Decentralized Finance (DeFi) lending platforms claim to use decentralized, community-driven governance. In practice, however, governance power remains concentrated among a limited number of users. This thesis investigates the behavior of users actively participating in governance in two major DeFi lending platforms, Aave and Compound, using a network science approach. Data covering governance actions and both governance and yield token transfers is combined in a multi-layered network model. By combining these layers, it is possible to investigate the characteristics of governance-active users. Specifically, how their governance activity relates to their behavior in the token transfer networks. The analysis studies correlations between token transfer patterns and governance participation, detects communities in governance token transfer networks, and uses a Susceptible-Infected (SI) process to estimate each user's structural influence in a spreading process to compare this to governance activity. The results show that governance token transfer volume strongly correlates with voting power, while other network features do not. Most governance-active users cluster together in a few communities in the governance token transfer network. Users with a greater simulated spreading potential only tend to partially hold higher actual voting power. Together, these findings suggest that DeFi governance is strongly shaped by token wealth and tight user clustering, highlighting that open design alone does not guarantee broad and equal participation.

Accurately predicting future interactions in temporal networks is essential for diverse applications, including disease modeling and transparent policy-making under regulatory requirements. While much research has focused on topological or temporal link prediction, relatively few studies address temporal weighted link prediction, where both the presence of future links and their intensity are critical. This paper develops an interpretable approach for forecasting weighted links over time, using contact networks that are aggregated into temporal weighted snapshots. By adapting the Self-Driven (SD) and Self-Cross-Driven (SCD) models introduced in Zou et al. (2023), we design prediction methods that incorporate memory decay while preserving explanatory clarity—avoiding the “black-box” nature common to deep learning.

We benchmark four methods - Baseline, SD, SCD, and an extended SCD* - across a range of physical (face-to-face) and virtual (online communication) contact networks, each exhibiting unique structural and dynamic properties when aggregated into temporal weighted networks. Our results show that the SCD model achieves a 32.86% reduction in Mean Squared Error (MSE) over the Baseline, while SD yields a 19.31% improvement. Moving from SD to SCD confers an additional 15.87% decrease in MSE, underscoring the benefits of incorporating both temporal and structural information. Although SCD* introduces further complexity, it did not show any consistent improvements in predictive performance across datasets.

Additional evaluations using the Area Under the Precision-Recall Curve (AUPRC) highlight dataset-specific variability in capturing active links. Correlation analysis reveals that MSE scales with average link weight distributions, whereas AUPRC correlates strongly with the proportion of active links per network snapshot. These findings emphasize that incorporating decay and structural context, in an interpretable manner, significantly enhances predictive accuracy, although parameter tuning remains crucial for different network topologies and interaction patterns.

Spreading processes are ubiquitous in nature and society, from the diffusion of information in social platforms to the spread of diseases within populations. Many real-world systems can be represented as networks, where a piece of information or a disease spreads along links connecting nodes. Different nodes and links often differ in their network properties and play distinct roles in a spreading process. Based on network properties of nodes or links, practitioners may be interested in identifying key nodes as the seed nodes to maximally diffuse a piece of information, or removing specific links to mitigate the spreading. In this thesis, we study the roles of a node or a link in a spreading process from three different perspectives and investigate how these roles relate to the properties of nodes and links within the underlying network.

We first explore how the network properties of a node can be used to predict the spreading influence of the node, defined as the average number of nodes that are ultimately infected when this node is the only seed node. Previous studies have shown that combining node properties derived from local and global topological information can better predict nodal influence than using a single metric. In Chapter 2, we investigate whether using relatively local information is sufficient for the prediction. To address this question, we define an iterative metric set by leveraging the iterative process used to derive classical nodal centralities like eigenvector centrality. The iterative metric set progressively incorporates more global information and is used as the feature set in a regression model to predict nodal spreading influence. The iterative metric set is then used as the feature set in a regression model to predict the spreading influence of a node. We find that the model using the iterative metric set that includes relatively local information achieves comparable prediction quality with the method that includes both local and global information, in various networks.

A spreading process can be mitigated by blocking social contacts, i.e., time-specific interactions. In Chapter 3, we investigate how the network properties of a contact are associated with the mitigation effect when the contact is blocked. We develop probabilistic contact blocking strategies, which remove contacts (temporal links) based on their properties in a temporal network, to mitigate the spread of a Susceptible-Infected-Recovered spreading process. The removal probability of a contact depends on a given centrality metric of the corresponding link in the time-aggregated network and the occurring time of the contact. We propose diverse link centrality metrics, and each centrality metric leads to a unique contact blocking strategy. Our results indicate that the spread of the epidemic is most effectively mitigated when contacts between node pairs that have fewer contacts and contacts that occur earlier in time are more likely to be removed.

