This thesis proposes a link-based approach to identify supporting and opposing groups in a Weblog community. We formulate the interaction behavior as a graph. Bloggers involved in the discussion of one specific issue are formulated as vertices. Semantic orientation is used to construct possible opposite opinion links. Bloggers with opposite stances will form an opposite link. A max-cut algorithm is used latter to obtain the optimal approximation of supporting and opposing groups. The categorization results are compared between semantic orientation classifier and simple link-based categorization. The simple link-based categorization compares then with the enhancement of link-based categorization using hypergraph.

The Max-Cut problem is a well-known NP-hard problem, for which numerous approximation algorithms have been developed over the years. In this thesis, we examine the special case where the degree of vertices in the graph is bounded. With minor modifications to existing algorithms, we are able to obtain an improved approximation ratio for general bounded-degree graphs. Furthermore we show additional improvements for graphs with at least a constant fraction of odd-degree vertices. We also identify some other possible areas for improvement in the general bounded-degree case. / Max-Cut-problemet är ett välkänt NP-svårt problem, för vilket ett flertal olika approximationsalgoritmer har utvecklats över åren. I det här arbetet undersöks specialfallet då grafens noder har begränsat gradtal. Med mindre förändringar av existerande algoritmer lyckas vi uppnå en förbättrad approximationskvot för generella gradtals-begränsade grafer. Vi visar också ytterligare förbättringar för grafer med minst en konstant andel noder med udda gradtal. Vi identifierar också några andra möjliga områden för förbättring i det allmänna gradtals-begränsade fallet.

Uma maneira de resolver problemas de programação linear de grande escala é explorar a relaxação lagrangeana das restrições \"difíceis\'\' e utilizar métodos de subgradientes. Populares por fornecerem rapidamente boas aproximações de soluções duais, eles não produzem diretamente as soluções primais. Para obtê-las com custo computacional adequado, pode-se construir seqüências ergódicas ou utilizar uma técnica proposta recentemente, denominada algoritmo do volume. As propriedades teóricas de convergência não foram bem estabelecidas nesse algoritmo, mas pequenas modificações permitem a demonstração da convergência dual. Destacam-se como adaptações o algoritmo do volume revisado, um método de feixes específico, e o algoritmo do volume incorporado ao método de variação do alvo. Este trabalho foi baseado no estudo desses algoritmos e de todos os conceitos envolvidos, em especial, análise convexa e otimização não diferenciável. Estudamos as principais diferenças teóricas desses métodos e realizamos comparações numéricas com problemas lineares e lineares inteiros, em particular, o corte máximo em grafos. / One way to solve large-scale linear programming problems is to exploit the Lagrangian relaxation of the difficult constraints and use subgradient methods. Such methods are popular as they give good approximations of dual solutions. Unfortunately, they do not directly yield primal solutions. Two alternatives to obtain primal solutions under reasonable computational cost are the construction of ergodic sequences and the use of the recently developed volume algorithm. While the convergence of ergodic sequences is well understood, the convergence properties of the volume algorithm is not well established in the original paper. This lead to some modifications of the original method to ease the proof of dual convergence. Three alternatives are the revised volume algorithm, a special case of the bundle method, and the volume algorithm incorporated by the variable target value method. The aim of this work is to study such algorithms and all related concepts, especially convex analysis and nondifferentiable optimization. We analysed the main theoretical differences among the methods and performed numerical experiments with linear and integer problems, in particular, the maximum cut problem on graphs.

Uma maneira de resolver problemas de programação linear de grande escala é explorar a relaxação lagrangeana das restrições \"difíceis\'\' e utilizar métodos de subgradientes. Populares por fornecerem rapidamente boas aproximações de soluções duais, eles não produzem diretamente as soluções primais. Para obtê-las com custo computacional adequado, pode-se construir seqüências ergódicas ou utilizar uma técnica proposta recentemente, denominada algoritmo do volume. As propriedades teóricas de convergência não foram bem estabelecidas nesse algoritmo, mas pequenas modificações permitem a demonstração da convergência dual. Destacam-se como adaptações o algoritmo do volume revisado, um método de feixes específico, e o algoritmo do volume incorporado ao método de variação do alvo. Este trabalho foi baseado no estudo desses algoritmos e de todos os conceitos envolvidos, em especial, análise convexa e otimização não diferenciável. Estudamos as principais diferenças teóricas desses métodos e realizamos comparações numéricas com problemas lineares e lineares inteiros, em particular, o corte máximo em grafos. / One way to solve large-scale linear programming problems is to exploit the Lagrangian relaxation of the difficult constraints and use subgradient methods. Such methods are popular as they give good approximations of dual solutions. Unfortunately, they do not directly yield primal solutions. Two alternatives to obtain primal solutions under reasonable computational cost are the construction of ergodic sequences and the use of the recently developed volume algorithm. While the convergence of ergodic sequences is well understood, the convergence properties of the volume algorithm is not well established in the original paper. This lead to some modifications of the original method to ease the proof of dual convergence. Three alternatives are the revised volume algorithm, a special case of the bundle method, and the volume algorithm incorporated by the variable target value method. The aim of this work is to study such algorithms and all related concepts, especially convex analysis and nondifferentiable optimization. We analysed the main theoretical differences among the methods and performed numerical experiments with linear and integer problems, in particular, the maximum cut problem on graphs.

