Data Science Research Centre

Contact: Gerardo Iñiguez, gerardo.iniguez [at] tuni.fi 

Universal behaviour in rank dynamics and other complex phenomena

Many complex systems develop rankings of their elements that emerge from networked interactions. These rankings evolve according to system-dependent mechanisms of interaction and reflect the relevance of elements in performing a function in the system. By analysing ranking data on many social, biological, and economic systems, as well as simple models of rank dynamics, we will explore generic features of rank stability that allow us to model and predict patterns of ranking behaviour across a variety of complex systems.

G. Iñiguez et al. Dynamics of ranking. Nature Communications 13, 1646 (2022). DOI: https://doi.org/10.1038/s41467-022-29256-x. arXiv: https://arxiv.org/abs/2104.13439

J. A. Morales et al. Rank dynamics of word usage at multiple scales. Frontiers in Physics 6, 45 (2018). DOI: https://doi.org/10.3389/fphy.2018.00045. arXiv: https://arxiv.org/abs/1802.07258

J. A. Morales et al. Generic temporal features of performance rankings in sports and games. EPJ Data Science 5, 33 (2016). DOI: https://doi.org/10.1140/epjds/s13688-016-0096-y. arXiv: https://arxiv.org/abs/1606.04153 

Underlying mechanisms of collective behaviour in social systems

With dynamical systems analysis of simple models of social network evolution, we will estimate how much of the structure and dynamical patterns in friendship and communication networks is a result of underlying mechanisms of social interaction, explaining the rise of segregated groups in society, among other collective social phenomena. This will also allow us to infer attitudinal space embeddings for political social networks.

A. F. Peralta et al. Multidimensional political polarization in online social networks. Physical Review Research 6, 013170 (2024). DOI: https://doi.org/10.1103/PhysRevResearch.6.013170. arXiv: https://arxiv.org/abs/2305.02941

G. Iñiguez et al. Universal patterns in egocentric communication networks. Nature Communications 14, 5217 (2023). DOI: https://doi.org/10.1038/s41467-023-40888-5. arXIv: https://arxiv.org/abs/2302.13972

A. Asikainen et al. Cumulative effects of triadic closure and homophily in social networks. Science Advances 6, eaax7310 (2020). DOI: https://doi.org/10.1126/sciadv.aax7310. arXiv: https://arxiv.org/abs/1809.06057 

Uncovering the temporal evolution on/of techno-social networks 

We will use large-scale data from online social platforms to study and predict the temporal evolution of social contagion processes and cascading behaviour related to the use of innovations and new digital markets. By considering data features like weighted or multiplex social interactions, product competition, and temporal social networks, we will aim at increasing the explanatory and predictability power of models of temporal networks, information diffusion, and social contagion. 

S. Unicomb et al. Dynamics of cascades on burstiness-controlled temporal networks. Nature Communications 12, 133 (2020). DOI: https://doi.org/10.1038/s41467-020-20398-4. arXiv: https://arxiv.org/abs/2007.06223

S. Unicomb et al. Threshold driven contagion on weighted networks. Scientific Reports 8, 3094 (2018). DOI: https://doi.org/10.1038/s41598-018-21261-9. arXiv: https://arxiv.org/abs/1707.02185

Z. Ruan et al. Kinetics of social contagion. Physical Review Letters 115, 218702 (2015). DOI: http://dx.doi.org/10.1103/PhysRevLett.115.218702. arXiv: https://arxiv.org/abs/1506.00251

Opinion formation, algorithmic bias, and deception on coevolving social networks

Opinions synthesise our perceptions and the knowledge we have of the external world, other people, and ourselves. To better understand the impact opinions have on social interactions and decision-making, we will explore data from small controlled experiments and analyse idealised models of opinion formation, deception, and network change, potentially under the effect of algorithmic bias, letting us gauge how opinions influence the structure and dynamics of off- and online society. 