The role of a link in a spreading process can also be reflected by the extent to which the link is used in the process. Many real-world systems may involve interactions among groups of more than two individuals and can therefore be represented as temporal higher-order networks. Chapter 4 explores the Susceptible-Infected threshold spreading process unfolding on temporal higher-order networks with two objectives: (1) to understand the contribution of each hyperlink to the spreading process, defined as the average number of nodes that are directly infected via the activation of the hyperlink starting from an arbitrary seed node, and (2) to investigate hyperlinks with what network properties tend to contribute more to the spreading process. This understanding is crucial for developing effective strategies to mitigate a spreading process. Given a temporal higher-order network, we propose to construct a weighted higher-order network, the so-called diffusion backbone, where the weight of each hyperlink denotes its contribution to the spreading process. We then systematically design centrality metrics for hyperlinks in a temporal higher-order network, where each centrality metric captures a specific property of the hyperlink within a temporal higher-order network and is used to estimate the ranking of hyperlinks by their weights in the backbone. We find and explain why certain centrality metrics can better estimate the contributions of hyperlinks under different parameters of the spreading process.

The last chapter reflects on the insights of this thesis and discusses possible future directions related to our research.

In criminal investigations, individuals may be connected to illicit activities by linking their personal phone to an otherwise anonymous, crime-related phone. Several methods have been published that use cell tower registrations to differentiate between same-user and different-user scenarios for the two phones. However, criminals may deviate in movement patterns and phone usage from the test subjects on which the methods are developed and evaluated. Whether the proposed methods are robust to such different behavioral profiles is unclear. The scarcity of readily available datasets on criminals' movements and phone usage further complicates this issue.

Lacking precise knowledge of the behavior of the population of interest, we propose a robustness analysis. Here, we present a tool for generating synthetic datasets, based on well-established models for the movement of individuals. We used the tool to generate data for a range of behavioral properties, encompassing variations in both underlying movement and phone usage. We evaluated three existing methods using our synthetic data. The first is a discriminatory approach that learns typical movement patterns and phone usage from a reference dataset. The second approach uses a model of cell tower behavior, making minimal assumptions on user behavior by choosing pairs of registrations close in time. The third is a generic statistical method for comparing event data. Additionally, we present a fourth method that combines the latter two, as conceptually, they use different aspects of the data.

Our analysis reveals that the discriminatory method performs best in a baseline scenario but is most sensitive to behavioral deviations. The cell tower method shows the lowest baseline performance yet exhibits the strongest resilience to variations. The generic model appears intermediate in terms of performance and sensitivity. Given the importance of robustness in evaluating evidence, we recommend using the combined approach, which is both reliable and effective across our defined variations.

A network, is defined as a collection of nodes interconnected by links. When this topology changes through time, we call it a temporal network. A specific class of networks, with only two types of nodes with no connections between one kind, is the bipartite network. An example is a telecommunications network, where nodes represent telecommunication base station and various mobile services like web-browsing, streaming etc. A link may exist only between a base station and a service. Moreover, each link is associated with a time-evolving weight, which represents the volume of the traffic between the corresponding base station and service over time. This weight associated with each link is also called the activity weight, with the link considered active only when the weight is non-zero. Predicting such a temporal weighted network in the future is crucial for telecommunications engineers, allowing for e.g., better traffic management. Prediction of the unweighted temporal network one step ahead, at time $t+1$, based on the network observed in the past between $[t-L-1; t]$, has been studied recently in contact networks. However, the prediction of weighted temporal networks, or equivalently, predicting the activity weight of each link, in the future has not been explored yet. Moreover, we also aim to uncover the mechanisms that enable the prediction of a weighted temporal network. We achieve this by devising several strategies that help us select the most relevant links within the network, whose activity weights in the past serve as the input for the interpretable, statistical learning algorithm, LASSO Regression, to predict the activity of a given target link at time $t+1$. The focus of the strategies is to capture a relationship of activity weights between the selected and target links. These selected links range from most active links (amount of timesteps the link weight is non-zero), those with largest activity weights or most similar to the target link using several metrics. In this thesis we apply this general methodology to two bipartite networks sourced from real world data and evaluate the performance of different strategies. Through the learned LASSO coefficients and prediction accuracy, we discover that past activity weight of a link is the best predictor for it's future weights. In terms of predicting power, most is coming from the past weights of the link we want to predict and one or two neighbouring links. Most of the selected links have minimal impact on the prediction accuracy. While different strategies of link selection excel in specific conditions, their improvement over the random link selection, is relatively low. The proposed method could be further applied to predict other weighted temporal networks with different properties to understand whether and how the the performance of link selection strategies depends on properties of the network to be predicted.

