黑料不打烊

Baigiamojo darbo tyrimo objektas yra gimusi懦 nuo 2017 iki 2021 met懦 Vilniaus mieste vaik懦 skai膷ius. 艩ie duomenys yra suskirstyti pagal Vilniaus miesto seni奴naitijas detalesnei prognozei atlikti. Darbe atliekami testai tinkamam statistiniam modeliui nustatyti. Tikrinamos 3 丑颈辫辞迟别锄臈s: duomenys pasiskirst臋 normaliuoju skirstiniu, duomenys pasiskirst臋 trikampiu skirstiniu ir duomenys pasiskirst臋 trapeciniu skirstiniu. Tinkamumas vertinamas chi-kvadratu kriterijumi. Darb膮 sudaro 9 dalys: 寞vadas, duomen懦 apra拧as, duomen懦 normali拧kumo tyrimo apra拧ymas, trikampio skirstinio apra拧as, trapecinio skirstinio i拧vedimas ir apra拧ymas, prognoz臈s sudarymas, i拧vados, literat奴ros s膮ra拧as, priedai. Darbo apimtis 鈥� 22 Atskirai pridedami darbo priedai.

The subject of the final thesis is the number of children born in Vilnius city from 2017 to 2021. This data is divided according to the districts of Vilnius city for a more detailed forecast. The research involves testing to determine a suitable statistical model. Three hypotheses are examined: data distributed according to a normal distribution, data distributed according to a triangular distribution, and data distributed according to a trapezoidal distribution. Suitability is evaluated using the chi-square criteria. The project consists of 9 parts: introduction, data description, description of the normality test of the data, description of the triangular distribution, derivation and description of the trapezoidal distribution, forecasting, conclusions, list of references, and appendices. Work includes 22 pages of text. The attachments to the project are included separately.

Baigiamajame bakalauro darbe kuriami tradiciniai generatyvini懦 prie拧i拧k懦 tinklai. Teorin臈je dalyje, remiantis Lietuvos ir u啪sienio autori懦 mokslin臈s literat奴ros ir straipsni懦 拧altiniais, i拧analizuojami pagrindiniai dirbtinio intelekto, ma拧ininio ir gilaus mokymosi teoriniai pagrindai, dirbtini懦 neuronini懦 tinkl懦 veikimo modelis ir atliekama generatyvini懦 prie拧i拧k懦 tinkl懦 ap啪valga. Tyrimo dalyje sukuriami tradiciniai generatyviniai prie拧i拧ki tinklai naudodami Python programavimo kalbos ma拧ininio mokymosi bibliotek膮 鈥� TensorFlow . I拧vadose aptariami gauti rezultatai. Darb膮 sudaro 7 dalys: 寞vadas, dirbtinio intelekto teoriniai aspektai, neuronini懦 tinkl懦 teoriniai pagrindai, generatyvini懦 prie拧i拧k懦 tinkl懦 ap啪valga, generatyvini懦 prie拧i拧k懦 tinkl懦 k奴rimas, i拧vados, literat奴ros s膮ra拧as ir si奴lymai. Darbo apimtis: 42 p. be pried懦, 21 pav., 33 bibliografiniai 拧altiniai. Atskirai pridedami darbo priedai.

The final bachelor's thesis examines the methods and development of generative adversarial networks. In the theoretical part, based on the sources of scientific literature and articles by Lithuanian and foreign authors, the thesis describes the main theoretical foundations of artificial intelligence, machine and deep learning, artificial neural network operation models are analyzed and an overview of generative adversarial networks is performed. In the research part, traditional generative adversarial networks are implemented using Python programming language鈥檚 machine learning library 鈥� TensorFlow. The results are discussed in the conclusions. Thesis consists of 7 parts: introduction, theoretical aspects of artificial intelligence, theoretical foundations of neural networks, overview of generative adversarial networks, development of generative adversarial networks, conclusions and suggestions, bibliography. Volume of work: 42 p. without annexes, 21 pictures, 33 bibliographical entries. Annexes are attached separately.