A. F. Peralta et al. The effect of algorithmic bias and network structure on coexistence, consensus, and polarization of opinions. Physical Review E 104, 044312 (2021). DOI: https://doi.org/10.1103/PhysRevE.104.044312. arXiv: https://arxiv.org/abs/2105.07703

G. Iñiguez et al. Effects of deception in social networks. Proceedings of the Royal Society B 281, 20141195 (2014). DOI: http://dx.doi.org/10.1098/rspb.2014.1195. arXiv: https://arxiv.org/abs/1406.0673

G. Iñiguez et al. Opinion and community formation in coevolving networks. Physical Review E 80, 066119 (2009). DOI: http://link.aps.org/doi/10.1103/PhysRevE.80.066119. arXiv: https://arxiv.org/abs/0908.1068v2

Contact: Jonathon Taylor, jonathon.taylor [at] tuni.fi

Image processing and analysis for building energy and climate adaptation modelling.

Building physics models can be used to simulate the energy and climate resilience of a building, which is essential to understand how we can adapt our buildings to be more resilient to the future climate and energy efficient. However, most studies in Finland rely on example case study buildings, or example buildings which do not represent the true range of the building stock. This proposal seeks to scrape floorplan data from real estate websites (for example Etuovi), vectorise the floorplans, and classify rooms according to their type. It will then explore intelligent/AI ways to aggregate floorplans into a representative ‘average’ buildings that we can use for simulations

Using distributed networks of low cost citizen science sensors to identify exposures to environmental hazards.

The growth of IoT-connected sensor devices (e.g. Netatmo personal weather stations or low cost air pollution sensors) means there is now vast amounts of citizen-acquired data on environmental conditions in cities for a more dense network that the official stations we have normally relied on. This big data can reveal information on how we can design our urban environments to be more healthy and climate-resilient, detect sudden events like fires, and understand environmental risks in cities (e.g. Taylor et al, 2024). But the data is vast and can be very messy and unreliable. This project would use this big data to clean and help identify, e.g signals in this data that can help us answer important questions about our cities.

Taylor, J., et al (2024). The potential of urban trees to reduce heat-related mortality in London. Environmental research letters, 19(5), 054004.

Household air pollution data aggregation and Bayesian hierarchical modelling

Air pollution from the household burning of polluting fuels (such as solid wood, coal, etc) is estimated to lead to over 3 million premature deaths worldwide every year. However, these estimates, from for example the World Health Organisation, are limited because they only consider exposures to those burning the fuels, not the wider population – for example those who do not burn these fuels, but breathe the polluted air from them (e.g. Mohajeri et al, 2023). This topic will use a database of aggregated research studies, link them to the nearest measurement station, and develop a Bayesian Hierarchical model for household air pollution exposure. Alternative calculations will examine how much this exposure could be reduced if populations switched to cleaner fuels.

Mohajeri, N., et al. (2023). Urban–rural disparity in global estimation of PM2· 5 household air pollution and its attributable health burden. The Lancet Planetary Health, 7(8), e660-e672.

Urban greenery in Finnish Cities

Greenspace in cities have proven benefits for the environment, such as storing and sequestering carbon, potentially reducing air pollution, and lowering heat exposures. They also provide mental health benefits for residents. Studies into urban greenspace can often rely on satellite imagery or city surveys of public property. But what about what we actually see from the ground? What is the greenest city in Finland? What is the greyest? This project would assess greenery in Finnish cities using streetview imagery (e.g. Lu et al, 2023) helping to understand the availability and different types of greenspace.

Lu, Y., et al. (2023). Assessing urban greenery by harvesting street view data: A review. Urban Forestry & Urban Greening, 83, 127917.

Contact: Kostas Stefanidis, konstantinos.stefanidis [at] tuni.fi

Explanations for Different Stakeholders 

Unexpected (either existing or missing) and/or unjustified recommendations may frustrate the users and can be detrimental to their trust and loyalty to the recommender system (RS). Explainable RS share a piece of information from the system with users, so that they understand how/why the recommendation was computed. This information is commonly called an explanation and can either be 1) system-agnostic and data-specific information (e.g., user/item characteristics, previous actions), or 2) system-specific information (e.g., a set of rules in a rule-based decision system). At the same time, different stakeholders (e.g., consumers, producers, or system administrators) require different levels of detail in the explanations. For example, consumers can be satisfied with more general explanations, while system administrators require detailed model-specific ones. 

Tasks: Design an Explainable RS that is able to provide explanations to different stakeholders based on their roles. Explanations for different stakeholders can be one of but not limited to the following: 

Consumer: Model-agnostic - Explanations may include which previous actions of the consumer caused the recommendation to be received. (Counterfactual Explanations) [3,4]. Model-specific (user-based Collaborative Filtering) - Analysis of what other similar users to the consumer preferred and why the recommendation was generated [2]. 

