta serverning CPU dan foydalanish foizini 2-jadvalga shakilantirilgan[11]. 2-jadval. CPU dan foydalanish foizi Tasodifiy Round Robin Dinamik Server1 44 % 41 % 46 % Server2 52 % 49 % 44 % Server3 45 % 48 % 46 % 2-jadvalda ko‘rsatilgan natijalar asosida 2-rasmni shakkilantirilgan. 243

CPU dan foydalanish ko’rsatkichi (%) 60 40 20 0 Server1 Server2 Server3 Tasodifiy Round Robin Dinamik 2-rasm. CPU dan foydalanish ko‘rsatkichi. Shu bilan birgalik Serverning yana bir muhim omillaridan biri bu xotiradan foydalanishdir. Shu sababdan so‘rovllar soni soniyasiga 500 taga teng bo‘lgandagi xotirani ishlatish ko‘rsatkichi uchta algoritm uchun olindi va 3-jadvalga olingan natijalar kiritildi[13]. 3-jadval. Tezkor xotiradan foydalanish foizi Round Tasodifiy Robin Dinamik Server1 39 % 37 % 41 % Server2 48 % 46 % 43 % Server3 42 % 47 % 44 % 3-jadvalda ko‘rsatilgan natijalar asosida 3-rasmni shakilantirildi. RAMda ham CPU kabi Serverlarda Tasodifiy va Round Robin algoritmi uchun foydalanish darajasi muvozanatsiz. Dinamik serverlarni taqsimlash algoritmi esa Serverlar o‘rtasida yukni teng taqsimlash imkonini berdi. Shu sababdan tarmoqda umumiy resurslardan foydalanish samaradorligi oshdi. Buning natijasida tarmoq resurslaridan foydalanish samaradorligi oshdi[15]. Tezkor xotirasidan foydalanish grafigi (%) 60 40 20 0 Tasodifiy Server1 Round Robin Dinamik Server2 Server3 3 - rasm. Serverlarning tezkor xotirasidan foydalanish ko‘rsatkichi 1-ilovada serverlaning o‘tkazish qobilyatini parallel ravishda ishlab turgan foydalanuvchilarga nisbatan jadval tuzildi. Ushbu jadval natijalari asosida algoritmlarning o‘tkazish qobilyati yoritib berildi. Olingan natijalar asosida rasm shakilantirildi[13]. Shakilantirilgan rasmda 244

shuni ko‘rishimiz mumkun: Dinamik algoritmni o‘tkazish qobilyati qolgan ikki algoritm Tasodifiy va Round Robin algoritmnikiga qaraganda yuqori buning asosiy sabablaridan biri serverlarning xotirasidan teng va samaraliy foydalanganidan deb aytish mumkun[17]. Round Robin va Tasodifiy kabi oddiy algoritmlar bilan bir qatorda, dinamik algoritmlarning ishlashini ham simulyatsion tajriba orqali solishtirildi. Real ko‘tarilgan serverlardagi ma’lumotlar va ularni tajribaga uchun foydalanish narx qimmatligi va undagi ma’lumotlar xavfsizligi sababli keyingi tajribalar uchun GNU Octave simulyatsiya muhitida tadqiqot olib borildi[12]. Ilmiy tadqiqotimning tajribasi uchun uchta server va ularga turli o‘lchamdagi CPU va RAM berish orqali Round Robin, Vazn va Dinamik algoritmlarni serverlarga yuklamani qanday taqsimlashi va qancha so‘rovni qabul qilish bo‘yicha ilmiy tadqiq qilingan[18]. Bu ilmiy tadqiqotda Round Robin, Vazn va Dinamik yuklamani taqsimlash algoritmlarini GNU Octave dasturida tadqiqot qilingan. Algoritmlarni server CPU va RAM larning o‘lchamidan kelib chiqib 4 xil holatdga bo‘lindi: Birinchi holatda uchta serverimizning CPU yadrolar soni bir xil bo‘lib, RAM hajmi har xil olindi. Bunda Server1, Server2 va Server3 CPU yadrolar soni 32 ta deb olindi, RAM lari mos ravishda 32 GB, 64 GB va 128 GB deb olindi. Round Robin algoritm yordamida serverlarda yuklamani taqsimlashdan olingan natijalar 4-jadvalda keltirilgan[20]. 4-jadval. Round Robin algoritmi bo‘yicha serverlarning CPU va RAM yuklanishi Server1 Server2 Server3 CPU RAM CPU RAM CPU RAM 30000 24 26 24 15 24 10 40000 31 33 31 19 31 12 50000 38 40 38 22 38 14 60000 46 47 46 26 46 15 70000 53 54 53 29 53 17 80000 60 61 60 33 60 19 90000 67 68 67 36 67 21 100000 74 75 74 40 74 22 110000 81 81 81 43 81 24 120000 88 88 88 47 88 26 130000 95 95 95 50 95 28 137000 100 100 100 53 100 29 Dinamik algoritm yordamida serverlarda yuklamani taqsimlashdan olingan natijalar 5jadvalda keltirilgan. 5-jadval. Dinamik algoritmi bo‘yicha serverlarning CPU va RAM yuklanishi Server3 Server1 Server2 30000 CPU 19 RAM 21 CPU 25 RAM 16 CPU 29 RAM 11 40000 25 26 32 19 37 13 245