Wavelength Division Multiplexing (WDM) has been a widely-used multiplexing technique in large-scale optical telecommunication networks during the past few decades. Routing and Wavelength Assignment (RWA) is a fundamental process that involves determining the optimal path (routing) and assigning specific wavelengths to incoming data streams to ensure efficient and reliable transmission in WDM networks. This dynamic allocation of wavelengths allows for flexibility and adaptability in handling diverse traffic demands, making RWA a crucial mechanism for maximizing the capacity and performance of optical communication systems.
This project aims to find intrinsic factors that influence RWA performance in WDM and propose novel RWA approaches with enhanced performance. Existing dynamic RWA methods are reviewed from the literature and simulated in a self-built performance evaluation model. As the availability of every edge at every wavelength is constantly changing, we can transform the WDM network into a multi-layer temporal network structure. In order to uncover the essential reasons for the differences between the performances of the different methods, we investigate the multi-layer temporal network with graph theoretic analysis to explore correlations between specific multi-layer metrics and RWA performance. A few single-layer network connectivity metrics are applied in multi-layer networks including the number of connected components, the size of the largest components, the spectral radius, the algebraic connectivity, the effective resistance, the sum of betweenness, and the number of reachable node pairs. The experimental results show that the maximum value of spectral radius and algebraic connectivity over all layers are the best 2 multi-layer metrics describing the performance of the RWA methods.
Building upon this analysis, four new routing methods are proposed based on the previous methods and the two best-adapted multi-layer graph metrics, including the Least Spectral Radius Deduction (LSRD), Least Algebraic Connectivity Deduction (LACD), Least Hopcount and Congestion Path (LHCP) and Congestion Weighted Shortest Path (CWSP) methods. All new methods are fitted in the evaluation model and it has been proven that the CWSP method has better performance compared to all other RWA methods based on its improvement of selected multi-layer graph metrics.

Temporal networks, like physical contact networks, are networks whose topology changes over time. However, this representation does not account for group interactions, when people gather in groups of more than two people, that can be represented as higher-order events of a temporal network. The prediction of these higher-order events is usually overlooked in traditional temporal network prediction methods, where each higher-order event of a group is regarded as a set of pairwise interactions between every pair of individuals within the group. However, pairwise interactions only allow to partially capture interactions among constituents of a system. Therefore, we want to be able to predict the occurrence of such higher-order interactions one step ahead, based on the higher-order topology observed in the past, and to understand which types of interactions are the most influential for the prediction. We find that the similarity in network topology is relatively high at two time steps with a small time lag between them and that this similarity decreases when the time lag increases. This motivates us to propose a memory-based model that can predict a higherorder temporal network at the next time step based on the network observed in the past. In particular, the occurrence of a group event will be predicted based on the past activity of this target group and of other groups that form a subset or a superset of the target group. Our model is network-based, so it has a relatively low computational cost and allows for a good interpretation of its underlying mechanisms. We propose as a baseline the memory-based method for the traditional pairwise network prediction problem. In this baseline model, the predicted higher-order events at a prediction time step are then deduced from the predicted pairwise network at the same prediction time step. We evaluate the prediction quality of all models in eight real-world physical contact networks and find that our model outperforms the baseline model. We also analyze the contribution of group events of different orders to the prediction quality. We find that the past activity of the target group is the most important factor for the prediction. Moreover, the past activity of groups of a larger size has, in general, a lower impact on the prediction of events of an arbitrary size than groups of a smaller size.

This research is a case study investigating the effect of participation in the development process of a network-based scientist-journalist recommender system on the mental model of digital innovation of a team of communication professionals from Delft University of Technology and Naturalis Biodiversity Center.

