Umeå University's logo

Predicting future interactions or novel links in networks is an indispensable tool across diverse domains, including genetic research, online social networks, and recommendation systems. Among the numerous techniques developed for link prediction, those leveraging the networks' community structure have proven highly effective. For example, the recently proposed MapSim predicts links based on a similarity measure derived from the code structure of the map equation, a community-detection objective function that operates on network flows. However, the standard map equation assumes complete observations and typically identifies many small modules in networks where the nodes connect through only a few links. This aspect can degrade MapSim's performance on sparse networks. To overcome this limitation, we propose to incorporate a global regularization method based on a Bayesian estimate of the transition rates along with three local regularization methods. The regularized versions of the map equation compensate for incomplete observations and mitigate spurious community fragmentation in sparse networks. The regularized methods outperform standard MapSim and several state-of-the-art embedding methods in highly sparse networks. This performance holds across multiple real-world networks with randomly removed links, simulating incomplete observations. Among the proposed regularization methods, the global approach provides the most reliable community detection and the highest link prediction performance across different network densities. The principled method requires no hyperparameter tuning and runs at least an order of magnitude faster than the embedding methods.

This work was conducted by the UArctic Thematic Network on Research Analytics and Bibliometrics. It was supported by Global Affairs Canada through the Global Arctic Leadership Initiative.

Providing necessary knowledge and skills for engineering students to become successful professionals is a tricky task. Besides disciplinary knowledge, e.g., communication skills, ability to work in teams, and international experience are often mentioned as important. Regarding internationalization, most engineering programs in Sweden rely on either student exchange or low-level internationalization-at-home, such as international literature and lecturers. This paper explores sustainable international experiences for students on their home turf provided through an international interdisciplinary collaboration where engineering students in Sweden and marketing students in Australia work together on a project. The setup simulates a consultancy firm with development and marketing offices in different countries that cooperate to launch an application for the Australian market. The paper is based on interviews and surveys with students and teachers participating in this, since 2017, ongoing project.

Findings reveal that students encountered several challenges that are hard to simulate in an ordinary university setting, e.g., language barriers, cultural differences, time differences, differences between disciplines, and varying work habits and values. The results also highlight opportunities such as learning from each other's perspectives and expertise, developing a more professional approach, presenting to people from other industry backgrounds, and gaining a better understanding of different cultures. The results show that the students gain professional experience that is of great value for their future profession. From a teacher's perspective, the paper discusses important issues when setting up an international inter-disciplinary collaboration, e.g., alignment of exercises, building a common ground, and the need for flexibility.

Programming has gradually become an essential skill for engineers and scientists across disciplines and is an important part of the CDIO Syllabus covering fundamental knowledge and reasoning. Recently, there has been a shift away from introductory programming languages like C and Java towards Python, especially in programs where the focus lies on handling and analysing large quantities of data, such as energy technology, biotechnology, and bioinformatics. This paper illustrates the successful setup of a one-week-long introductory Python programming course with a hands-on approach. Given the limited time, a challenge is how to effectively teach students a meaningful set of skills that enables them to self-guide their future learning. Moreover, since the course does not include any summative assessment, we need other means of measuring students’ learning and guiding course development. We address these challenges by coupling short lectures with short quizzes for formative assessment, adding another learning activity to the course. We find that, in the absence of summative assessment, short, frequent quizzes with immediate feedback are an excellent tool to track the learning of a class as a whole. Students report that the quizzes, albeit challenging, improved their understanding of programming concepts, made them aware of potential mistakes, and were a fun learning experience. Furthermore, the results from this paper illustrate how a new programming language can be taught to students without prior programming skills in a short period of time. We summarise our lessons learnt for designing and integrating quizzes in short-format programming courses.

Higher education should provide learning situations that prepare students for a future profession and make them world-ready. This paper reports insights from an international interdisciplinary collaborative project aiming to create learning experiences that are close to a professional situation. The collaboration setup simulates a setting of a digital agency with a development team in Sweden and a marketing team in Australia working together to solve a task. The collaborative project has been active since 2017, completing its fifth iteration in 2021. Postcourse survey results show that the students felt that a real situation was created with a high level of collaboration and commitment, internationalization, well selected digital collaborative tools, and that an interdisciplinary community of practice was created among the students.