Virtually all genome assemblers to date are designed for use with data from haploid or homozygous diploid genomes. Their use on heterozygous genomic datasets generally results in highly-fragmented, error-prone assemblies, owing to the violation of assumptions during both the contigging and scaffolding phases. Of the two phases, scaffolding is more particularly impacted and algorithms to facilitate the scaffolding of heterozygous data are lacking. We present a stand-alone scaffolding algorithm, ScaffoldScaffolder, designed specifically for scaffolding diploid genomes. A fundamental step in the scaffolding phase is the assignment of sequence orientations to contigs within scaffolds. Deciding such an assignment in the presence of ambiguous evidence is what is termed the contig orientation problem. We define this problem using bidirected graph theory and show that it is equivalent to the weighted MAX-CUT problem. We present a greedy heuristic solution which we comparatively assess with other solutions to the contig orientation problem, including an advanced MAX-CUT heuristic. We illustrate how a solution to this problem provides a simple means of simultaneously identifying inverted haplotypes, which are uniquely found in diploid genomes and which have been shown to be involved in the genetic mechanisms of several diseases. Ultimately our findings show that due to the inherent biases in the underlying biological model, a greedy heuristic algorithm performs very well in practice, retaining a higher total percent of edge weight than a branch-and-bound semidefinite programming heuristic. This application exemplifies how existing graph theory algorithms can be applied in the development of new algorithms for more accurate assembly of heterozygous diploid genomes.

A cut is a partition of a graph's nodes into two sets, and we say that an edge crosses the cut if it connects two nodes belonging to different sets. A maximum cut is a cut that maximises the number of crossing edges. We show that for any sufficiently small ε > 0 it is NP-hard to distinguish between graphs for which at least a fraction 1 - ε of all edges crosses the maximum cut and graphs for which at most a fraction 1 - 1.4568 ε of all edges crosses the maximum cut. The previous state of the art had a constant smaller than 1.375 in place of 1.4568. / Ett snitt är en partition av en grafs noder i två mängder, och vi säger att en kant korsar snittet om dess ändpunkter tillhör olika mängder. Ett maximalt snitt är ett snitt som maximerar antalet kanter som korsar snittet. Vi bevisar att det för alla tillräckligt små konstanter ε > 0 är NP-svårt att skilja mellan grafer för vilka minst en andel 1 - ε av alla kanter korsar det maximala snittet och grafer för vilka högst en andel 1 - 1.4568 ε av alla kanter korsar det maximala snittet. Detta är en förbättring jämfört med ett tidigare resultat som hade en konstant mindre än 1.375 istället för 1.4568.

MAX−CUT is one of the well-studied NP-hard combinatorial optimization problems. It can be formulated as an Integer Quadratic Programming problem and admits a simple relaxation obtained by replacing the integer “spin” variables xi by unitary vectors v⃗ i. The Goemans–Williamson rounding algorithm assigns the solution vectors of the relaxed quadratic program to a corresponding integer spin depending on the sign of the scalar product v⃗ i⋅r⃗ with a random vector r⃗ . Here, we investigate whether better graph cuts can be obtained by instead using a more sophisticated clustering algorithm. We answer this question affirmatively. Different initializations of k-means and k-medoids clustering produce better cuts for the graph instances of the most well known benchmark for MAX−CUT. In particular, we found a strong correlation of cluster quality and cut weights during the evolution of the clustering algorithms. Finally, since in general the maximal cut weight of a graph is not known beforehand, we derived instance-specific lower bounds for the approximation ratio, which give information of how close a solution is to the global optima for a particular instance. For the graphs in our benchmark, the instance specific lower bounds significantly exceed the Goemans–Williamson guarantee.

[pt] Nesta dissertação é apresentado um framework para desenvolver critérios de split para lidar com atributos nominais multi-valorados em árvores de decisão. Critérios gerados por este framework podem ser implementados para rodar em tempo polinomial no número de classes e valores, com garantia teórica de produzir um split próximo do ótimo. Apresenta-se também um estudo experimental, utilizando datasets reais, onde o tempo de execução e acurácia de métodos oriundos do framework são avaliados. / [en] In this dissertation we propose a framework for designing splitting criteria for handling multi-valued nominal attributes for decision trees. Criteria derived from our framework can be implemented to run in polynomial time in the number of classes and values, with theoretical guarantee of producing a split that is close to the optimal one. We also present an experimental study, using real datasets, where the running time and accuracy of the methods obtained from the framework are evaluated.