Baigiamajame bakalauro darbe, taikant ekonometrin寞 modeliavim膮, yra modeliuojami Covid-19 infekcijos rodikli懦 鈥� 鈥瀖ir膷i懦鈥� ir 鈥瀙atvirtint懦鈥� atvej懦 modeliai, skai膷iuojamas reprodukcinis skai膷ius R, kuris nusako kiek vidutini拧kai asmen懦 sergantis asmuo perduos virus膮. 艩iame darbe yra taikomi du modeliai 鈥� tiesin臈 regresin臈 analiz臈 ir laikin臈s sekos. Pirmoje darbo dalyje yra aptariamas reprodukcinis skai膷ius R, jo skai膷iavimo metodai, rodiklis prakti拧kai pritaikomas turimiems Covid-19 infekcijos plitimo rodikliams. Antroje dalyje aptariama tiesin臈s regresijos analiz臈s tinkamumas, modelio prielaidos, modelis pritaikomas turimiems duomenis. Tre膷ioje dalyje apra拧omas laikini懦 sek懦 modelis, atliekami skai膷iavimai su turimais duomenimis. Darb膮 sudaro penkios dalys: 寞vadas, duomen懦 apra拧ymas, reprodukcinio skai膷iaus R aktualumas, du tiriamos dalies skyriai. Darbo apimtis 鈥� 44 p. teksto be pried懦, 27 iliustracijos, 10 lenteli懦, 35 bibliografiniai 拧altiniai

In the final bachelor's thesis, econometric modelling is used to create models of Covid-19 infection rates, such as 鈥渄eaths鈥� and 鈥渃onfirmed鈥� cases, and to calculate the reproductive number R, which presents how many people the virus will be transferred to by an infected individual. Two models are used in this work: linear regression analysis and temporal sequences. The first part of the work discusses the reproductive number R, its calculation methods, and the indicator is applied to the available Covid-19 transmission rates in practice. The second part discusses the appropriateness of linear regression analysis, model assumptions and models applied to the dataused. The third part describes the model of the temporary sequence and calculations are performed using the provided data. The work consists of five parts: introduction, data description, relevance of the reproductive number R, and two sections of reasearch and calculation. The scope of the work is 44 pages of text without attachments, 27 illustrations, 10 tables, 35 bibliographic sources.

Baigiamajame bakalauro darbe did啪iausias d臈mesys buvo skirtas epidemiologini懦 modeli懦 analizei ir j懦 taikymui modeliuojant choleros plitimo dinamik膮. 艩io darbo tikslas 鈥� susipa啪inti su matematiniais modeliais biologijoje ir atskleisti, kaip modeliuojant t膮 pat寞 proces膮 naudojant skirtingus matematinius modelius gali pasikeisti ligos plitimo dinamika. Taikant studij懦 metu 寞gytas 啪inias, atlikome keli懦 modeli懦 kokybin臋 bei kiekybin臋 analiz臋 ir palyginome gautas i拧rai拧kas. Darb膮 sudaro 8 dalys: 漠vadas, epidemiologiniai modeliai, epidemiologini懦 modeli懦 tyrimo teorija, SIR modelis, keli choleros infekcijos matematiniai modeliai, gaut懦 rezultat懦 analiz臈 ir palyginimas, i拧vados ir literat奴ra. Darbo apimtis 鈥� 42 p. teksto be pried懦, 9 iliustr., 11 lent., 22 bibliografiniai 拧altiniai.

In the final bachelor's thesis, the main focus was on the analysis of epidemiological models and their application in modeling the dynamics of cholera spread. The aim of this work is to get acquainted with mathematical models in biology and to reveal how the dynamics of disease spread can vary when different mathematical models are used to simulate the same process. Applying the knowledge acquired during the studies, we conducted qualitative and quantitative analyses of several models and compared the obtained expressions. The work consists of 8 parts: Introduction, Epidemiological Models, Theory of Epidemiological Model Analysis, SIR Model, Several Mathematical Models of Cholera Infection, Analysis and Comparison of Obtained Results, Conclusions, and Literature. The scope of the work includes 42 pages of text without extras, 9 pictures, 11 tables, and 22 bibliographic sources.