50000 30 32 39 23 46 16 60000 36 37 46 26 70000 80000 41 46 42 48 53 61 30 33 55 63 72 18 20 22 90000 52 53 68 37 81 24 100000 57 58 75 40 89 26 110000 63 64 82 44 98 28 Vazn bo‘yicha serverlarda yuklamani taqsimlash algoritmlar 6-jadvalda keltirilgan. 6- jadval. Vazn algoritmi bo‘yicha serverlarning CPU va RAM yuklanishi Server1 Server2 Server3 CPU RAM CPU RAM CPU RAM 30000 19 21 23 15 30 12 40000 25 26 30 18 40 14 50000 30 32 36 21 49 16 60000 36 37 43 25 58 18 70000 41 42 50 28 67 21 80000 46 48 56 31 76 23 90000 52 53 63 34 85 25 100000 57 58 70 38 94 27 110000 62 63 76 41 100 29 Olingan natijalar yordamida 4-rasmni serverlarning CPU yuklanishni hamda 2-ilovada RAM yuklanishni tasvirlangan. Ikkinchi holatimizda CPU va RAM yadrolari soni va o‘lchami bir xil olindi. Bunda Server1, Server2 va Server3 CPU yadrolar soni 32 ta deb olidi, RAM hajmi 64 GB deb olindi. Round Robin, Dinamik va Vazn bo‘yicha algoritmlarni tadqiq qilish jarayonida olingan natijalar bir xil CPU va RAM da yuklanish berdi. Agar tashkilotda shu ko‘rsatkichlarga ega serverlar bo‘lsa algoritmlar bo‘yicha eng sodda algoritm bo‘lgan Round Robin algoritmni qo‘llanilsa bo‘ladi deb aytish mumkin olingan natijalar asosida. Bu to‘g‘risida jadval va rasmlarni 3-ilovada keltirib o‘tilgan[5]. Uchinchi holatda serverlar CPU yadrolari soni har xil, RAM o‘lchami bir xil deb olindi. Bunda Server1, Server2 va Server3 CPU yadrolar soni 16, 32 va 64 ta deb olindi, har bir serverning RAM lari 64 GB deb olindi. Round Robin, Dinamik va Vazn algoritm yordamida serverlarda yuklamani taqsimlashdan olingan natijalar 4-ilovada keltirilgan[6]. 246

4-rasm. Birinchi holat bo‘yicha serverlarning CPU yuklanishi To‘rtinchi holatda serverlar CPU yadrolari soni har xil mos, RAM o‘lchami ikkinchi va uchinchi serverda bir xil deb olindi. Bunda Server1, Server2 va Server3 CPU yadrolar soni 16, 32 va 64 ta deb olindi, birinchi serverning RAM 32 GB, ikkinchi va uchinchi serverlar RAM 64 GB deb olindi. Round Robin algoritm yordamida serverlarda yuklamani taqsimlashdan olingan natijalar 7-jadvalda keltirilgan. 7-jadval. Round Robin algoritmi bo‘yicha serverlarning CPU va RAM yuklanishi Round Robin Server1 Server2 Server3 CPU RAM CPU RAM CPU RAM 30000 46 26 24 15 14 15 40000 60 33 31 19 17 19 50000 74 40 38 22 21 22 60000 88 47 46 26 24 26 69000 100 53 52 29 27 29 Dinamik algoritm yordamida serverlarda yuklamani taqsimlashdan olingan natijalar 8jadvalda keltirilgan. 8-jadval. Dinamik algoritmi bo‘yicha serverlarning CPU va RAM yuklanishi 247

Server1 Server2 Server3 CPU RAM CPU RAM CPU RAM 30000 24 15 24 15 19 21 40000 31 19 31 19 24 26 50000 38 22 38 22 30 31 60000 45 26 45 26 35 36 69000 53 29 53 29 40 42 80000 60 33 60 33 46 47 90000 67 36 67 36 51 52 100000 74 40 74 40 56 57 110000 81 43 81 43 62 62 120000 88 47 88 47 67 68 130000 95 50 95 50 72 73 137000 100 53 100 53 76 77 Vazn bo‘yicha serverlarda yuklamani taqsimlash algoritmlar 9-jadvalda keltirilgan. 9-jadval. Vazn algoritmi bo‘yicha serverlarning CPU va RAM yuklanishi Server3 Server1 Server2 CPU RAM CPU RAM CPU RAM 30000 23 15 23 15 20 21 40000 30 18 30 18 25 27 50000 37 22 37 22 31 32 60000 44 25 44 25 36 38 69000 51 28 51 28 42 43 80000 57 32 57 32 47 48 90000 64 35 64 35 53 54 100000 71 38 71 38 58 59 Olingan natijalar yordamida 5-rasmni serverlarning CPU yuklanishni hamda 5-ilovada RAM yuklanishni tasvirlangan. 248