The goal of this thesis is to study the complexity of NP-Hard problems, using the Max-Cut and the Max-k-Cut problems, and the study of fitness landscapes. The Max-Cut and Max-k-Cut problems are well studied NP-hard problems specially since the approximation algorithm of Goemans and Williamson (1995) which introduced the use of SDP to solve relaxed problems. In order to prove the existence of a performance guarantee, the rounding step from the SDP solution to a Max-Cut solution is simple and randomized. For the Max-k-Cut problem, there exist several approximation algorithms but many of them have been proved to be equivalent. Similarly as in Max-Cut, these approximation algorithms use a simple randomized rounding to be able to get a performance guarantee. Ignoring for now the performance guarantee, one could ask if there is a rounding process that takes into account the structure of the relaxed solution since it is the result of an optimization problem. In this thesis we answered this question positively by using clustering as a rounding method. In order to compare the performance of both algorithms, a series of experiments were performed using the so-called G-set benchmark for the Max-Cut problem and using the Random Graph Benchmark of Goemans1995 for the Max-k-Cut problem. With this new rounding, larger cut values are found both for the Max-Cut and the Max-k-Cut problems, and always above the value of the performance guarantee of the approximation algorithm. This suggests that taking into account the structure of the problem to design algorithms can lead to better results, possibly at the cost of a worse performance guarantee. An example for the vertex k-center problem can be seen in Garcia-Diaz et al. (2017), where a 3-approximation algorithm performs better than a 2-approximation algorithm despite having a worse performance guarantee. Landscapes over discrete configurations spaces are an important model in evolutionary and structural biology, as well as many other areas of science, from the physics of disordered systems to operations research. A landscape is a function defined on a very large discrete set V that carries an additional metric or at least topological structure into the real numbers R. We will consider landscapes defined on the vertex set of undirected graphs. Thus let G=G(V,E) be an undirected graph and f an arbitrary real-valued function taking values from V . We will refer to the triple (V,E,f) as a landscape over G. We say two configurations x,y in V are neutral if f(x)=f(y). We colloquially refer to a landscape as 'neutral'' if a substantial fraction of adjacent pairs of configurations are neutral. A flat landscape is one where f is constant. The opposite of flatness is ruggedness and it is defined as the number of local optima or by means of pair correlation functions. These two key features of a landscape, ruggedness and neutrality, appear to be two sides of the same coin. Ruggedness can be measured either by correlation properties, which are sensitive to monotonic transformation of the landscape, and by combinatorial properties such as the lengths of downhill paths and the number of local optima, which are invariant under monotonic transformations. The connection between the two views has remained largely unexplored and poorly understood. For this thesis, a survey on fitness landscapes is presented, together with the first steps in the direction to find this connection together with a relation between the covariance matrix of a random landscape model and its ruggedness.

The inapproximability for NP-hard combinatorial optimization problems lies in the heart of theoretical computer science. A negative result can be either conditional, where the starting point is a complexity assumption, or unconditional, where the inapproximability holds for a restricted model of computation. Algorithms based on Linear Programming (LP) and Semidefinite Programming (SDP) relaxations are among the most prominent models of computation. The related and common measure of efficiency is the integrality gap, which sets the limitations of the associated algorithmic schemes. A number of systematic procedures, known as lift-and-project systems, have been proposed to improve the integrality gap of standard relaxations. These systems build strong hierarchies of either LP relaxations, such as the Lovasz-Schrijver (LS) and the Sherali-Adams (SA) systems, or SDP relaxations, such as the Lovasz-Schrijver SDP (LS+), the Sherali-Adams SDP (SA+) and the Lasserre (La) systems. In this thesis we prove integrality gap lower bounds for the aforementioned lift-and-project systems and for a number of combinatorial optimization problems, whose inapproximability is yet unresolved. Given that lift-and-project systems produce relaxations that have given the best algorithms known for a series of combinatorial problems, the lower bounds can be read as strong evidence of the inapproximability of the corresponding optimization problems. From the results found in the thesis we highlight the following: For every epsilon>0, the level-Omega(sqrt(log n/ log log n)) LS+ relaxation of the Vertex Cover polytope has integrality gap 2-epsilon. The integrality gap of the standard SDP for Vertex Cover remains 2-o(1) even if all hypermetric inequalities are added to the relaxation. The resulting relaxations are incomparable to the SDP relaxations derived by the LS+ system. Finally, the addition of all ell1 inequalities eliminates all solutions not in the integral hull. For every epsilon>0, the level-Omega(sqrt(log n/ log log n)) SA relaxation of Vertex Cover has integrality gap 2-epsilon. The integrality gap remains tight even for superconstant-level SA+ relaxations. We prove a tight lower bound for the number of tightenings that the SA system needs in order to prove the Pigeonhole Principle. We also prove sublinear and linear rank bounds for the La and SA systems respectively for the Tseitin tautology. Linear levels of the SA+ system treat highly unsatisfiable instances of fixed predicate-P constraint satisfaction problems over q-ary alphabets as fully satisfiable, when the satisfying assignments of the predicates P can be equipped with a balanced and pairwise independent distribution. We study the performance of the Lasserre system on the cut polytope. When the input is the complete graph on 2d+1 vertices, we show that the integrality gap is at least 1+1/(4d(d+1)) for the level-d SDP relaxation.