To measure node importance, network scientists employ centrality scores that typically take a microscopic or macroscopic perspective, relying on node features or global network structure. However, traditional centrality measures such as degree centrality, betweenness centrality, or PageRank neglect the community structure found in real-world networks. To study node importance based on network flows from a mesoscopic perspective, we analytically derive a community-aware information-theoretic centrality score based on network flow and the coding principles behind the map equation: map equation centrality. Map equation centrality measures how much further we can compress the network's modular description by not coding for random walker transitions to the respective node, using an adapted coding scheme and determining node importance from a network flow-based point of view. The information-theoretic centrality measure can be determined from a node's local network context alone because changes to the coding scheme only affect other nodes in the same module. Map equation centrality is agnostic to the chosen network flow model and allows researchers to select the model that best reflects the dynamics of the process under study. Applied to synthetic networks, we highlight how our approach enables a more fine-grained differentiation between nodes than node-local or network-global measures. Predicting influential nodes for two different dynamical processes on real-world networks with traditional and other community-aware centrality measures, we find that activating nodes based on map equation centrality scores tends to create the largest cascades in a linear threshold model.

Node similarity scores are a foundation for machine learning in graphs for clustering, node classification, anomaly detection, and link prediction with applications in biological systems, information networks, and recommender systems. Recent works on link prediction use vector space embeddings to calculate node similarities in undirected networks with good performance. Still, they have several disadvantages: limited interpretability, need for hyperparameter tuning, manual model fitting through dimensionality reduction, and poor performance from symmetric similarities in directed link prediction. We propose MapSim, an information-theoretic measure to assess node similarities based on modular compression of network flows. Unlike vector space embeddings, MapSim represents nodes in a discrete, non-metric space of communities and yields asymmetric similarities in an unsupervised fashion. We compare MapSim on a link prediction task to popular embedding-based algorithms across 47 networks and find that MapSim's average performance across all networks is more than 7% higher than its closest competitor, outperforming all embedding methods in 11 of the 47 networks. Our method demonstrates the potential of compression-based approaches in graph representation learning, with promising applications in other graph learning tasks.

We live in a highly-connected world and find networks wherever we look: social networks, public transport networks, telecommunication networks, financial networks, and more. These networks can be immensely complex, comprising potentially millions or even billions of inter-connected objects. Answering questions such as how to control disease spreading in contact networks, how to optimise public transport networks, or how to diversify investment portfolios requires understanding each network's function and working principles.

Network scientists analyse the structure of networks in search of communities: groups of objects that form clusters and are more connected to each other than the rest. Communities form the building blocks of networks, corresponding to their sub-systems, and allow us to represent networks with coarse-grained models. Analysing communities and their interactions helps us unravel how networks function.

In this thesis, we use the so-called map equation framework, an information-theoretic community-detection approach. The map equation follows the minimum description length principle and assumes complete data in networks with one node type. We challenge these assumptions and adapt the map equation for community detection in networks with two node types and incomplete networks where some data is missing. We move beyond detecting communities and derive approaches for how, based on communities, we can identify influential objects in networks, and predict links that do not (yet) exist.

Vi lever i en värld som blir mer och mer sammanlänkad. Vart vi än tittar hittar vi nätverk: sociala nätverk, kollektivtrafiknätverk, telekommunikationsnätverk, finansiella nätverk och så vidare. Dessa nätverk kan vara oerhört komplexa och omfatta potentiellt miljoner eller till och med miljarder sammankopplade objekt. För att kunna besvara frågor som: hur kontrollerar vi sjukdomsspridning i kontaktnät, hur optimerar vi kollektivtrafiksnätverk eller hur diversifierar vi investeringsportföljer, krävs det att vi förstår varje nätverks funktion och principer.

Nätverksforskare analyserar strukturen i nätverk i jakt på kluster: grupper av objekt som är mer kopplade till varandra än till resten av nätverket. Kluster utgör byggstenarna, eller delsystemen, i nätverken och låter oss representera dessa med förenklade modeller. Att analysera kluster och deras interaktioner hjälper oss att ta reda på hur nätverk fungerar.

