|
|
Data mining in social network analysis
Zvirinský, Peter ; Mrázová, Iveta (advisor) ; Drotár, Peter (referee) ; Vidnerová, Petra (referee)
Title: Data mining in social network analysis Author: Mgr. Peter Zvirinský Department: Department of Theoretical Computer Science and Mathematical Logic Supervisor: doc. RNDr. Iveta Mrázová CSc., Department of Theoretical Com- puter Science and Mathematical Logic Abstract. In the past several years, the global economy has experienced a sig- nificant increase in overall debt, reaching 238% of the world GDP in 2022, as reported by the International Monetary Fund. This growing indebtedness raises concerns about the stability of the financial system and the welfare of individuals and institutions. It also underscores the need for e ective strategies to under- stand the intricate relationships between debtors and creditors and to mitigate associated risks. In response, this thesis proposes a novel approach based on data mining methods for the comprehensive analysis of debt formation patterns among individuals and companies, focusing on the largely untapped data from the Insolvency Register (IR) of the Czech Republic. We aim to leverage social network analysis (SNA) methods to model and analyze the interactions among subjects participating in insolvencies, namely debtors, creditors, and insolvency administrators. Additionally, we focus our research on dynamic social networks that capture structural changes in...
|
|
|
Knowledge Extraction with Deep Belief Networks
Bronec, Jan ; Mrázová, Iveta (advisor) ; Červíčková, Věra (referee)
Deep Belief Networks (DBNs) are multi-layered neural networks constructed as a series of Restricted Boltzmann Machines stacked on each other. Like several other types of neural networks, increasing the size of a DBN will generally improve its performance. However, this comes at the cost of increased computational complexity and memory requirements. It is usually necessary to reduce a deep neural network's size to deploy it on a mobile device. To address this issue, we focus on a size-reduction technique called pruning. Pruning aims to zero out a large portion of the network's weights without significantly affecting its accuracy. We apply selected pruning algorithms to DBNs and evaluate their performance on both grayscale and color images. We also investigate the performance of the so-called confidence rules extracted from a trained DBN. These rules offer a knowledge representation that is easy to interpret. We investigate whether they also provide an accurate low-cost alternative to the original network. 1
|
|
|
Social Networks: Analysis of Evolution and Sentiment
Fanči, Samuel ; Mrázová, Iveta (advisor) ; Vomlelová, Marta (referee)
Nowadays, social networks form an essential part of our lives. Their analysis helps us better understand various social phenomena, identify individuals influencing society, and model future developments of communities. Often, real-world social networks con- form to power-law degree distribution. We oriented our research toward investigating communities surrounding two well-known companies: GameStop and Enron. Using the data obtained from Reddit and Twitter, we have trained machine learning models like Support vector machines and Neural networks to assess the sentiment of the GameStop community. The results confirm the expected positive sentiment following the GameStop price spike in 2021. We constructed the respective social networks based on the available datasets and identified their vital individuals according to selected centrality measures. Publicly known figures like Ryan Cohen in the case of GameStop and Jeff Skilling in the case of Enron are ranked high according to PageRank and Authority scores. On the other hand, minor influencers from the GameStop community and the upper management of Enron were assigned top ranks of the Hub score and Betweenness centrality. A statistical analysis using the goodness-of-fit test for the power-law degree distribution was performed for both networks. Results...
|
|
|
Detection of Influential Individuals, Communities, and Link Prediction in Social Networks
König, Matúš ; Mrázová, Iveta (advisor) ; Hric, Jan (referee)
Social network analysis provides several means to better understand the structure of the underlying social networks. This thesis is focused on the area of community detection in social networks. We discuss six of the main community detection algorithms and their hybrid variants involving a com- bination of rough and fine partitioning techniques. The text explains the measures used to quantify the detected communities' properties. For dif- ferent problem sizes, the Zachary's karate club and Enron email datasets were used. Further, the work concentrates on experiments that provide per- formance assessment for the investigated methods. Based on the obtained results, we draw conclusions towards recommendations for a reliable usage of the findings in practice. At the same time, we aim to identify the appropriate number of communities in the data at hand since this is a parameter of many community detection algorithms. For the same reason, we also investigate whether non-hierarchical clustering algorithms could be used to form a sub- community hierarchy. All of the mentioned experiments were run by means of a community detection system CGAT - Config-based Graph Analysis Tool we developed and implemented as a part of the thesis.