Producer: Explanations for producers may include 1) which users prefer their products, 2) the product’s exposure to the consumers, 3) the differences between the product’s interested consumers and a specific user, and 4) statistics about their products (ratings received, positive-negative feedback, exposure).

System Administrator: Detailed explanations about the specific model variables that were the cause for the recommendation in question [1]. 

[1] M. Stratigi et al. Why-not questions & explanations for collaborative filtering. In Web Information Systems Engineering – WISE 2020. 

[2] Y. Zhang and X. Chen. Explainable Recommendation: A Survey and New Perspectives. Found. Trends Inf. Retr. 14(1): 1-101 (2020). 

[3] V. Kaffes et al. Model-agnostic counterfactual explanations of recommendations. In Proceedings of the 29th ACM conference on user modeling, adaptation and personalization, 2021. 

[4] J. Tan et al. Counterfactual explainable recommendation. In Proceedings of the 30th ACM International Conference on Information & Knowledge Management  – CIKM 2021. 

Identify and Explain Bias in Recommendation Systems

Recommendation systems (RS) can strongly influence the information we see online, e.g., on social media, and thus impact our beliefs, decisions, and actions. At the same time, these systems can create substantial business value for different shareholders. Given the growing potential impact of such systems on individuals, organizations, and society, questions of fairness have gained increased attention in recent years. 

Tasks: Employ a state-of-the-art recommendation system, identify the bias present in it, and provide a consumer-friendly explanation of its origins.

Identify bias: Biases in RS are typically divided into data bias and model bias. 

1. Data bias can come from data generation, collection, storage, etc. 
2. Model bias can arise from biases in model designing, training, and evaluating processes, such as the biased model architecture, improper use of specific optimization methods or estimators, and inappropriate benchmarks. 

Provide explanations: After identifying the bias, provide a user-friendly explanation about its origin. 

E. Pitoura, K. Stefanidis and G. Koutrika. Fairness in Rankings and Recommendations: An Overview. The VLDB Journal, 2022.

Yongfeng Zhang and Xu Chen. Explainable Recommendation: A Survey and New Perspectives. Found. Trends Inf. Retr. 14(1): 1-101 (2020)

Li, Yunqi, et al. Fairness in recommendation: Foundations, methods, and applications. ACM Transactions on Intelligent Systems and Technology 14.5 (2023).

Harini Suresh and John V Guttag. 2019. A framework for understanding unintended consequences of machine learning. arXiv preprint arXiv:1901.10002 2.

Alessandro Castelnovo, Riccardo Crupi, Greta Greco, Daniele Regoli, Ilaria Giuseppina Penco, and Andrea Claudio Cosentini. 2022. A clarification of the nuances in the fairness metrics landscape. Scientific Reports 12, 1.

Personalized summaries using LLMs and RAGs  

There have been several works focusing on summarizing knowledge graphs. However, the development of Large Language Models (LLMs) and Retrieval-Augmented Generation (RAG) frameworks, leaves an unexplored direction on how these novel AI techniques could be combined with existing works to increase the quality of the generated graph summary and potentially make it personalized. Further, the interrelation of textual summary with a summary graph is currently unexplored and could be further investigated.

Giannis Vassiliou, Nikolaos Papadakis, Haridimos Kondylakis: SummaryGPT: Leveraging ChatGPT for Summarizing Knowledge Graphs. ESWC (Satellite Events) 2023: 164-168

Sejla Cebiric et al. Summarizing semantic graphs: A Survey. VLDB J. 28(3): 295-327 (2019)

Contact: Tarmo Lipping, tarmo.lipping [at] tuni.fi 

Recording and analysis of physiological data in real-life environment 

1. Brain Computer Interfaces using consumer-oriented EEG devices

The thesis project will involve developing an environment of realt-time analysis of EEG signal. The signal will be obtained using a consumer-oriented device such as MUSE headband, for example. The analysis results will be used to control an external environment such as a computer game, robotic arm or music generation software. Also, interface between brain function and generative AI models can be considered. The specific setup for the experiments will be determined at the beginning of the project. The work is related to the EHEÄ (Well-being, vitality and smart services through experience production and technology), LuovAIn! (http://luovain.ai) and MindFlow (https://www.tuni.fi/en/research/mindflow-measuring-flow-state) projects of the DAO 
research group. 