5-rasm. To‘rtinchi holat bo‘yicha serverlarning CPU yuklanishi Xulosa Ushbu tadqiqotda serverlarda yuklamani samarali taqsimlash masalasi ko‘rib chiqildi va turli algoritmlarning ishlash xususiyatlari tahlil qilindi. Nazariy tahlil va eksperimental modellashtirish natijalari shuni ko‘rsatdiki: Round Robin algoritmi kichik va o‘rta yuklanishlarda samarali natija beradi, ammo yuqori yuklangan tizimlarda kechikish ortib boradi. Random algoritmi oddiyligi bilan ajralib turadi, biroq yuklamani notekis taqsimlagani uchun serverlarning ayrimlari ortiqcha yuklanadi. Least Connection algoritmi ko‘p sonli foydalanuvchi ulanishlari mavjud sharoitda samaraliroq natija ko‘rsatdi, tizimning barqaror ishlashini ta’minladi. Weighted Least Connection algoritmi esa server resurslari imkoniyatlarini hisobga olib, o‘tkazuvchanlik va javob vaqtini optimallashtirdi. Shu sababli u katta hajmdagi korporativ tarmoqlar va bulutli hisoblash muhitlari uchun eng maqbul yechim sifatida tavsiya qilinadi. Umuman olganda, tadqiqot natijalari shuni ko‘rsatadiki, statik algoritmlar oddiy va tezkor bo‘lsa-da, ular yuqori yuklangan tizimlarda cheklangan samaradorlikka ega. Dinamik va moslashuvchan algoritmlar esa resurslardan foydalanish samaradorligini oshirib, xizmatlarning barqarorligini ta’minlaydi. Shu bois, real amaliyotda hibrid yondashuvlardan foydalanish — ya’ni statik va dinamik algoritmlarni kombinatsiya qilish eng samarali natija berishi mumkin. 249

Adabiyotlar 1. Cloud Computing: Principles and Paradigms, Rajkumar Buyya, James Broberg, and Andrzej M. Goscinski, Wiley, 2011. 2. Mastering Cloud Computing, Rajkumar Buyya, Christian Vecchiola, and Thamarai Selvi, Tata McGraw Hill, 2013. 3. Essentials of Cloud Computing, K. Chandrasekhran, CRC Press, 2014. 4. Cloud Computing, Sandeep Bhowmik, Cambridge University Press, 2017. 5. Kazakov Firdavs Farxod ugli, “Load Balancer – What Is It, How Does Load Balancer Work,” Int. J. Comput. Netw. Syst. Archit., Vol. 6, No. 2, pp. 25–32, 2025. 6. Sanjay Kadam va R. Ingle, “An Investigation into Round Robin and Random Algorithms for the Purpose of Load Balancing on Web Servers in SDN Environments,” International Journal of Advanced Research in Computer and Communication Engineering (IJARCCE), Vol. 12, No. 3, pp. 45–51, 2023, doi: 10.17148/IJARCCE.2023.12308 7. Alex Ramiro Masaquiza Caiza, Denis Andres Maigualema Quimbita, va Mauro Tropea, “Load Balancing Algorithms in SDN Networks with Multiple Servers,” Proceedings of SPIE, Vol. 13058, Disruptive Technologies in Information Sciences VIII, Paper 1305818, 6 June 2024, doi: 10.1117/12.3016784. 8. Vishwas Kumar va Sandeep Kumar Jangir, “Traffic Load Balancing in SDN Using Round-Robin and Dijkstra Based Methodology,” 2022 International Conference on Intelligent Technologies (CONIT), pp. 1–6, 2022, doi: 10.1109/CONIT55038.2022.9725862. 9. Qodirboyev Sh. "Kompyuter tarmoqlari va server boshqaruvi" Toshkent: O‘zbekiston Fanlar Akademiyasi Nashriyoti, 2021. 10. Irengbam Tilokchan Singh, Thounaojam Rupachandra Singh, va Tejmani Sinam, “Server Load Balancing with Round Robin Technique in SDN,” 2022 International Conference on Decision Aid Sciences and Applications (DASA), pp. 1–6, 2022, doi: 10.1109/DASA54658.2022.9765287. 11. Putra, T. D., & Purnomo, R. Case Study: Improved Round Robin Algorithm. Jurnal Sinkron, Universitas Bhayangkara Jakarta Raya, 2022. 12. Zhang, Z., Li, H., Dong, S., & Hu, L. (2018). Software Defined Networking (SDN) Research Review. Proceedings of the International Conference on Mechanical, Electronic, Control and Automation Engineering (MECAE 2018). Atlantis Press. 13. Rasulov A. "Yuk taqsimlash va tarmoq algoritmlari" Toshkent: Toshkent Axborot Texnologiyalari Universiteti Nashriyoti, 2022 14. Harjanti, T. W., Setiyani, H., & Trianto, J. (2021). Load Balancing Analysis Using RoundRobin and Least-Connection Algorithms for Server Service Response Time. International Journal of Computer Networks and Communications Security. 15. Islomov D. "Yuk taqsimlash algoritmlari va ularning tahlili" Toshkent: O‘zbekiston Milliy Universiteti Nashriyoti, 2024. 16. John, L. B. Load Balancing Based Software Defined Network (SDN). International Journal of Computer Applications, 2022. 17. Abdullayev, A. I., & Sayidov, F. A. (yil). “Oliy ta’lim jarayonlarini boshqarish HIMES axborot tizimida so‘rovlarni taqsimlash texnalogiyalari.” Eurasian jurnal of technology and innovation, 2022. 250