|
|
|
Knowledge representation in deep neural networks
Georgiev, Georgi Stoyanov ; Mrázová, Iveta (advisor) ; Pešková, Klára (referee)
Convolutional neural networks (CNNs) are known to outperform humans in numerous image classification and object detection tasks. They also excel at captioning, image segmentation, and feature extraction. CNNs are precise at recognition and generalize well, yet analyzing their decision-making process remains challenging. A means to study their internal knowledge representation provide the so-called heat maps and their variants like the saliency, SmoothGrad, and Grad-CAM maps. The techniques such as t-SNE, UMAP, and ivis can, on the other hand, help visualize the multi-dimensional features formed in different convolutional layers. Inspired by the results obtained when analyzing the capabilities of CNNs, we introduce two novel size-reduction algorithms: Iterative Top Cut and Iterative Feature Top Cut. Both algorithms successively remove the layers of a CNN starting from its top until a stopping criterion is activated. The stopping criteria involve the model's performance and the formed internal knowledge representation. In particular, the Iterative Top Cut method exceeds our expectations by shrinking some models, such as EfficientNetV2S, up to 3.15 times while preserving their accuracy on the Cars-196 dataset. Moreover, the algorithm generalizes well and proves to be stable. 1
|
|
|
Predicting the Outcomes of Darts Matches
Konečný, Tomáš ; Pilát, Martin (advisor) ; Mrázová, Iveta (referee)
This thesis deals with various approaches to modeling darts matches. We compare rating models, models based on statistics and a model that views the game states and random transitions between them as a Markov chain. As a part of the thesis, we propose a method for calculating statistics reflecting both long-term and short-term form of the players. Using a detailed dataset containing individual darts, we also derive how to choose a target based on the state of the match. The models are evaluated according to standard criteria for classification problems, but in addition, using bookmakers' odds, we estimate the profitability if betting would take place in practice according to the models' predictions. 1
|
|
|
Graph neural networks and their application to social network analysis
Behún, Marek ; Mrázová, Iveta (advisor) ; Vomlelová, Marta (referee)
Recently, the research on Graph Neural Networks (GNNs) made it possi- ble to apply deep learning techniques to graph-structured data. In this thesis, we explore the application of GNNs to Social Network Analysis (SNA). We build and compare deep learning models for the prediction of hotel review ratings, hotel classes, and hotel scores on data scraped from the Tripadvisor website. We consider the resulting models precise enough to be used by rec- ommender systems. A non-trivial part of this thesis is also the description of the theory behind GNNs and visualization techniques for high-dimensional data. We also provide software suitable for further experimentation on this topic.
|
|
|
Approximation of functions continuous on compact sets by layered neural networks
Fojtík, Vít ; Hakl, František (advisor) ; Mrázová, Iveta (referee)
Despite abundant research into neural network applications, many areas of the under- lying mathematics remain largely unexplored. The study of neural network expressivity is vital for understanding their capabilities and limitations. However, even for shallow networks this topic is far from solved. We provide an upper bound on the number of neurons of a shallow neural network required to approximate a function continuous on a compact set with given accuracy. Dividing the compact set into small polytopes, we ap- proximate the indicator function of each of them by a neural network and combine these into an approximation of the target function. This method, inspired by a specific proof of the Stone-Weierstrass Theorem, is more general than previous bounds of this character, with regards to approximation of continuous functions. Also, it is purely constructive. 1
|
|
|
Associative recall of damaged data
Lukešová, Jana ; Štanclová, Jana (advisor) ; Mrázová, Iveta (referee)
The focus of this work are asociative memories as one type of neural networks. We compare models of asociative memories with respect to recall of damaged spatial patterns. We deal with three types of asociative memories: Hopeld network know also as standard associative memory, hierarchical associative memory and cascade associative memory. Based on dened comparison criteria, we test the models on test data. Comparison and evaluation of the models based on test results concludes our work.
|
|
|
Data and their clustering
Pilmann, Jindřich ; Mrázová, Iveta (advisor) ; Kukačka, Marek (referee)
This master thesis descripes known methods of data clustering and examines their possible application on data from the area of social networks. Because of this we recapitulated how we describes objects using data and which technics we use for specifying their similarity. After that we recapitulated known clustering methods and possibilities of their validation. Consequently we have suggested method how perform clustering in the social networks and we tested it. We have applied this method on data from the area of international trade in 2008. We have evaluated and summarized results of this experiments. In the end of this work we have suggested possibilities of further research in this area.
|
guest ::
RSS feed.