In May 2019, the Dutch Council of State rejected the national approach for reducing nitrogen emissions in Dutch nature. Farmers were targeted by the policy change: all licenses for agricultural expansion were revoked, affecting the financial livelihoods of farmers. Farmers did not take this well and, using social media platforms, started organising large-scale demonstrations. Twitter flushed with posts about the demonstrations, and pictures and videos of the event went viral. Demonstrations result from social unrest, and social unrest starts with public dissatisfaction. The Nitrogen Crisis showed it is in the interest of decision-makers to monitor what negative feelings the public holds towards policies and act on these, before these feelings grow into social unrest. Twitter is a social media platform many users come to for expressing their opinions. Because of this, this study looks at what insights can be derived from Twitter about events, like demonstrations, that took place during the Nitrogen Crisis. For this, this study applies two Natural Language Processing methods: sentiment analysis and topic modelling (LDA). These methods are combined in order to create more insightful and interpretable results than the methods individually could provide. Two interviews are held with an expert on the Nitrogen Crisis to provide context on the crisis and to identify major events that received a lot of media attention. The events are plotted with- and compared to the results of sentiment analysis and topic modelling. In doing so, the following research question is answered:
How can sentiment analysis and topic modelling be applied to Twitter data to provide insights for decision-makers retrospectively about major events during the Dutch Nitrogen Crisis?
For sentiment analysis, two Dutch sentiment analysis tools are implemented and compared to the sentiment scores of 100 tweets by three annotators to select the best performing one. For topic modelling, a grid search is performed to choose the combination of timeframe and number of topics that result in the set of topic models that have the highest mean topic coherence. Also, a method is proposed for using topic models to represent changes in the topics discussed on Twitter over time. This is used not only to compare subsequent topic models per sentiment that are one week apart, but also topic models that are 4 weeks apart.
This research develops a fully functioning pipeline for collecting and processing tweets, applying sentiment analysis and topic modelling and plotting the outcomes. This pipeline has been validated at various points, leading to a scientifically viable methodology.
Unexpectedly, it is not sentiment analysis or topic modelling results that have the most obvious connection with the events identified: it is an increase in tweets during events. Therefore, while lacking better tools, decision-makers are recommended to monitor pre-determined topics on Twitter and implement a way to be notified when a significant change in volume of tweets takes place. The combination of sentiment analysis and topic modelling as implemented in this research is either not advanced enough to provide useful information to decision-makers, or sentiment analysis and topic modelling simply cannot provide insightful results on the Dutch Nitrogen Crisis. However, because there is an extensive amount of research applying these methods to social media data around various political events with valuable results, it is recommended to perform more experiments with this approach, and the quality of each research step needs to be further improved in order to draw final conclusions on the usefulness of the combination of sentiment analysis and topic modelling for decision-makers during policy crises. Various improvements for each step in this research are suggested to gain more precise, insightful and interpretable results. In summary, this study and the pipeline it proposes can serve as a solid basis for further development into a process that provides ready-to-use information to decision-makers.

Stance detection is a Natural Language Processing task that can detect if the input text is in favour, against or neutral towards a target. Research on stance detection has been growing and evolving over the last decade. In this paper, the current approaches for stance detection are discussed with a focus on the deep learning approaches. The organized competitions are discussed, and the most used traditional and deep learning approaches are shown. The challenges that arise with deep learning approaches are looked into further. Finally, an experiment was performed to examine and demonstrate the effects of a small data set on various stance detection models, this was done using the SVM, CNN, and BERT models on the SemEval 2016 data set. This experiment shows that a smaller data set has a greater negative impact on the CNN model than theSVM model. BERT is affected the least and outperforms the other models significantly.