Communication professionals at research institutes are tasked with connecting scientists and journalists. The recommender system supports this process by recommending scientist-journalist connections based on data from previous collaborations. A scientist collaboration network, a journalist collaboration network and a scientist-journalist collaboration network are combined into a multilayer network. A recommender system is designed based on centrality metrics in the scientist and journalist collaboration networks and distance metrics in the multilayer network. In contrast to traditional link prediction problems - which aim to predict what links are most likely to form in the network - the problem in this thesis is how to recommend the most likely link for a single node, i.e. the most likely scientist links for a given journalist or most likely journalist links for a given scientist. A novel evaluation method is created to evaluate the performance of the recommender system.

The development of this system is used as a vessel to research how participation in a digital development process affects the mental model of digital innovation. This research contributes to addressing the lack of understanding of how to develop a mental model that facilitates innovation in the context of digital transformation. Three themes were identified in their mental model change: The extent to which innovation requires involvement, the complexity of innovation processes and what outcomes can realistically be expected of a digital innovation process. The team went from a model of digital innovation as 'a mysterious black box' - something external, where they could hand in a list of requirements and walk away with a digital tool - to a 'super puppy' that can do remarkable things, but has to be trained and interacted with to get a desired effect.

This thesis introduces the concept of Fairness Aware Influence (FAI), which is dependent on prevalence and fairness. Prevalence of a given seed node is the number of infected nodes. Fairness can be defined differently based on the application or problem statement. In this case, the fairness is defined as the variation in the fraction of infections in all the communities. This variation is measured using standard deviation (SD). A lower SD corresponds to better fairness. FAI for a given seed node is defined as the ratio of prevalence to fairness, where a higher FAI score corresponds to higher prevalence and lower SD.

The primary objective of this thesis is to study how network properties such as degree and community size, relate with FAI. Network properties are measured using centrality metrics, which are categorized into two types.
The first type, referred to as "simple" or classic centrality metrics, do not account for community information. The second type, known as community-aware centrality metrics, incorporate community information but were not originally designed for FAI. These serve as baselines for ranking nodes in terms of FAI. Thus, two new classes of metrics are designed specifically for FAI in the attempt to perform better than the baselines.

Six real world networks are employed to evaluate the metrics. Local centrality and Community-Hub-Bridge are found to be good baselines in their respective categories, and the newly proposed metrics surpass the existing ones at the epidemic threshold. Additionally, a discussion is presented to compare and analyze these metrics, considering their performance under varying infection rates using an SIR infection spreading model.

In real life, the phenomenon of delayed payment of invoices often occurs. Malicious delayed payment may even lead to the break of capital flow of companies. Debtor payment prediction according to the historical payment behaviour of debtors can help give companies more insight into the risk of late payment. The debtor payment prediction is important for financial risk analysis because companies can better manage their cash flow to avoid the risk of capital chain breakage if they know when the invoice they present will be paid in advance. However, real-world data is far from perfect. With the inflow of various free-form data streams, it is challenging to integrate invoice payment records regarding debtors' historical payment behavior. As a result, the debtor payment prediction model sometimes performs sub-optimal because sufficient historical payment data cannot be retrieved for debtors. To improve debtor payment predictions, the data sparseness problem is studied to be addressed through entity resolution technology. Instead of fine-tuning a more complex model or adding new financial features, the prediction performance is studied to be improved by mining the information hidden in the data. More specifically, an end-to-end entity resolution workflow is proposed for invoice payment data to match debtors representing the same entity so that more historical payment data is available for the debtor that is known to be matched with other debtors. With the incorporation of entity resolution, the debtor payment prediction is verified to be improved in this project.

The COVID-19 pandemic is influencing the Dutch economy heavily. More so, small and medium-sized enterprises, also known as SMEs, are notoriously unstable and as a result, could be even more heavily affected by the coronavirus outbreak. The first major lockdown in The Netherlands was instated on March 23, 2020, which introduced several new measures, such as the prohibition of gatherings, the closing of food and beverage outlets, and the prohibition of all contact-based professions.

In such a time of economic instability as caused by the coronavirus outbreak, it is very useful for a company to know in what financial state they are going to be such that they can actively take precautions, such as liquidating their assets or decreasing their expenses. The financial state of a company is often reflected using Key Performance Indicators, or KPIs for short. These KPIs include metrics like the revenue, cost, and cash flow of a company. The forecasting of these KPIs can help a company in informing in what financial state they are going to be and are usually done using historical data of the company. Whereas the decrease in economic activity of business partners of a company is not reflected in the historical KPI data of the company itself, it can be seen in a network of companies that indicates whether there exists a relationship between two companies by using data on monetary transactions between companies. For this reason, we think that enriching historical KPI data using node features extracted from a dynamic network of companies can help improve the quality of KPI predictions during a period of economic instability such as the COVID-19 pandemic.