Baigiamajame bakalauro darbe yra nagrin臈jami automatinio vaizd懦 generavimo pasitelkiant tekstin臋 寞vest寞 metodai. Tyrimo metu yra aptariamos skirtingos tok寞 u啪davin寞 sprend啪ian膷i懦 algoritm懦 technologijos ir atliekama j懦 analiz臈. Daugiausiai d臈mesio skiriama naujausiems difuzija pagr寞stiems modeliams bei j懦 veikimo matematiniam pagrindimui, optimizavimo ir apmokymo metodams. Baigiamajame darbe apra拧omos trys skirtingos difuzijos modeli懦 realizacijos bei j懦 architekt奴ros. Taip pat analizuojami automatinio vaizd懦 generavimo pasitelkiant tekstin臋 寞vest寞 vertinimo metodai ir j懦 veikimo principai. I拧nagrin臈jus difuzijos modeli懦 teorinius ir praktinius veikimo aspektus, pateikiama gilesn臈 Stable Diffusion difuzijos modelio architekt奴ros analiz臈. Baigiamajame bakalauro darbe taip pat pateikiama difuzijos modeli懦 papildomo apmokymo technika, kuria remiantis atliekami bandymai su architekt奴ros parametrais. Tyrimo dalyje palyginama keli懦 skirting懦 parametr懦 parink膷i懦 寞taka galutiniam generuojamam vaizdui bei modelio veikimui. Papildomo apmokymo technika realizuojama bei 寞vertinama realizuojant Python kod膮 Google Colaboratory platformoje. Darb膮 sudaro 8 dalys: 寞vadas, automatinio vaizd懦 generavimo technologij懦 analiz臈, difuzijos modelio veikimas, difuzijos modeli懦 realizacijos ir j懦 寞vertinimo metodai, architekt奴ros tyrimo dalis, i拧vados, literat奴ros s膮ra拧as, priedai. Darbo apimtis: 40 p. be pried懦, 30 iliustracij懦, 38 bibliografiniai 拧altiniai. Atskirai pridedami darbo priedai.

The final bachelor's thesis examines the methods of automatic image generation using text input. During the study, different technologies of algorithms solving such a problem and their analysis are discussed. Most of the attention is paid to the latest diffusion-based models and the mathematical justification of their operation, optimization and training methods. The thesis describes three different implementations of diffusion models and their architectures. Thesis also contains evaluation methods for automatic image generation using text input and their operating principles. After examining the theoretical and practical operational aspects of diffusion models, a deeper analysis of the architecture of the Stable Diffusion model is presented. The bachelor's thesis also presents a technique for additional training of diffusion models, based on which experiments with architectural parameters are carried out. The research part compares the influence of several different parameter options on the final generated image and the performance of the model. The additional training technique is implemented and evaluated by implementing Python code on the Google Collaboratory platform. Thesis consists of 8 parts: introduction, analysis of automatic image generation technologies, operation of the diffusion model, implementation of diffusion models and their evaluation methods, architectural research part, conclusions, bibliography, appendices. Volume of work: 40 p. without annexes, 30 illustrations, 38 bibliographical entries. Annexes are attached separately.

Baigiamojo darbo tikslas - susipa啪inti su 寞vairiomis duomen懦 suspaudimo technologijomis, pasirinkti tam tikrus algoritmus bei suspaud啪iam懦 duomen懦 tip膮 ir atlikti skai膷iavimo eksperimentus. Gautus rezultatus palyginti tarpusavyje. Darbo analitin臈je dalyje bus ap啪velgiama: populiariausi duomen懦 suspaudimo algoritmai, j懦 prana拧umai ir tr奴kumai. Praktin臈je darbo dalyje bus palyginamos duomen懦 suspaudimo algoritm懦 ypatyb臈s, atlikti skai膷iavimo eksperimentai, bei palyginami egzistuojantys duomen懦 suspaudimo programiniai produktai. Ap啪velgiamojoje dalyje bus pateiktos gaut懦 skai膷iavim懦 rezultat懦 analiz臈s, bei i拧vados ir rekomendacijos. Darbo apimtis: 56 puslapiai, 22 paveiksl臈liai, 18 lentel臈s, 16 拧altini懦.

The aim of the thesis is to get acquainted with various data compression technologies, choose certain algorithms and the type of data to be compressed, and perform calculation experiments. Compare the results. The analytical part will review: the most popular data compression algorithms, their advantages and disadvantages. In the practical part of the work, the characteristics of different data compression algorithms will be compared, computational experiments will be performed, and existing data compression software products will be compared. The overview will present the analysis of the results as well as conclusions and recommendations. Scope of work: 56 pages, 22 pictures, 18 tables, 16 sources.