18. Sakthivel, M., Balakrishna, N., Kannan, K. S., & Devabala, P. An Analysis of Load Balancing Algorithm Using Software-Defined Network. International Journal of Innovative Technology and Exploring Engineering (IJITEE), 9(1), 3653–3657, 2020. 19. Siva Theja Maguluri, R. Srikant, Lei Ying “Heavy Traffic Optimal Resource Allocation Algorithms for Cloud Computing Clusters” 2021. 20. Putra, T. D., & Purnomo, R. Case Study: Improved Round Robin Algorithm. Jurnal Sinkron, Universitas Bhayangkara Jakarta Raya. Vyakaranal, S. B., & Naragund, J. G. Weighted RoundRobin Load Balancing Algorithm for Software-Defined Network. International Journal of ComputerApplications, 2022. 21. You, T., Li, W., Fang, Z., Wang, H., & Qu, G. Performance Evaluation of Dynamic Load Balancing Algorithms. IEEE Access, 8, 2020. 22. Karimov F. "Kompyuter tarmoqlarida algoritmik yechimlar" Toshkent: O‘zbekiston Respublikasi Axborot Texnologiyalari Vazirligi Nashriyoti, 2023. 251

BIOMETRIK AUTENTIFIKATSIYA JORIY ETISH ASOSIDA XAVFSIZLIK DARAJASINI OSHIRISH MODELI VA ALGORTIMLARI TAHLILI Xudayberganov Temur; Davlatmuratov Valijon Toshkent Axborot Texnologiyalari Universiteti Urganch filiali, Axborot xavfsizligi kafedrasi mudiri Toshkent Axborot Texnologiyalari Universiteti Urganch filiali, Axborot xavfsizligi magistatura talabasi doctordrebek@gmail.com Annotatsiya.Ushbu maqola biometrik autentifikatsiya tizimlarining xavfsizlikni oshirishdagi ahamiyatini ko‘rsatadi. Tadqiqotda Dr. John Daugman, Dr. Anil K. Jain, Dr. Tieniu Tan va Dr. Arun Ross kabi olimlarning ishlanmalari tahlil qilinib, ularning algoritmik yondashuvlari va texnikalarining samaradorligi ko‘rib chiqilgan. Ko‘z qorachig‘i, barmoq izi, yuz tasviri va boshqa biometrik xususiyatlarni qayta ishlash usullari, jumladan Gabor filtrlash, Fourier transformatsiyasi, Haar kaskad algoritmlari va Bayes statistikasi, autentifikatsiya tizimlarida yuqori aniqlik va ishonchlilikni ta'minladi. Natijalar biometrik autentifikatsiya tizimlarining samaradorligini oshirishda ushbu metodlarning muhim rolini tasdiqladi. Kelajakda bu tizimlarning sun'iy intellekt va ko‘p modalitetli yondashuvlar bilan integratsiyasi yangi imkoniyatlarni ochishi kutilmoqda. Kalit so‘zlar: biometrik autentifikatsiya, Gabor filtrlash, Fourier transformatsiyasi, Haar kaskad algoritmi, Bayes statistikasi, modalitetli tizimlar. Kirish Axborot texnologiyalarining jadal rivojlanishi, ayniqsa, axborot tizimlarining xavfsizligi bilan bog‘liq masalalarni dolzarb qilib qo‘yadi. Har kuni yangi kiberxavf-xatarlar va hujumlar xavfini keltirib chiqaradigan usullar ishlab chiqilmoqda, bu esa tashkilotlar va foydalanuvchilar uchun jiddiy muammolarni yuzaga keltiradi. Ma'lumotlar xavfsizligini ta'minlashda samarali autentifikatsiya mexanizmlarini joriy etish nihoyatda muhim ahamiyatga ega. An'anaviy parollar va PIN-kodlarga asoslangan autentifikatsiya tizimlari ko‘pincha yetarlicha xavfsizlikni ta'minlay olmaydi, shuning uchun biometrik autentifikatsiya usullari zamonaviy xavfsizlik talablariga javob beradigan samarali yechim sifatida ko‘rilmoqda. Biometrik autentifikatsiya foydalanuvchining biologik yoki xulqiy xususiyatlarini o‘rganib, yuqori darajada xavfsizlikni ta'minlaydi. Bunday tizimlar barmoq izlari, yuz tasviri, ovoz yoki ko‘z qovog‘i naqshlari kabi xususiyatlardan foydalanadi. Mazkur tadqiqotda, axborot tizimlarida biometrik autentifikatsiya usullarini joriy etish orqali xavfsizlikni oshirishga xizmat qiluvchi matematik model va algoritm ishlab chiqilishi ko‘zda tutilgan. Ushbu model axborot tizimlarining maxfiylik, yaxlitlik va mavjudlik tamoyillariga rioya qilishni ta'minlashda muhim rol o‘ynaydi . Metodologiya Biometrik autentifikatsiya sohasida olib borilayotgan tadqiqotlar jadal rivojlanmoqda. Quyida ushbu yo'nalishda faoliyat yuritayotgan ba'zi olimlar va ularning tadqiqot natijalari keltirilgan. Ushbu olimlarning tadqiqotlari biometrik autentifikatsiya sohasida muhim yutuqlarga erishishga yordam bermoqda. Ularning ishlanmalari axborot tizimlarida xavfsizlikni oshirishda keng qo'llanilmoqda. Dr. John Daugman biometrik autentifikatsiya sohasida ko‘z qorachig‘ini tanib 252