I denna avhandling vidareutvecklar vi den så kallade kartekvationen, en informationsteoretisk klusterdetekteringsmetod. Kartekvationen följer principen om minsta beskrivningslängd och förutsätter fullständiga data i nätverk som bara består av en typ av noder. Vi utmanar dessa antaganden och anpassar kartekvationen för klusterdetektering i nätverk som består av två typer av noder och ofullständiga nätverk där viss data saknas. Vi dyker också djupare in i kluster och härleder lösningar för hur vi, baserat på kluster, kan identifiera inflytelserika objekt i nätverk och förutsäga länkar som (ännu) inte existerar.

Wir leben in einer hochgradig vernetzten Welt und finden Netzwerke wo auch immer wir hinschauen: soziale Netzwerke, öffentliche Verkehrsnetze, Telekommunikationsnetze, Finanznetzwerke und mehr. Diese Netzwerke können immens komplex sein und potenziell Millionen oder sogar Milliarden miteinander verbundener Objekten umfassen. Um beantworten zu können, wie wir die Ausbreitung von Krankheiten in Kontaktnetzwerken kontrollieren, öffentliche Verkehrsnetze optimieren oder Anlageportfolios diversifizieren können, müssen wir die Funktionsweise und Arbeitsprinzipien dieser Netzwerke verstehen.

Netzwerkwissenschaftler analysieren die Struktur von Netzwerken auf der Suche nach Communities: Gruppen von Objekten, die Cluster bilden und stärker miteinander verbunden sind als mit dem Rest. Communities repräsentieren die Bausteine von Netzwerken, entsprechen ihren Subsystemen und erlauben es uns, Netzwerke mit vereinfachten Modellen darzustellen. Communities und ihre Interaktionen untereinander zu verstehen hilft uns dabei, zu enträtseln, wie Netzwerke funktionieren.

In dieser Doktorarbeit verwenden wir die sogenannte Kartengleichung, ein informationstheoretischer Ansatz zur Community-Erkennung. Die Kartengleichung folgt dem Prinzip der minimalen Beschreibungslänge und nimmt an, dass Netzwerke einen Knotentypen haben und ihre zugrundeliegenden Daten vollständig sind. Wir stellen diese Annahmen infrage und passen die Kartengleichung zur Community-Erkennung in Netzwerken mit zwei Knotentypen und unvollständigen Daten an. Darüber hinaus leiten wir Ansätze ab, die, basierend auf Communities, einflussreiche Objekte in Netzwerken identifizieren und (noch) nicht existierende Verbindungen zwischen Objekten vorhersagen.

Up-regulation of long non-coding RNAs (lncRNAs), colon-cancer associated transcript (CCAT) 1 and 2, was associated with worse prognosis in colorectal cancer (CRC). Nevertheless, their role in predicting metastasis in early-stage CRC is unclear. We measured the expression of CCAT1, CCAT2 and their oncotarget, c-Myc, in 150 matched mucosa-tumour samples of early-stage microsatellite-stable Chinese CRC patients with definitive metastasis status by multiplex real-time RT-PCR assay. Expression of CCAT1, CCAT2 and c-Myc were significantly up-regulated in the tumours compared to matched mucosa (p < 0.0001). The expression of c-Myc in the tumours was significantly correlated to time to metastasis [hazard ratio = 1.47 (1.10–1.97)] and the risk genotype (GG) of rs6983267, located within CCAT2. Expression of c-Myc and CCAT2 in the tumour were also significantly up-regulated in metastasis-positive compared to metastasis-negative patients (p = 0.009 and p = 0.04 respectively). Nevertheless, integrating the expression of CCAT1 and CCAT2 by the Random Forest classifier did not improve the predictive values of ColoMet19, the mRNA-based predictor for metastasis previously developed on the same series of tumours. The role of these two lncRNAs is probably mitigated via their oncotarget, c-Myc, which was not ranked high enough previously to be included in ColoMet19.

In today’s global society, international experience isimportant for students studying all subjects. This paper providesinsights and learnings from a long-term project with the purposeto provide international interdisciplinary experience forengineering students in Sweden as well as for marketing studentsin Australia. The paper discusses the design of the latest iterationof a long-term collaborative project that enables students who donot have the opportunity to engage in exchange studies in aprofessional international setting. The main objective of this paperis to give inspiration and a starting point to the implementation ofinternational learning experiences as an integrated part ofstudents’ education.