Baigiamajame darbe atliekama Solitaire 拧ifravimo algoritmo analiz臈. M奴s懦 laikais vis daug臈ja internetinio 寞silau啪imo, duomen懦 padirbimo ir kitokios nusikalstomos veiklos atvej懦, tod臈l norint u啪kirsti keli膮 pavojui, kuris sukeliamas konfidencialiai informacijai, yra naudojama kriptografija. Kriptografija tiria duomen懦 拧ifravimo metodus ir algoritmus. 漠vairios 拧ifravimo algoritm懦 grup臈s pritaiko skirtingus b奴dus, kaip apsaugoti informacij膮 nuo tre膷i懦j懦 拧ali懦. Siekiant i拧analizuoti Solitaire algoritmo saugum膮, buvo i拧keltas tyrimo objektas: Solitaire 拧ifravimo algoritmo analiz臈. Tam, 寞gyvendinimui buvo i拧keltas tikslas: atlikti Solitaire 拧ifravimo algoritmo analiz臋, realizuoti Solitaire 拧ifravimo algoritm膮 ir atlikti algoritmo nulau啪imo atakas. Tikslui realizuoti buvo i拧kelti 6 u啪daviniai: 1. Susipa啪inti su skirtingomis duomen懦 拧ifravimo sistem懦 grup臈mis. 2. Per啪i奴r臈ti mokslin臋 literat奴r膮 susijusia su Solitaire 拧ifravimo algoritmu. 3. Aptarti Solitaire 拧ifravimo algoritmo etapus. 4. Aptarti Solitaire 拧ifravimo algoritmo saugum膮. 5. Realizuoti Solitaire 拧ifravimo algoritm膮. 6. Atlikti Solitaire 拧ifravimo algoritmo nulau啪imo atakas. Pirmame 啪ingsnyje, buvo susipa啪inta su slaptojo ir vie拧ojo rakto 拧ifravimo sistem懦聽grup臈mis. Slaptojo rakto sistemos naudoja tik vien膮 slapt膮j寞 rakt膮 duomen懦 拧ifravimui, tuo tarpu, vie拧ojo rakto sistemos naudoja specialias聽vie拧ojo ir slaptojo rakt懦 poras. Antrajame 啪ingsnyje aptar臈me Solitaire 拧ifravimo algoritm膮 ir jo 拧ifravimo etapus. Generavimo etape i拧 slaptojo rakto i拧gaunamas keystream raktas. Toliau, 拧ifravimo etape vyksta paprasto teksto u啪拧ifravimas su keystream raktu. Tre膷iajame etape aptar臈me Solitaire 拧ifravimo algoritmo saugum膮. Nustat臈me algoritmo 拧ali拧kum膮, bei aptar臈me nulau啪imo metodus. Ketvirtame 啪ingsnyje realizavome Solitaire 拧ifravimo algoritm膮. Algoritmas realizuotas Python ir C++ programavimo kalbomis. Galiausiai atlikome聽Solitaire 拧ifravimo algoritmo nulau啪imo atakas. 漠vykd臈me Brute force nulau啪imo atakas Python ir C++ programomis.

The BA thesis analyzes the encryption algorithm of Solitaire. Nowadays, with the increasing number of cases of online hacking, data falsification and other criminal activities, cryptography is used to prevent the danger posed to confidential information. Cryptography is the study of data encryption methods and algorithms. Different groups of encryption algorithms apply different ways to protect information from third parties. In order to analyze the security of the Solitaire algorithm, a research subject was raised: analysis of the Solitaire encryption algorithm. For that, the aim was set: to analyze the Solitaire encryption algorithm, to implement the Solitaire encryption algorithm and to perform algorithm cracking attacks. To realize the aim, 6 objectives were set: 1. Get to know different groups of data encryption systems. 2. Review the scientific literature related to the Solitaire encryption algorithm. 3. Discuss the stages of the Solitaire encryption algorithm. 4. Discuss the security of Solitaire's encryption algorithm. 5. Implement the Solitaire encryption algorithm. 6. Perform hacking attacks on the Solitaire encryption algorithm. In the first step, the groups of secret and public key encryption systems were introduced. Private key systems use only one secret key to encrypt data, whereas public key systems use special pairs of public and private keys. In the second step, we discussed the Solitaire encryption algorithm and its encryption steps. In the generation phase, the keystream key is extracted from the secret key. Next, the encryption step involves encrypting the plaintext with the keystream key. In the third step, we discussed the security of Solitaire's encryption algorithm. We determined the bias of the algorithm and discussed hacking methods. In the fourth step, we implemented the Solitaire encryption algorithm. The algorithm is implemented in Python and C++ programming languages. Finally, we performed attacks to crack Solitaire's encryption algorithm. We performed brute force hacking attacks using Python and C++ programs.