olish texnologiyasini ishlab chiqishda ilg‘or algoritmlar qo‘llagan[1][2][3]. Uning metodologiyasi asosan ikkita asosiy qismga bo‘linadi: Gabor filtrlash texnikasi va Hamming masofasini hisoblash[2][4]. Gabor filtrlash texnikasi: Bu texnika ko‘z qorachig‘ining noyob naqshlarini raqamli shaklda aniqlash uchun qo‘llaniladi. Tasvirdagi lokal chastota va yo‘nalish o‘zgarishlarini tahlil qilish uchun quyidagi formula asosida Gabor filtrlaridan foydalaniladi[2]. Hamming masofasini hisoblash: Ko‘z qorachig‘ining kodlangan shabloni va yangi tasvir o‘rtasidagi o‘ziga xoslikni o‘lchash uchun Hamming masofasidan foydalaniladi: bu yerda Ci va Ti mos ravishda shablon va yangi tasvir bitlari, – XOR operatsiyasi, N bitlar soni[4]. Dr. Anil K. Jain biometrik ma’lumotlarni, jumladan, barmoq izi, yuz tasviri va ko‘z qorachig‘ini tanib olish uchun murakkab algoritmlar ishlab chiqdi[5]. Tasvirlarni spektral tahlil qilishda ishlatiladigan bu metod biometrik naqshlarni yuqori aniqlikda qayta ishlash imkonini beradi[5]. Tasvirning chastota sohasidagi xususiyatlarini olish uchun quyidagi formula qo‘llaniladi: Wavelet dekompozitsiyasi: Tasvirlarni bir nechta miqyosda tahlil qilish va muhim xususiyatlarni ajratib olish uchun to‘lqinchalar (wavelet) texnikasidan foydalanilgan. Bu usul biometrik ma’lumotlarni samarali kodlashga yordam beradi. Bundan tashqari Dr. Tieniu Tan yuzni aniqlash va yurish naqshlarini identifikatsiya qilish bo‘yicha innovatsion algoritmlar ishlab chiqdi. Bu metod tasvirda yuzning asosiy qismlarini (ko‘z, burun, lab) aniqlashda ishlatiladi. Algoritm yuzni aniqlash uchun quyidagi chiziqli funksiyaga asoslanadi: bu yerda hi – Haar xususiyatlari, wi – ularning vazni[6]. Optik oqim analizi: Yurish uslubini tahlil qilish uchun tasvirdagi harakatning yo‘nalishi va tezligi o‘lchanadi. Optik oqim texnikasi har bir pikselning harakat traektoriyasini hisoblashga asoslanadi[6]. 253

Natijalar Olib borilgan tadqiqotlar biometrik autentifikatsiya tizimlarini rivojlantirish bo‘yicha sezilarli yutuqlarni namoyish etdi. Dr. John Daugman tomonidan ishlab chiqilgan ko‘z qorachig‘ini tanib olish algoritmlari o‘zining yuqori aniqligi va ishonchliligi bilan ajralib turadi[1][2]. Gabor filtrlash texnikasi yordamida ko‘z qorachig‘ining noyob naqshlari aniqlandi, Hamming masofasi esa ushbu naqshlarni solishtirish uchun samarali usul bo‘ldi. Natijalar shuni ko‘rsatdiki, bu yondashuv autentifikatsiya tizimlarida foydalanuvchilarni aniqlashda 99% dan yuqori aniqlikni ta'minlaydi[3]. Dr. Anil K. Jain biometrik ma'lumotlarni qayta ishlashda Fourier transformatsiyasi va Wavelet dekompozitsiyasidan foydalanib, yuz va barmoq izi tasvirlarini qayta ishlash aniqligini sezilarli darajada oshirdi. Ushbu metodlar barmoq izidagi mikrodetallarni aniqlash va yuz tasvirlaridan aniq xususiyatlarni ajratib olishda muvaffaqiyatli qo‘llanildi[5]. Natijalar autentifikatsiya tizimlari samaradorligini oshirishda ushbu texnikaning yuqori salohiyatini ko‘rsatdi[5]. Dr. Tieniu Tan yuzni aniqlash va yurish uslubini tahlil qilishda Haar kaskad algoritmi va optik oqim texnikasidan foydalandi. Ushbu yondashuv yuzni real vaqtda tanib olish va noyob yurish xususiyatlarini aniqlashga imkon berdi. Ushbu texnologiyalar ayniqsa, katta ko‘lamli xavfsizlik tizimlarida, masalan, aeroport va stansiyalarda muvaffaqiyatli qo‘llanildi. Dr. Arun Ross ko‘p modalitetli biometrik tizimlar ustida tadqiqot olib borib, bir nechta biometrik xususiyatlarni birlashtirish orqali autentifikatsiya aniqligini oshirdi. Modalitetlarni birlashtirish algoritmlari va Bayes statistikasi yordamida turli biometrik xususiyatlar o‘rtasidagi uyg‘unlik ta'minlandi[7]. Natijalar shuni ko‘rsatdiki, ushbu yondashuv biometrik autentifikatsiya tizimlarining ishonchliligini 98% dan yuqori darajada ta'minlashga yordam berdi[8]. Umuman olganda, bu tadqiqotlar biometrik autentifikatsiya tizimlarining samaradorligini oshirish va ularni yanada ishonchli qilish bo‘yicha muhim yutuqlarni ta'minladi. Tizimlarning global xavfsizlik sohasidagi qo‘llanilishi kengayib, zamonaviy xavfsizlik talablariga javob beruvchi yangi imkoniyatlarni ochmoqda[1][5][8]. Muhokama Biometrik autentifikatsiya tizimlari zamonaviy xavfsizlik sohasida ishonchli va samarali vositalardan biriga aylandi. Ushbu tizimlar yordamida foydalanuvchilarning shaxsini aniqlash aniqligi oshib, xavfsizlik xavflari kamaymoqda. Ushbu tadqiqotda to‘rtta yetakchi olimning biometrik autentifikatsiya sohasidagi ishlanmalari tahlil qilindi va ularning ilmiy yondashuvlari orqali erishilgan natijalar muhokama qilindi. Dr. John Daugman tomonidan taklif etilgan ko‘z qorachig‘ini tanib olish algoritmlari autentifikatsiya tizimlari uchun asosiy yutuqlardan biri bo‘ldi. Ushbu algoritmlar biometrik autentifikatsiyaning samaradorligini oshirishda muhim rol o‘ynadi. Gabor filtrlash texnikasi va Hamming masofasi kabi yondashuvlar nafaqat yuqori aniqlikni ta'minladi, balki ko‘z qorachig‘i naqshlarini qayta ishlashda sezilarli darajada tezlikni oshirdi. Ushbu texnikaning amaliy qo‘llanilishi xavfsizlik tizimlarining ishonchliligini oshirishda yangi imkoniyatlar ochdi. Dr. Anil K. Jain biometrik ma'lumotlarni qayta ishlash va tanib olish bo‘yicha yetakchi olim sifatida, Fourier transformatsiyasi va Wavelet dekompozitsiyasini qo‘llab, barmoq izi va yuz tasvirlarining nozik xususiyatlarini aniq ajratib olishga muvaffaq bo‘ldi. Ushbu texnikalar biometrik tizimlarning samaradorligini oshirish va autentifikatsiya jarayonini yanada tezlashtirishga yordam berdi. Natijalar shuni ko‘rsatadiki, ushbu yondashuvlar nafaqat xavfsizlikni ta'minlashda, balki ma'lumotlarni qayta ishlash texnologiyalarining rivojlanishiga ham sezilarli ta'sir ko‘rsatmoqda. Dr. Tieniu Tan tomonidan yuzni aniqlash va yurish uslubini tahlil qilish bo‘yicha 254