The increasing amount of untrusted content on the internet is a worrisome trend. The headline of an article can be adjusted to influence a potential readers attention and click-through rate. This clickbait or sensationalism can mislead the reader as the headline does not accurately represent the information in the corresponding article body. This headline incongruity problem has received some recent attention from the research community with proposed datasets and approaches. However, there still lacks an overarching paper that tries to answer the current state of tackling the problem. This paper aims to fill this gap, look at recent proposed datasets and approaches, and compare them. These results will be discussed, reflected on, and formulated to answer the research question in conclusion. The main conclusions from this research are that the most suited dataset for comparison purposes is the Real-News based dataset and the Graph Neural Network by Yoon et al. performed the best.

A side-channel attack is performed by analyzing unwanted physical leakage to achieve a more effective attack on the cryptographic key. An attacker performs a profiled attack when he has a physical and identical copy of the target device, meaning the attacker is in full control of the target device. Therefore, these profiled attacks are known as the most powerful attacks in the side-channel analysis. This physical leakage is analyzed by machine learning and, in the last years, mostly deep learning, which both are used as a profiling tool to perform a side-channel attack. The best known deep learning technique for side-channel analysis at this moment is the convolutional neural network (CNN). However, this thesis investigates a well-known deep learning model that is never used before in side-channel analysis. The deep learning models RNN, LSTM, and GRU are tested and evaluated to look for the best hyperparameters. We show the influence of different models, amount of layers, dropout, activation function, units, recurrent dropout, and batch sizes in the experiments. We also show that using different sequence length gives a speedup in training. To reduce the sequence length, we use a linear regression technique. After that, we show that sequential data models are a suitable alternative for side-channel analysis; however, their results do not surpass the CNNs. After this, we experiment with an autoencoder as a preprocessing algorithm to "clean" noisy traces. We show that the LSTM autoencoder easily removes a hiding countermeasure with noise. However, a hiding countermeasure with delay is more challenging for the LSTM autoencoder. Combining both countermeasures seems impossible for the LSTM autoencoder. The performance we see when cleaning the traces also affects the guessing entropy. Lastly, we use an embedding layer as the first layer for MLP, CNN, and a sequential data model in the side-channel analysis. We experiment with different output dimensions and conclude that an embedding layer is a valid alternative to change the data dimension when using an MLP or a sequential data model.

To detect malicious activities in a network, intrusion detection systems are used. Even though these solutions are widely deployed for this purpose they have one serious shortcoming which is the huge amount of false alarms that they are generating. Different measures are taken to tackle this problem such as manually changing the settings of the intrusion detection systems. However, this is an infeasible approach for organisations since a network is changing regularly and specialists that have good knowledge of both the environment and the solution are required. The existing unsupervised approaches cannot be implemented as fully automated solutions because of the need to tune hyper-parameters. Additionally,the implementation of the existing solutions is complicated since often multiple models and the constant updating thereof is required which is a computationally intensive process considering the selected algorithms. In this work, the possibilities to reduce the false alarms in an automated manner are investigated. This is done by applying unsupervised anomaly detection techniques on the resulting alert data to distinguish regular alarms from high priority ones. Real alert data is collected from a network of a large organisation and an additional synthetically generated data set is used to evaluate the proposed approach. Four unsupervised anomaly detection algorithms are chosen to model the regular alerts. These are Local Outlier Factor (LOF), Isolation Forest (IF), Histogram-based Outlier Score(HBOS) and Cluster-based Local Outlier Factor (CBLOF). We show that this approach can greatly reduce the false alarms in real environments. By adding noise to the data we evaluate the performance of the models and propose a method that can be used to determine when the model needs to be retrained. This is done by deriving a metric that is used to trigger the system to automatically retrain on the most recent historic data. This is necessary in order to make the system automated and adaptable to changes in the network.

