Publications by students
Analysing political biases in danish newspapers using sentiment analysis
By Kenneth C Enevoldsen, Lasse Hansen
Traditionally, the evaluation of political biases in Danish newspapers has been carried out through highly subjective methods. The conventional approach has been surveys asking samples of the population to place various newspapers on the political spectrum, coupled with analysing voting habits of the newspapers’ readers (Hjarvard, 2007). This paper seeks to examine whether it is possible to use sentiment analysis to objectively assess political biases in Danish newspapers. By using the sentiment dictionary AFINN (Nielsen et al., 2011), the mean sentiment scores for 360 articles was calculated. The articles were published in the Danish newspapers Berlingske and Information and were all regarding the political parties Alternativet and Liberal Alliance. A significant interaction effect between the parties and newspapers was discovered. This effect was mainly driven by Information’s coverage of the two parties. Moreover, Berlingske was found to publish a disproportionately greater number of articles concerning Liberal Alliance than Alternativet. Based on these findings, an integration of sentiment analysis into the evaluation of biases in news outlets is proposed. Furthermore, future studies are suggested to construct datasets for evaluation of AFINN on news and to utilize web-mining methods to gather greater amounts of data in order to analyse more parties and newspapers.
Emotional Metacognition: Stimulus Valence Modulates Cardiac Arousal and Metamemory
By Sebastian Scott Engen
Cognition and Emotion
Emotion alters how we feel, see, and experience the world. In the domain of memory, the emotional valence and arousal of memorized stimuli can modulate both the acuity and content of episodic recall. However, no experiment has investigated whether arousal and valence also influence metacognition for memory (i.e., the process of self-monitoring memories). In a pre-registered study, we applied a novel psychophysiological design together with computational models of metacognition to assess the influence of stimulus valence and arousal on the sensitivity, bias, and efficiency of metamemory. To estimate the role of physiological arousal in mediating these effects, we recorded cardiac measures through pulse oximetry. We found that negative valence globally and substantially decreased both memory performance and subjective confidence, in particular for low-arousal words. Simultaneously, we found that emotional valence modulated both heart rate and heart-rate variability (HRV), indicating a robust effect of negative valence on physiological arousal during recognition memory. Exploratory trial-level analyses further revealed that subjective confidence was encoded in instantaneous heart-rate fluctuations, and that this relationship was modulated by emotional valence. Our results demonstrate that both recognition memory and metacognition are influenced by the emotional contents of encoded items and that this correlation is in part related to cardiac activity.
Sentida: A New Tool for Sentiment Analysis in Danish
By Gustav Aarup Lauridsen, Jacob Aarup Dalsgaard, and Lars Kjartan Bacher Svendsen
In the midst of the Era of Big Data, tools for analysing and processing unstructured data are needed more than ever. Being among these, sentiment analysis has experienced both a substantial proliferation in popularity and major developmental progress. However, the development of sentiment analysis tools in Danish has not experienced the same rapid development as e.g. English tools. Few Danish tools exist, and often the ones available are either ineffective or outdated. Moreover, authoritative validation tests in low-resource languages, are missing, which is why little can be deduced about the competence of current Danish models. We present SENTIDA, a simple and effective model for general sentiment analysis in Danish, and compare its competence to the current benchmark within the field of Danish sentiment analysis, AFINN. Combining a lexical approach with several incorporated functions, we construct SENTIDA and categorise it as a domain-independent sentiment analysis tool focusing on polarity strength. Subsequently, we run different validation tests, including a binary classification test of Trustpilot reviews and a correlation test based on manually rated texts from different domains. The results show that SENTIDA excels across all tests, predicting reviews with an accuracy above 80% in all trials and providing significant correlations with manually annotated texts.