This thesis answers the question of whether we can use utilize a dynamic network of SMEs to improve the quality of KPI predictions during the COVID-19 lockdown. To answer this question, we first focus on creating a dynamic network consisting of SMEs and the transactions between them out of unstandardized data by proposing a novel, lightweight entity resolution algorithm that is used to find a mapping between companies. The resulting network is analyzed, and we found that the effects of the coronavirus lockdown are visible in the network. Next, we examine several KPIs, such as the revenue or the cash flow of a company, and we found that we can also see the effects of the COVID-19 lockdown in several KPIs. Lastly, this thesis describes an analysis of whether node features can be used to improve the quality of the forecasting of several of these KPIs, where we found that node features such as the degree and clustering coefficient of a node can indeed help with improving KPI forecasting under certain conditions.

The study of epidemic spreading processes on contact based complex networks has gained a lot of traction in recent years. These processes can entail a variety of problems such as disease spreading, opinion spreading in social networks or even airport congestion in airline networks. One of the key tasks in this area of research and also of this work is the prediction of the final epidemic size of an outbreak in a network, given that a contagion process has been initiated by a seed node. More specifically, the objective is to predict to which extent a seed node is able to activate the rest of the nodes in a network using a supervised learning model. In this work, this task is termed: “The node influence prediction problem”. Being able to predict the epidemic footprint of a node allows the design of robust networks and the application of efficient intervention strategies.

Recently, a limited number of studies have proposed methods on how to utilize classical network topology based features to predict the nodal influence. However, two main challenges still persist: (1) individual topology based features do not fully capture the information of a node and (2) it is tedious to obtain these features for nodes in large scale networks. As an alternative solution, this work aims to utilize network embedding based features instead, where feature vectors of the nodes are learned from the network topology. In this research we assume that the network topology and the nodal influence of a small subset of the nodes are known. We then proceed to show how to build and optimize a machine learning framework where only 10\% of the nodes are used as training data and which could even be applicable on large scale networks. Additionally, we also demonstrate why network embedding based features are applicable in the node influence prediction task.

The findings show that node pairs which are closer in proximity in the network, are also embedded closer in the embedding space (exhibiting a higher similarity). The performance evaluation of the predictive models illustrate that network embedding based features can compete with classical topological metrics, despite the disadvantage of their higher dimensionality. This is achieved by combining the embedding features with individual low cost topology features such as the degree.

The value of data has increased enormously over the last couple of years. Many datasets contain valuable information that can, for example, be used to make forecasts. In this thesis, the dataset of a company in the self-storage industry is analyzed. The company offers customers rental storage facilities, such as lockers, rooms, containers at several locations in the Netherlands. The rooms, also called units, are characterized by three features: location, volume, and floor. The dataset contains for each storage facility the entire renting history of each self-storage unit. For the company, it is of interest to understand and predict the demand of its customers, such that it can flexibly adjust the prices of the rooms and services based on demand. Beyond, forecasting users' demand for various types of products is a common and essential problem in many different domains (e.g., recommendation systems, transportation systems).

This thesis aims to predict the demand for the next week by applying white-box and black-box models. The problem is represented as a temporal weighted bipartite network prediction problem. Specifically, the goal is to predict the network structure at a time T+1 based on the bipartite network observed at time T-k+1, T-k+2, ..., T, where k is an integer and needs to be optimized such that the prediction error is minimized. By analyzing the data in its temporal dimension, using the autocorrelation and cross-correlation among different storage locations, floors and volumes, it was shown that the autocorrelation is high and the cross-correlation is low. This suggests that the temporal bipartite network is possibly predictable.

