http://hdl.handle.net/123456789/4 http://hdl.handle.net/123456789/5801 IMPACT OF TEXT CLASSIFICATION ON NATURAL LANGUAGE PROCESSING APPLICATIONS Šandrih, Branislava The main goal of this dissertation is to put different text classification tasks in the same frame, by mapping the input data into the common vector space of linguistic attributes. Subsequently, several classification problems of great importance for natural language processing are solved by applying the appropriate classification algorithms. The dissertation deals with the problem of validation of bilingual translation pairs, so that the final goal is to construct a classifier which provides a substitute for human evalu- ation and which decides whether the pair is a proper translation between the appropriate languages by means of applying a variety of linguistic information and methods. In dictionaries it is useful to have a sentence that demonstrates use for a particular dictio- nary entry. This task is called the classification of good dictionary examples. In this thesis, a method is developed which automatically estimates whether an example is good or bad for a specific dictionary entry. Two cases of short message classification are also discussed in this dissertation. In the first case, classes are the authors of the messages, and the task is to assign each message to its author from that fixed set. This task is called authorship identification. The other observed classification of short messages is called opinion mining, or sentiment analysis. Starting from the assumption that a short message carries a positive or negative attitude about a thing, or is purely informative, classes can be: positive, negative and neutral. These tasks are of great importance in the field of natural language processing and the proposed solutions are language-independent, based on machine learning methods: sup- port vector machines, decision trees and gradient boosting. For all of these tasks, a demonstration of the effectiveness of the proposed methods is shown on for the Serbian language. http://hdl.handle.net/123456789/5800 POLINOMIJALNA ENTROPIJA ZA MORSOVE GRADIJENTNE SISTEME I LOGISTIČKO PRESLIKAVANJE Perić, Milan This thesis presents a method for calculating the polynomial entropy of the topolog- ical dynamic system with finitely many non-wandering points. A special coding is adapted for such systems. Thanks to this coding the polynomial entropy can be bounded by the number of specific mutually singular points in the closures of stable manifolds of non-wandering points. This method was applied to Morse gradient systems. It is shown that the polynomial entropy of the Morse gradient system is bounded by n(F ) − 1, where n(F ) is the number of different Morse indices of critical points of the Morse function F. If Morse gradient systems on mani- folds of dimension n has only critical points of indices 0 and n, it is proved that the polynomial entropy is equal to 1, and if the system has critical points of indices 0, n/2 and n, it is proved that polynomial entropy is equal to 2. The polynomial entropy for different parameter values in logistic map has also been calculated, and it has been shown that the polynomial entropy distinguishes the systems of low complexity with drastically different behaviours, which cannot be distinguished by the topological entropy. The example of the homeomorphism of the con- nected compact metric space that is not equicontinuous and with vanishing polynomial entropy is also given. http://hdl.handle.net/123456789/5799 UTAPANJA PROSTORA HARMONIJSKIH FUNKCIJA SA MEŠOVITOM NORMOM U OGRANIČENIM OBLASTIMA U Rn Jovanović Spasojević, Tanja In this thesis, subjects of consideration are the embeddings theorems of weighted Bergman spaces in Lp-spaces, as well as embeddings theorems of harmonic mixed norm spaces. The first part of the thesis generalizes the theorems of embeddings Bergman spaces into Lp(μ)-spaces, where μ is a Borel measure on a given domain. They have been earlier studied on domains such as unit ball and upper half-space. Generalization refers to bounded domains Ω ⊂ Rn with C1 boundary. This embedding will be valid to any p > 0, whenever the measure of the spaces Lp satisfies the Carledon condition. Reverse the direction will be valid only in case if p > 1 + α+2 n−2 . The second part of the dissertation also generalizes the embeddings theorems of mixed norm spaces of harmonic functions on a unit ball, where the generalization is applied to the domain Ω ⊂ Rn with C1 boundary. However, in addition we are obtaining another important result relating to the limitation of the maximum operators in the mixed norm on the general domain for the class of QNS functions. http://hdl.handle.net/123456789/5798 EKSTREMALNI PROBLEMI BRAUNOVOG KRETANJA I DRUGIH SLUČAJNIH PROCESA Jovalekić, Milica Let M be a maximum and let N be a minimum of the non-negative martingale X1, X2, . . . , Xn. It is well known, that if X1 = 1, then γ(‖M ‖1) ≤ E (Xn log Xn) and γ(‖N ‖1) ≤ E (Xn log Xn) , where γ(x) = x − 1 − log x, for all x > 0. In this thesis, we prove the analogue of this result in the case when 1 < p < ∞, by proving that δp (‖M ‖p p ) ≤ ‖Xn‖p and δp (‖N ‖p p ) ≤ ‖Xn‖p, where δp(x) = ( 1 − 1 p ) x 1 p + 1 p x 1 p −1, for all x > 0. We also obtain a probabilistic proof of the fact min ρ∈D(Qn) ∫ Qn dx1 . . . dxn ρ (x1, . . . , xn)p−1 ∏n j=1 xαj +1 j = n∏ j=1 ( p p − αj − 1 )p , where p > 1, αj < p − 1 for j = 1, . . . , n and D (Qn) is family of all densities on the n-dimensional unit cube Qn = (0, 1)n in Rn. This provides the proof of the multidimensional weighted Hardy inequality. Namely, if f : Rn + → (0, ∞) is a measurable function, p > 1 and αj < p − 1 for j = 1, . . . , n, then ∫ Rn + n∏ j=1 xαj j Hnf (x)p dx ≤ n∏ j=1 ( p p − αj − 1 )p ∫ Rn + n∏ j=1 xαj j f (x)p dx, where Hnf (x) = 1 x1 . . . xn ∫ x1 0 · · · ∫ xn 0 f (t) dt, is a multidimensional Hardy operator, x = (x1, . . . , xn) ∈ Rn +, t = (t1, . . . , tn) and dt = dt1 . . . dtn. Let B(t) be a standard planar Brownian motion and r(θ) be the length of the projection of B[0, 1] on the line generated by the unit vector eθ = (cos θ, sin θ), where 0 ≤ θ ≤ π. We nd the common distribution function F of the random variables r(θ). Namely, we prove that F(x) = 8 ∞∑ n=1 ( 1 x2 + 1 (2n − 1)2π2 ) exp ( − (2n − 1)2π2 2x2 ) , for every x > 0. As immediate consequence, lower bound for the expected diameter of the set B[0, 1], better than known, is obtained. Namely, it is known that Ed ≥ 1.601, where d is the diameter of the set B[0, 1]. In this thesis we show Ed ≥ 1.856.