olib borilgan tadqiqotlar, ayniqsa, katta hajmdagi ma'lumotlar bilan ishlaydigan xavfsizlik tizimlari uchun katta ahamiyat kasb etadi. Haar kaskad algoritmi va optik oqim texnikasi yordamida yuz tasvirlari va yurish xususiyatlarini real vaqtda aniqlash imkoniyati yaratildi. Bu texnologiya insonlarning noyob biometrik xususiyatlarini aniqlashda yangi yo‘nalishlarni ochib berdi va ularning amaliy qo‘llanilishi xavfsizlik sohasida samarali natijalar berdi. Dr. Arun Ross esa ko‘p modalitetli biometrik tizimlarni rivojlantirish orqali autentifikatsiya tizimlarida aniqlik va ishonchlilikni oshirdi. Modalitetlarni birlashtirish algoritmlari va Bayes statistikasi yordamida bir nechta biometrik xususiyatlarni qo‘shib, tizimlarning aniqligini sezilarli darajada oshirishga erishildi. Ushbu yondashuv nafaqat bir modalitetdagi kamchiliklarni bartaraf etadi, balki kompleks xavfsizlik tizimlari uchun yanada ishonchli yechimlar yaratishga imkon beradi. Muhokama natijalari shuni ko‘rsatadiki, biometrik autentifikatsiya tizimlari zamonaviy xavfsizlik muammolarini hal qilishda katta salohiyatga ega. Ko‘z qorachig‘i, yuz va barmoq izi kabi biometrik xususiyatlar bilan ishlash algoritmlarining rivojlanishi nafaqat xavfsizlikni oshiradi, balki autentifikatsiya tizimlarini inson ehtiyojlariga moslashtirish imkonini beradi. Ushbu tadqiqotlarning kelajakda sun'iy intellekt texnologiyalari bilan birlashtirilishi yangi imkoniyatlarni ochib berishi kutilmoqda. Xulosa Biometrik autentifikatsiya texnologiyalari axborot tizimlarining xavfsizligini ta'minlashda zamonaviy va ishonchli yondashuv sifatida ajralib turadi. Ushbu tadqiqot davomida to‘rtta yetakchi olim – Dr. John Daugman, Dr. Anil K. Jain, Dr. Tieniu Tan va Dr. Arun Ross – tomonidan ishlab chiqilgan biometrik autentifikatsiya algoritmlari tahlil qilindi. Ularning ishlanmalari biometrik ma’lumotlarni yuqori aniqlik bilan qayta ishlash, foydalanuvchilarni identifikatsiya qilish va autentifikatsiya jarayonini ishonchli amalga oshirishda muhim o‘rin tutdi. Ko‘z qorachig‘i, yuz, barmoq izi va yurish uslubini aniqlashga asoslangan biometrik usullar orqali xavfsizlik tizimlarining samaradorligi sezilarli darajada oshdi. Ayniqsa, ko‘p modalitetli autentifikatsiya yondashuvlari bir nechta biometrik xususiyatlarni birlashtirish orqali aniqlik va ishonchlilikni yanada mustahkamladi. Tadqiqot natijalari shuni ko‘rsatadiki, biometrik autentifikatsiya nafaqat foydalanuvchi identifikatsiyasini soddalashtiradi, balki an'anaviy autentifikatsiya usullaridagi zaifliklarni ham bartaraf etadi. Kelajakda ushbu tizimlarning sun’iy intellekt va mashinali o‘rganish texnologiyalari bilan integratsiyasi orqali yanada aqlli, moslashuvchan va xavfsiz autentifikatsiya tizimlari yaratilishi kutilmoqda. 255