We have explored different state-of-the-art predictive techniques. Markov chain model, LSTM, and ConvLSTM have been selected because of their fundamental difference in the way they learn and predict. A performance comparison is given where the techniques have been applied on the storage data, and it shows that LSTM outperforms the Markov chain and ConvLSTM based on the following evaluation metrics: RMSE, MAE, and accuracy. According to our dataset, higher predictability was achieved when only the data of a single link was exploited. The Markov chain and the LSTM utilize the information of a single link to predict. On the contrary, the ConvLSTM utilizes the information of the entire network to predict. The low cross-correlation between the links explains why the LSTM outperforms the ConvLSTM. The ConvLSTM tries to capture spatio-temporal dependencies, while this, in general, does not contain much valuable predictive information. Thus, the model is introduced to more noise, making it harder to predict accurately. The LSTM also outperforms the Markov chain model, which is used as a baseline method. This proves that it is beneficial to use a complex deep learning model for this dataset to predict. However, the Markov chain performs comparable to the ConvLSTM, showing that a black-box model does not always outperform a white-box model. This emphasizes that the most suitable predictive algorithm depends on the statistical properties of the dataset.

The theoretical upper bound of the predictability of the network is computed. It is the upper bound that can be used to compare the realized performance to the maximum achievable prediction performance for any predictive algorithm. The difference between the performance of the best performing algorithm to our dataset, LSTM, and the theoretical upper bound is still large, indicating that there is still room for improvement.

Small and medium enterprises (SMEs) bring a significant contribution to each country’s economy, ensuring both a high employment rate and financial prosperity. Despite their essential role, these type of companies presents a higher vulnerability to default than large corporates. The default event implies that the SME could not properly reimburse the money owned to its suppliers. If their default could be forecast, experts could adopt measures in order to prevent it or its impactful consequences. For this reason, developing a credit scoring model which is able to predict which SMEs are endangered of default is crucial. There are plenty of credit scoring prediction models available in the literature. However, most of them are only relying on the financial status of one company. In this work, we are presenting a novel method for credit scoring prediction which does not only take into account the SMEs’ financial situation, but also their position and role within a transactional network. A transactional network is a graph, whereby the nodes are represented by SMEs and the edges show that between two nodes there should be at least one transaction. In our work, we highlight the limitations that traditional models face and provide an alternative to overcome them. Furthermore, our findings show that combining network features with financial features could lead to a more accurate prediction and increase the robustness of the model. For this reason, we believe that the transactional network carries significant insights and could be a meaningful addition to financial based credit scoring prediction models.

Healthcare industry is an ever-emerging field in the 21st century. The statistics from Centers for Medicare services (CMS) [15] website shows that in 2017, in USA, the healthcare industry has invested USD 7.4 billion for research collaborations with physicians. These research collaborations in CMS is spread across multiple facets ranging from contributing towards research, developing new products, running clinical trials, royalty, licences, patents, providing innovative ideas etc. In this thesis, we make an assumption that, a relationship between investment made by the healthcare company and the research profile of a physician exists. We aim to answer, what could possibly be the relation between payments made by the healthcare company; on the physicians and the research profile of the physicians. The research profile of a physician includes factors like h-index, years of research experience, citation count, physician citation network, etc. To validate this relationship we use the data corresponding to returns of the healthcare company. Some of the measures of returns, from a research collaboration between physicians, include, innovation, good will, fame, market share, Return on Investment (ROI). We choose ROI, as a measure of return, due to the availability of data and to determine the relationships mentioned above.

To understand the above mentioned relationship, we explore two types of relationships, i.e., direct and indirect relationship. In the direct relationship, we use multiple regression model to understand the direct relationship between the research payments and the research profile of the physicians, by making an assumption that the research profile of the physician describes the research quality of the physician. In the indirect relationship, we make use of a weighted physician co-author citation network, to investigate the relationship between his/her co-author interactions and the research payments he/she received from the healthcare company. To accomplish this, we developed a spreading process that models influence diffusion in a physician citation network. The diffusion of influence is dependent on the topological property of the node in the network.

Our models are an exemplification of the direct and indirect relationships, which exists in the real world. To evaluate our models, we use metrics such as coefficient of determination, Pearson correlation coefficient and Spearman's rank correlation. Once the models were evaluated, we inferred that the model for indirect relationship, explains the relationship between research profile, investments, and return 96.3% more than the model for direct relationship. We also perform a deep analysis, by investigating the nature of the distributions of the variables and scatter plots to understand the relationship between the variables used in understanding the direct and indirect relationship. Lastly, we propose two different redistribution methods, where the original payments made to physicians are redistributed to a potential group of physicians in the physician citation network. These potential physicians are identified based on their topological property. In consequence, our redistribution methods may inspire the healthcare companies, to design their future investments made to physicians.