tomsup: An implementation of computational Theory of Mind in Python
By Kenneth C Enevoldsen, Peter Thestrup Waade
Computational implementations of Theory of Mind (ToM), the ability to attribute mental states to others, has been used to investigate a variety of issues. This includes the effect of framing effects on, or inter-species differences in, ability to do ToM (Devaine et al., 2014a, 2017), ToM in autists (d’Arc et al., 2018), or providing an explanation for the apparent limits on human ability to do ToM recursively (Devaine et al., 2014b). It has been implemented in the VBA package for Matlab (Daunizeau et al., 2014), but not in any free and open-source software. Therefore this thesis presents the Theory of Mind simulation using Python (tomsup) package. The tomsup package provides accessible tools for running agent-based models in a game theory context, and allows the implementation of a computational model of ToM, either in agent-based models or in interaction with a human player. The implementation of the ToM model was originally proposed by Yoshida et al. (2008), and was developed by drawing on the Free Energy Principle (Friston, 2010) to its current form as it is in Devaine et al. (2017), where it is generalized to any 2-player game which can be operationalized as a 2-by-2 payoff matrix. Importantly, the ToM implementation introduces a sophistication level k, which determines how many recursive simulations of its opponent it can perform, hereby assuming bounded rationality (Kahneman, 2003). An agent using the ToM model, denoted as k-ToM, uses a variational Bayes Laplace approximation (Daunizeau, 2017b) on a turn-by-turn basis to infer its opponent’s model parameters and sophistication level, based on which it predicts the opponent’s choice and acts accordingly. An agent-based model simulation using the competitive matching pennies game was done to perform a preliminary investigation of the behaviour of the k-ToM model. Most importantly, it was found that k-ToM’s prior beliefs about its opponent have a notable effect on its performance, even over many trials, warranting further research into how its priors should be formed. Various ways are suggested in which the tomsup package and the k-ToM model could be applied and developed further, as well as a discussion on how to make it broadly available, so as to scaffold future research using computational ToM models.
Confucian Free Energy: The Predictive Mind in Ancient China
By Peter Thestrup Waade
Roger Ames presents an interpretation of the classic Confucian philosophy where the world is seen as an always ongoing, radically interrelational, holistic process best described in terms of focus and field. Without any permanent ground truth, its inhabitants must be action-oriented way-makers: irreducibly social optimizers, and contextual co-creators, of their shared cosmos. Meanwhile, a pragmatic turn is happening in the cognitive sciences, instantiated in a new and ambitious theoretical framework: Karl Friston’s free energy principle, which aims to be a unifying paradigm for the sciences of brain, life and mind. Based on first principles from statistical physics, it describes humans - and self-organizing systems in general - as continually modelling and predicting the world. The full spectrum of human cognition and behavior is then generated by a single mechanism: the minimization of the difference between predicted and actual sensory inputs. Although still controversial, the free energy principle is gaining traction within robotics, biology, neuroscience, psychology, social sciences and philosophy of mind. I here compare these two worldviews from so vastly different backgrounds, and find, perhaps surprisingly, that they in many respects are immediately compatible. Social co-creativity across spatio-temporal scales, contextualization, action-orientation and the Confucian virtues naturally translate to or integrate with free energy principle terms. Even when in seeming disagreement, for example on the question of the appearance/reality dichotomy, or that of reductive materialism which often separates the sciences and humanities in general, the two frameworks seem able to positively inform or nuance each other. This multidisciplinary comparison is interesting because it can potentially inform or guide the two theories, and address recent critiques of Ames, and as a proof of concept that bridges can be built across times, cultures and academic traditions
SEIZ Matters: Modelling the spread of concepts on Twitter
By Jonathan H. Rystrøm
How do different concepts spread on social media? This question is becoming increasingly important as much of our time, discussion, and news consumption move online. This paper investigates the use of two models from epidemiology, namely the classical SI-model and the sociology-inspired SEIZ-model, to model and understand this phenomenon. I study the spread of two concepts during the 2019 Danish national election, klimatosse (climate fool) and Paludan on Twitter, both of which played key roles in the election season and had epidemic qualities in their usage throughout social media. I find that although both models can provide decent fits of the data, the SEIZ-model outperforms the SI-model by a wide margin. Furthermore, the parameters can be interpreted to provide a deeper understanding of the two phenomena and how they spread.