2. Assessing inter-subject brainwave synchronization during joint tasks

There is a lot of evidence that if two or more subjects are engaged in joint activity, their brainwaves get synchronized. These kind of recordings are called hyperscanning. We have previously collected some hyperscanning data using the MUSE and g.tec Unicorn devices. The thesis project will involve carrying out a more systematic data collection in hyperscanning setting. The more specific tasks and research questions will be determined at the beginning of the thesis project. The project will also deal with testing and evaluating methods to detect synchronization in time series. 

3. Assessment of the physiological effects of cultural and recreational interventions

It is commonly accepted that cultural experiences have positive impact on mental well-being and recovery from stress. However, evidence to support this claim is still lacking. The thesis project will involve designing and implementing a body sensory system tailored for the assessment physiological effects of cultural experiences. In addition, the project may deal with using physiological data to augment cultural experiences by, e.g., using performers' physilogical data in controlling visual or audiotory environment during cultural performance. A specific experimental setup, determined at the beginning of the thesis project will be used. Thesis project will involve data recording and preliminari analysis.

Related references:

Beiramvand, M, Shahbakhti, M, Karttunen, N, Koivula, R, Turunen, J & Lipping, T 2024, 'Assessment of Mental Workload Using a Transformer Network and Two Prefrontal EEG Channels: An Unparameterized Approach', IEEE Transactions on Instrumentation and Measurement. https://doi.org/10.1109/TIM.2024.3395312

T. Lipping and M. Beiramvand, "Assessment of Mental Workload in Real-Life Setup using EEG Synchronization Measures,"  2024 IEEE International Workshop on Metrology for Industry 4.0 & IoT (MetroInd4.0 & IoT), Firenze, Italy, 2024, pp. 412-416, doi: 10.1109/MetroInd4.0IoT61288.2024.10584156.

Beiram Vand, M, Shahbakhti, M & Lipping, T 2023, Cross-Entropy-Based Assessment of Mental Workloads Using Two Prefrontal EEG Channels. In 2023 IEEE EMBS Special Topic Conference on Data Science and Engineering in Healthcare, Medicine and Biology. IEEE, pp. 49-50, International Conference on Data Science and Engineering in Healthcare, Medicine & Biology, St. Julians, Malta, 7/12/23. https://doi.org/10.1109/IEEECONF58974.2023.10404709

Hakim, U. et al. 2023 'Quantification of inter-brain coupling: A review of current methods used in haemodynamic and electrophysiological hyperscanning studies. NeuroImage, vol 280, 120354, https://doi.org/10.1016/j.neuroimage.2023.120354 

Applied machine learning, computational models and statistics in agriculture 

1. Remote sensing data in crop field characterization and zoning

Crop fields are monitored using remote sensing observations, often from satellites and drones. These result in multiband images with varying spatial and spectral resolutions. Using these images we can follow the progression of the growing season, identify differences within and between crop fields and 
identify patterns and anomalies. Find ways to divide fields into meaningful zones based on remote sensing images and other relevant data sources, to support field activities for precision field management and targeted information gathering.

2. Topological Data Analysis in Agriculture

Topological Data Analysis (TDA) is an emerging set of tools for analysing complex, multi-dimensional data. TDA has solid mathematical foundations and an existing set of mature software tools, yet it has not yet widely been used with agricultural data. Explore the capabilities of TDA with agricultural datasets such as remote sensing imagery, in extracting meaningful features with predictive power. Demonstrate how to use TDA as a standalone data analysis tool or how to combine it together with existing machine learning methods in solving practical problems.

3. Simulation models for crop growth

Dynamic crop models represent the growth process of a plant from seeding to maturity. These kind of models aim to encode scientific knowledge about the relevant natural processes, allowing us to explain and predict the growing behavior, given the environmental conditions during the season. Process models typically need to be calibrated to local conditions and their behavior analysed statistically, due to the many uncertainties involved. Analyze and calibrate the behavior and results of crop growth models using statistical methods such as Monte carlo simulations and sensitivity analysis, to support tasks such as biomass estimation and irrigation planning.