Link prediction in complex networks has attracted increasing attention. The link prediction algorithms can be used to retrieve missing information, identify spurious interactions, capturing net- work evolution, and so on. Recently, network embedding has been proposed as a new strategy to embed network into low-dimensional vector space. By embedding nodes into vectors, the link pre- diction problem can be converted into a similarity comparison task. Nodes with similar vectors are more likely to connect. Some traditional network embedding methods include matrix factorization, random walk paradigm and deep neural network models.
In this thesis, we propose SISNE, a diffusion based paradigm for node embedding, applying Susceptible-Infected-Susceptible (SIS) model to extract node neighborhood structure. Both random walk based algorithms and our proposed method sample node sequences as input and feed them into a Skip-gram model, a representative language model that embeds words into vectors. Specially, our proposed model provides flexibility to explore the network topology by operating in- formation spreading on networks. Another contribution of the proposed model is that SISNE takes into the account of the evolving nature of complex networks. To verify the efficacy, we conduct experiments on missing link prediction task and show that our SIS diffusion based model outperforms other state-of-the-art network embedding algorithms across all four empirical datasets, reaching a maximal 7% improvement. Importantly, even when the input size is small, the performance remains stable whereas other baseline models drop dramatically, which indicates that our proposed model is less sensitive to input size and suggests that the model is applicable to large-scale networks. Moreover, we further show that as long as the infection probability β is larger than the threshold value of the diffusion model, we can obtain a relatively high performance for link prediction task. Taken together, our work has shown great effectiveness and efficiency in learning embeddings in temporal networks.

This thesis concludes the final report which has been performed for the company Gymplanner. At the same time it is the summary for our bachelor end project, thus it is considered the bachelor thesis for our Computer Science Bachelor Degree

There is an increasing attention towards link prediction in complex networks both in physical and computer science communities. Particularly Online Social Networks (OSNs) are
becoming the most popular platforms for information sharing, content creation and communication between users on the Internet. However, most of the research was done considering only a static snapshot of the network and without using relevant information from
other types of activities.
In that direction, the present thesis proposes a novel method for link prediction using temporal information in Stack Overflow with the assistance of interactions from Github. The
developed multilayer network enhanced with temporal interactions is aiming to improve
the performance of the prediction compared to the traditional methods while the design
choices intend to investigate the evolution of the network through time. In the end, the
generalized framework could be used not only to make accurate link prediction that translate to human interactions over time, but also as a tool to characterize the behavior of the
users in the targeted network.

Keeping a steady cash flow is one of the biggest if not the biggest problem that Small to Medium Enterprises (SMEs) deal with daily. Within the different types of cash flow, Accounts Receivable (AR) classifies the balance of money that needs to be paid by the company's customers. In the most typical case, after receiving goods or services, the customer receives an invoice with the amount that is owed to the supplier. However, this often does not happen before the aforementioned date, meaning that the invoice is often paid late. Intervention requires resources and over-intervention could cause unwanted customer dissatisfaction. Knowing whether an invoice is going to be paid late can be vital information. Current methods of late payment prediction focus only on the history between the seller and the buyer and are unusable when this history is not present. Intuitively, one's business depends on the relationships and transactions that it has with its neighbors. Suggesting that neighbor behavior could be useful when predicting the cash flow of a company. Unfortunately, this type of information is not always given and needs to be data mining from non-relational data. This work presents a method for building a relational network of SMEs using entity resolution and improving the current state of the art of late payment prediction using features extracted from the graph.

Information systems, such as information retrieval machines and recommendation systems, utilize various user information and history behaviors to provide personalized content to users. However, a debate on whether the personalization in information systems can trigger the online echo chamber effect has emerged. The online echo chamber effect describes the situation that Internet users are segregated into groups based on common interests or opinions and their existing views or confirmation bias are reinforced by repetition. Based on the idea that the strong community structure of the user network suggests the emergence of the echo chamber effect, we propose a brand new methodology based on temporal community evolution to detect the echo chamber effect in recommender systems. A two-layer temporal user network is constructed, with the first layer representing the user taste similarity and the second layer encoding potential information flows between users. Then, we apply an estrangement confinement based algorithm to detect the temporal communities in the two-layer temporal network. Our experiment results on the MovieLens dataset suggest the emergence of the echo chamber effect. Moreover, we find that the echo chamber effect is becoming more remarkable over time. In addition, we observe that some users tend to stay in one community over time. These users are potentially affected by the echo chamber effect and have a higher mean node strength in both network layers.