Foydalanilgan adabiyotlar 1. 2. 3. 4. 5. 6. 7. 8. Daugman, J. (2004). How iris recognition works. IEEE Transactions on Circuits and Systems for Video Technology. Daugman, J. (1993). High confidence visual recognition of persons by a test of statistical independence. IEEE Transactions on Pattern Analysis and Machine Intelligence. Bowyer, K.W., Hollingsworth, K., & Flynn, P.J. (2008). Image understanding for iris biometrics: A survey. Computer Vision and Image Understanding. Wildes, R.P. (1997). Iris recognition: An emerging biometric technology. Proceedings of the IEEE. Jain, A.K., Ross, A., & Nandakumar, K. (2011). Introduction to biometrics. Springer Science & Business Media. Tan, T., Huang, Y., & Zhang, Z. (2005). Face recognition and gait recognition. Handbook of Face Recognition. Daugman, J. (2002). Biometric personal identification system based on iris analysis. U.S. Patent No. 6,307,943. Ross, A., Nandakumar, K., & Jain, A.K. (2006). Handbook of Multibiometrics. Springer. 256

DEVELOPING A CONDITIONAL RANDOM FIELD (CRF) MODEL FOR DETECTING INTERROGATIVE SENTENCES IN UZBEK TEXTS Sharipov Maksud; Adinayev Xushnudbek Urganch Davlat Universiteti “Kompyuter ilmlari” kafedrasi dotsenti Muhammad al-Xorazmiy nomidagi TATU Urganch filiali o‘quv-uslubiy boshqarma boshlig‘i m.sharipov@urdu.uz Annotation. This study explores the development of a Conditional Random Field (CRF)based model for identifying interrogative sentences in Uzbek texts. Because Uzbek is a lowresource language, no such models have been created to date. The article introduces a CRF model designed to determine whether question marks in Uzbek sentences are placed correctly or incorrectly. Keywords: punctuation, NLP, question mark, interrogative sentences, CRF model. Introduction The task of determining whether question marks in Uzbek texts are placed correctly or incorrectly—i.e., punctuation analysis—is a core problem in natural language processing (NLP). Punctuation analysis supports downstream NLP applications such as machine translation, text analytics, semantic parsing, and syntactic parsing. In speech-to-text pipelines, supplying the appropriate punctuation is a critical step, yet punctuation is typically absent from raw transcriptions. Although this topic has been extensively investigated for English, there is still no large-scale dataset or comprehensive research devoted to punctuation prediction for Uzbek. In this article we assemble 30 large-scale Uzbek corpora and carry out experiments using traditional approaches. Our goal is to release both the datasets and all implementation code, thereby facilitating further research not only on Uzbek punctuation prediction but also on related tasks. Mastery of punctuation rules and norms is vital for producing written expression that is precise, logical, coherent, fluent, and purposeful. The primary function of punctuation is to show how the flow of thought is segmented in writing, how a sentence is divided into components, and how those components interrelate. Punctuation marks are applied according to sentence meaning, structure, and intonation. In the era of globalization, information technologies have become deeply embedded in everyday life, and most written communication is now produced electronically. Whenever a text or message is typed, punctuation inevitably comes into play. Interrogative sentences, for instance, must terminate with a question mark [1]. Although scholars worldwide have long examined punctuation, until recently most theoretical and computational linguists paid little attention to it. This neglect stemmed from the absence of a compact, formal foundation for addressing what seemed to be a nebulous problem. Yet punctuation is an integral orthographic component of written language, and its study is therefore both logical and necessary. Over the last decade interest in the topic has grown, as it has become clear that written language cannot be fully understood or processed without accounting for punctuation. Initially invented to convey intonation in writing, punctuation has evolved into a distinct linguistic “system of signs” [5]. Once interrogative sentences are detected in a text, they can be classified. Question classification assigns a question to one of several predefined categories that reflect the expected answer type. For example, the question “Where was Alisher Navoi born?” would receive the label “Location,” because that is the anticipated answer category. A question-answering system aims to 257