In recent time, there has been a rise in the number of coworking and activity-based working spaces. This increase has brought a high return to the real-estate developers but at the same time has compromised with occupant satisfaction. It brings an additional challenge for the architects and designer to look into the user-centred design for workspaces and offices. The current design process relies heavily on the experience. There is a need to evaluate the design from a user’s perspective.
It is known that user’s in an office environment are impacted by various Indoor environmental quality factors. The most important of those factors is the noise disturbance which leads to decreased concentration, reduced satisfaction, and sick building syndrome in some cases. The noise disturbance is caused by speech interactions, appliances and services. These disturbances are relatively unknown to the designer in the early design phase and depend on user behaviour. This graduation project challenges the current design process by proving a digital intervention in the form of workflow. The workflow aims to help architects and designers in the early design process to evaluate spatial layout with respect to disturbances in the office.
This research combines the knowledge of design informatics, psychology and building physics to come up with a simulation workflow. It uses the knowledge of the psychology of workspaces to understand the user behaviour towards the noise disturbance. This knowledge is put to use through an agent-based model simulating the user movement and interactions through the office layout. It is made possible by combining existing methodologies of acoustic simulation, trajectory evaluation with the newly developed agent-based model that is simulating user behaviour. The results are then visualized on the spatial layout to get further insights into the design by testing with two case studies. As a result of prototyping, a simulation workflow is proposed which can be used to evaluate workspace layout in terms of user behaviour towards the noise.

In the past years, society has become increasingly more reliant on the Internet. Consequently, the security of the Internet became of critical importance. This thesis focusses on the security of one of the Internet's main protocols. This protocol, called the Border Gateway Protocol (BGP), is used to exchange information that allows Internet traffic to reach its intended destination. BGP is vulnerable to misconfigurations and attacks that can cause a range of problems. This thesis focusses on one of them: BGP origin hijacks. In this thesis, a year of possible origin hijacks is analysed. These possible origin hijacks were detected by BGPStream between 20 May 2018 and 31 May 2019. Analysing these hijacks gives insight into the causes and characteristics of origin hijacks. This can help to find the most pressing issues and may provide guidance in securing BGP. Various data sources are used to collect and compute features that give more information on each hijack. These features are used to find relations between hijacks and to label them using labels that indicate a cause or a certain aspect of the hijack. These relations and labels are used to analyse groups of similar hijacks. This approach is very effective. Using the context of a group of hijacks gives much more insight than looking at hijacks individually. It shows that many of the possible hijacks are likely not a hijack at all and that hijacks that look like origin hijacks are often the result of another type of attack called a path hijack. In addition, this thesis provides a way to detect several types of misconfigurations and points out weaknesses in the detection system used by BGPStream. It also gives an overview of the characteristics of hijacks and how often specific behaviour occurs.

A side-channel attack (SCA) recovers secret data from a device by exploiting unintended physical leakages such as power consumption. In a profiled SCA, we assume an adversary has control over a target and copy device. Using the copy device the adversary learns a profile of the device. With the profile, the adversary exploits the measurements from a target device and recovers the secret key. As SCAs have shown to be a realistic attack vector, countermeasures have been invented to harden these kinds of attacks. In the last few years, deep learning has been applied in a wide variety of domains. For example, convolutional neural networks have shown to be effective for object recognition in images and recurrent neural networks for text generation. In the side-channel analysis domain, deep learning has shown to be successful. Up until recently, no deep learning layer existed that was specifically designed for SCAs. In this work, we analyze this layer, called the spread layer, and demonstrate the flaws of this layer. We improve the flaws and show the spread layer does not enhance the performance of SCAs. Additionally, we show there is no need to develop a deep learning layer specifically for SCAs on unprotected implementations. For implementations where countermeasures are present, literature demonstrated that convolutional neural networks are the most successful. However, for both the masking and random delay countermeasure, little is known about the influence of the kernel size and depth of the network. In this work, we illustrate that increasing the kernel size and depth of the network both increase the attack efficiency for the random delay countermeasure. For the masking countermeasure, we demonstrate that higher kernel sizes and shallow networks perform the best. Additionally, in this work, we consider a portability setting where the probe position has been changed in between the measurements of the profiling and attack measurements. Here, we show that the probe position causes a typical deep learning SCA to be ineffective. We introduce a normalization method such that the attack becomes effective, and show this method enables the attack to perform as expected.