Baigiamojo darbo tikslas yra i拧analizuoti ma拧ininio mokymosi metod懦 tikslum膮 sprend啪iant prognozavimo u啪davin寞. Darbe i拧analizuoti trys ma拧ininio mokymosi algoritmai: sprendim懦 medis, naivusis Bajesas ir atramini懦 vektori懦 klasifikatorius. Metodai lyginami apskai膷iuojant tikslumo metrikas, vertinant mai拧aties matricas bei plot膮 po ROC kreive. Eksperimentai atlikti su vie拧ai prieinamu telekomunikacij懦 duomen懦 rinkiniu, sprend啪iant klient懦 praradimo u啪davin寞. I拧 baigiamajame darbe i拧analizuot懦 metod懦 tiksliausias buvo 鈥� sprendim懦 med啪io klasifikatorius. Darb膮 sudaro 寞vadas, literat奴ros, skirtos klient懦 praradimo u啪daviniui spr臋sti, analiz臈, ma拧ininio mokymosi srities ap啪valga, klasifikavimo metod懦 analiz臈, telekomunikacij懦 duomen懦 rinkinio analiz臈, metod懦 palyginimas, i拧vados. Darbo apimtis - 45 p., 18 pav., 4 lentel臈s, 31 bibliografinis 拧altinis.

The aim of the final work is to analyze the accuracy of machine learning methods in solving the prediction problem. Three algorithms were analyzed: decision tree, naive Bayes and support vector machine classifier. The methods are compared by calculating precision metrics, estimating the confusion matrices and the area under the ROC curve. The experiments were performed with a publicly available set of telecommunications data to solve the problem of customer loss. The most accurate of the methods analyzed in the thesis was the decision tree classifier. The work consists of an introduction, analysis of the literature for solving the problem of customer loss, an overview of machine learning, analysis of classification methods, analysis of telecommunications data set, comparison of methods, findings. Thesis consists of 45 pages, 18 figures, 4 tables, 31 bibliographical entries.

Baigiamojo darbo modeliavimo tikslas yra pagerinti gamybos proces懦 valdym膮 simuliacijos ir tam tikr懦 valdymo dali懦 automatizacijos pagalba. Darbe susipa啪inome su 鈥濼eltonika LoT Group鈥� 寞moni懦 grup臈s gamybin臋 veikla. Apibr臈啪臈m 拧iuo metu esan膷i膮 gamybin臈s veiklos problem膮, kad esantis gamybos proces懦 valdymas yra neefektyvus. Susipa啪inome su gamybos procesais Teltonika gamykloje ir j懦 modeliavimo b奴dais. Darbe suformuluojamas optimizavimo u啪davinys, apra拧yti realizacijos ypatumai. I拧nagrin臈tas 寞rankis, kuris atkartoja gamykloje vykstan膷ius procesus. Modeliuojant proces膮 diskre膷i懦 寞vyki懦 simuliacijos b奴du 鈥� panaudotas sekan膷io 寞vykio modeliavimo metodas. Sprend啪iant optimizavimo u啪davin寞 panaudojamas Monte Karlo metodas. I拧tirta papildymo ir patobulinimo egzistuojan膷i懦 sprendim懦 galimyb臈s, tokios kaip euristik懦 寞traukimas ir sprendini懦 vizualizavimas. Simuliacijos vizualizavimas leid啪ia 啪mogui geriau analizuoti sprendin寞. Viena i拧 nagrin臈t懦 darbe euristik懦 remiasi u啪sakym懦 grupavimu pagal kod膮, test懦 rezultatai parod臈, kad ie拧kant optimalaus sprendimo su duomen懦 grupavimu, skai膷iavimo laikas buvo i拧gaunamas geresnis, bet optimizavimo rezultatai gaunami blogesni, nei skai膷iuojant be duomen懦 grupavimo. Siekiant pagerinti rezultatus, u啪davinys buvo kei膷iamas 寞 kitok寞, buvo panaudojama porcij懦 dyd啪i懦 keitimo euristika. Testavimo rezultate gavome suma啪臈jus寞 laik膮, bet optimizavimo rezultatai tampa blogesni. Galiausiai atlikti lygiagretinimo testai, siekiant i拧gauti geresnio skai膷iavimo laiko ir neblogesnio optimizavimo rezultatus. Gauti rezultatai parod臈, kad skai膷iuojant su 4 gij懦 skai膷iumi, spartinimo koeficientas yra lygus ~3.55, bei efektyvumo koeficientas ~0.89. Taip pat, optimizavimo rezultatai taikant lygiagretinim膮 pablog臈ja ne啪enkliai. Darb膮 sudaro 寞vadas, gamyklos proces懦 modeliavimas, sukurto 寞rankio apra拧ymas, vizualizavimas, rezultatai, i拧vados. Darbo apimtis 鈥� 42 p., 20 pav., 5 lent., 13 literat奴rini懦 拧altini懦.