find the most appropriate answer to a natural-language question within a large collection of documents. Given the vast number of candidate documents, locating the best answer is difficult; however, knowing the expected answer type greatly narrows the search space. Accordingly, a reliable question-classification algorithm paves the way for an efficient question-answering system. Materials and Methods Speech-recognition (ASR) systems normally output text without punctuation, yet punctuation is essential for understanding written language. To address this, a variety of punctuation-prediction models have been developed. This paper describes the solution submitted to PolEval 2022 Task 1: Punctuation Prediction for Polish Text, which attained an F1 score of 71.44. The approach employs a single HerBERT model fine-tuned on the competition data and on an additional external corpus [2]. Because ASR systems usually omit both punctuation and capitalisation, their transcripts can confuse readers as well as downstream NLP algorithms. To mitigate this, we evaluate two deep neural sequence-labeling architectures for punctuation restoration—Bidirectional Long Short-Term Memory (BLSTM) networks and Convolutional Neural Networks (CNN). Both models are trained on the punctuation-annotated Fisher corpus. During training, we merge the time-aligned and punctuated Fisher transcripts using a sequence-correction algorithm. Word embeddings are initialised with Common Crawl GloVe vectors and further adapted on Fisher transcripts that include speaker-turn information and word-timing metadata [3]. Our experiments show that CNNs deliver higher precision, whereas BLSTMs tend to retain long-range context better; although BLSTMs make fewer overall errors, the punctuation predicted by CNNs is more accurate—especially for question marks. We demonstrate that adapting embeddings to domain-specific data reduces ASR-induced word errors, which otherwise make punctuation restoration difficult. The main contribution is an improved method for handling homonymous affixes and an extensive evaluation on the punctuation-prediction task, achieving absolute F1 improvements of 6.2 percentage points for question marks and up to 9 points for full stops over the previous state of the art [4]. To date, most punctuation-analysis models and techniques have been designed for IndoEuropean languages [8]. Semantic relatedness among words is a central concept in NLP, making semantic evaluation an important challenge [7]. As in many other fields, linguistics is rapidly adapting to digital technologies; processing traditional linguistic norms within computer programs and information systems is therefore critical [9]. Paraphrasing tools (“synonymators”) are now popular worldwide: they raise textual originality, automate rewriting, and help avoid repetition. Alongside traditional rule-based methods, machine-learning approaches have recently become widespread for building such applications [10]. When designing a morphological analyser, researchers can choose from a range of tools whose functions and characteristics have been discussed in detail [11]. Work on the low-resource Karakalpak language, spoken by about two million people in Uzbekistan, tackles the automatic extraction of out-of-vocabulary words for NLP systems [12]. Finally, lemmatization is another fundamental concept in NLP. Reference [13] presents an algorithm for Uzbek that uses a finite-state machine to strip affixes and identify each word’s lemma (the dictionary form). II.1 Structure of Interrogative Sentences A question mark (?) is one of the end-of-sentence punctuation marks. It is placed after sentences whose meaning is explicitly interrogative and signals that the sentence is a question. 258

In Uzbek, interrogative sentences contain lexical or grammatical devices that express the question: interrogative pronouns, interrogative particles, or interrogative intonation. The main patterns are illustrated below. 1. Interrogative pronouns (kim “who”, nima “what”, qayerda “where”, qachon “when”, nima uchun “why”, qanday “how”, etc.) • Bu kim? — “Who is this?” • Isming nima? — “What is your name?” 2. Interrogative particles (-mi, -chi, -a, -ya) • Ertaga universitetga boramizmi? — “Shall we go to the university tomorrow?” • Men universitetdaman, o‘zingiz-chi? — “I’m at the university—what about you?” • Shu universitetda o‘qiysiz-a? — “You study at this university, right?” 3. Intonation-only questions – sentences that contain neither interrogative pronouns nor particles; the questioning meaning is conveyed purely by rising intonation. • Vazifani bajarib bo‘ldim. — “I’ve finished the task.” • Vazifani bajarib bo‘ldim? — “Have I really finished the task?” (intonation creates the question) Note. Even when interrogative pronouns or particles appear in a sentence, it is not always an interrogative sentence. Avval kitobni o‘qib ko‘ray-chi, oh, naqadar so‘lim joylar-a! “Let me read the book first—oh, what beautiful places!” Here, -chi and -a are present, but the sentence is exclamatory, not interrogative. Developing the Interrogative-Sentence Detection Model Mathematical formulation In a linear-chain CRF, xususiyat.txt denote the token at position xt and yt∈{Q,O}yt {Q, O}yt ∈{Q,O} its label, where Q marks a token inside a question and O marks a token outside a question. The question-pronoun feature is defined as 1, 𝑖𝑓 𝑥𝑡 ꞒP (𝑞𝑢𝑒𝑠𝑡𝑖𝑜𝑛 − 𝑝𝑟𝑜𝑛𝑜𝑢𝑛) 𝑎𝑛𝑑 𝑦𝑡 = 𝑄, Fpron(yt,x,t)={ 0, 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒 where xususiyat.txt collection read from P={kim, nima, qchon, …} Results We tested our question-sentence detection algorithm on Uzbek texts written in the Latin script. The evaluation set contained 30 distinct text categories comprising 137 ,848 sentences and 1 ,642 ,860 words. Within this corpus there were 7 ,480 interrogative sentences. • The system correctly identified 7 ,180 of these questions and mistakenly treated 294 of them as non-interrogative. • Among the 130 ,368 declarative sentences, it wrongly labeled 554 as questions. These figures yield an overall accuracy of 89 %. To our knowledge, no earlier system has tackled Uzbek question detection, but the results can still be improved. One promising avenue is to integrate a statistical n-gram model. In future work we intend to combine such an n-gram component with our current rule-based algorithm to achieve even better performance. Table X (not shown here) presents the detailed breakdown: • Column 1 – index; • Column 2 – text type; • Column 3 – number of sentences in the text; 259