The aim of the thesis is to improve the control of manufacturing processes with the help of simulation and automation of certain control parts. In the work, we got acquainted with the manufacturing activities of the Teltonika LoT Group. We have defined the current problem of manufacturing activities. The existing management of manufacturing processes is inefficient. We got familiar with the manufacturing processes and their modeling methods at the Teltonika plant. In this work we formulated optimization problem which is described in detail. The tool that replicates factory processes was analyzed. For process modeling by discrete event simulation the next event modeling method was used. To solve optimization problem, we used Monte Karlo optimizing method. We have explored the possibilities for supplementing and improving the existing approach, such as the inclusion of heuristics and the visualization of solutions. Simulation visualization allows a person to better analyze the solution. One of the heuristics analyzed in the work is based on the grouping of orders by code. Test results showed that the search for the optimal solution with data grouping resulted in better computation time, but the optimization results are worse than the ones without data grouping. Seeking for the improvement of the computations speed, the task was changed to a different one, where heuristics of changing the portion sizes were used. This resulted in a faster calculation at the cost of better results. Finally, parallelization tests were performed to obtain better computation time while achieving good optimization results. The obtained results showed that calculations performed on 4 threads are performed faster ~ 3.55 times with the efficiency coefficient of ~0.89. Results of optimization with parallelization deteriorate insignificantly. The work consists of an introduction, modeling of factory processes, description of the developed tool, visualization, results, conclusions. Thesis consists of: 42 pages, 20 figures, 5 tables, 13 references.

Baigiamajame bakalauro darbe yra nagrin臈jami nat奴ralios kalbos apdorojimo u啪daviniai, tokie kaip trumpo teksto klasifikavimas pagal tekste esant寞 sentiment膮 bei klausim懦-atsakym懦 u啪davinys lietuvi懦 kalba ir j懦 sprendimas taikant dirbtiniais neuroniniais tinklais paremtus modelius bei sistemas. Aptariama nat奴ralios kalbos apdorojimo samprata, dirbtini懦 neuronini懦 tinkl懦 sudarymo ir veikimo principai, logistin臈 regresija ir jos taikymas binarinio klasifikavimo u啪davini懦 sprendimui. Ap啪velgiami 啪ingsniai, skirti paruo拧ti duomenis, pateiktus nat奴raliosios kalbos pavidalu, apdorojimui. Duomen懦 paruo拧imo 啪ingsniai ir teksto klasifikavimo u啪davinys, skirtas klasifikuoti internetinius komentarus 寞 teigiamus ir neigiamus ir realizuotas naudojant 鈥濸ython鈥� programavimo kalb膮. Sprend啪iant klausim懦-atsakymo u啪davin寞 pasitelkta sistema 鈥濿atson Assistant鈥�, sukurtas veikiantis virtualus agentas, kuris dirbtini懦 neuronini懦 tinkl懦 pagalba aptinka intencij膮 naudotojo klausime ir pagal j膮 parenka labiausiai tinkant寞 atsakym膮. Darb膮 sudaro 9 dalys: 寞vadas, nat奴ralios kalbos apdorojimas, dirbtiniai neuroniniai tinklai, logistin臈 regresija, trumpo teksto klasifikavimo u啪davinys, virtual奴s agentai, Vilnius Tech virtualus agentas, i拧vados, literat奴ros s膮ra拧as. Darbo apimtis 鈥� 42 p. teksto, 24 pav., 3 lent., 20 bibliografini懦 拧altini懦. Atskirai pridedami darbo priedai.

The subject of this bachelor鈥榮 thesis is Natural Language Processing tasks, such as the short text classification according to the sentiment in the text, the question-and-answer problem in Lithuanian language, and their solution using models and systems based on artificial neural networks. It consists the concepts of natural language processing, the construction and features of artificial neural networks, logistic regression. The steps of data processing and logistic regression are implemented using 鈥濸ython鈥� programming language. 鈥濿atson Assistant鈥� was used to solve the question answering task and to create virtual agent which is capable to detect intent of user question using neural networks and to choose the best suitable answer. Structure: introduction, natural language processing, short text classification task, virtual agents, virtual agent of Vilnius Tech, conclusions, references Thesis consist of: 42 p. text without appendixes, 24 pictures, 3 tables, 22 bibliographical entries. Appendixes included.