Het woord ‘bias’ komt naar voren in zowel maatschappelijk als wetenschappelijk debat over de inzet van artificiële intelligentie (ai). Het verwijst doorgaans naar een vooroordeel dat iets of iemand vaak onbedoeld heeft. Wanneer dit vooroordeel leidt tot een afwijking in besluitvorming vergeleken met een situatie wanneer dit vooroordeel er niet zou zijn, dan is een bias doorgaans onwenselijk.
LINK
With the proliferation of misinformation on the web, automatic misinformation detection methods are becoming an increasingly important subject of study. Large language models have produced the best results among content-based methods, which rely on the text of the article rather than the metadata or network features. However, finetuning such a model requires significant training data, which has led to the automatic creation of large-scale misinformation detection datasets. In these datasets, articles are not labelled directly. Rather, each news site is labelled for reliability by an established fact-checking organisation and every article is subsequently assigned the corresponding label based on the reliability score of the news source in question. A recent paper has explored the biases present in one such dataset, NELA-GT-2018, and shown that the models are at least partly learning the stylistic and other features of different news sources rather than the features of unreliable news. We confirm a part of their findings. Apart from studying the characteristics and potential biases of the datasets, we also find it important to examine in what way the model architecture influences the results. We therefore explore which text features or combinations of features are learned by models based on contextual word embeddings as opposed to basic bag-of-words models. To elucidate this, we perform extensive error analysis aided by the SHAP post-hoc explanation technique on a debiased portion of the dataset. We validate the explanation technique on our inherently interpretable baseline model.
DOCUMENT
Valuation judgement bias has been a research topic for several years due to its proclaimed effect on valuation accuracy. However, little is known on the emphasis of literature on judgement bias, with regard to, for instance, research methodologies, research context and robustness of research evidence. A synthesis of available research will establish consistency in the current knowledge base on valuer judgement, identify future research opportunities and support decision-making policy by educational and regulatory stakeholders how to cope with judgement bias. This article therefore, provides a systematic review of empirical research on real estate valuer judgement over the last 30 years. Based on a number of inclusion and exclusion criteria, we have systematically analysed 32 relevant papers on valuation judgement bias. Although we find some consistency in evidence, we also find the underlying research to be biased; the methodology adopted is dominated by a quantitative approach; research context is skewed by timing and origination; and research evidence seems fragmented and needs replication. In order to obtain a deeper understanding of valuation judgement processes and thus extend the current knowledge base, we advocate more use of qualitative research methods and scholars to adopt an interpretative paradigm when studying judgement behaviour.
DOCUMENT
Albeit the widespread application of recommender systems (RecSys) in our daily lives, rather limited research has been done on quantifying unfairness and biases present in such systems. Prior work largely focuses on determining whether a RecSys is discriminating or not but does not compute the amount of bias present in these systems. Biased recommendations may lead to decisions that can potentially have adverse effects on individuals, sensitive user groups, and society. Hence, it is important to quantify these biases for fair and safe commercial applications of these systems. This paper focuses on quantifying popularity bias that stems directly from the output of RecSys models, leading to over recommendation of popular items that are likely to be misaligned with user preferences. Four metrics to quantify popularity bias in RescSys over time in dynamic setting across different sensitive user groups have been proposed. These metrics have been demonstrated for four collaborative filteri ng based RecSys algorithms trained on two commonly used benchmark datasets in the literature. Results obtained show that the metrics proposed provide a comprehensive understanding of growing disparities in treatment between sensitive groups over time when used conjointly.
DOCUMENT
With the proliferation of misinformation on the web, automatic methods for detecting misinformation are becoming an increasingly important subject of study. If automatic misinformation detection is applied in a real-world setting, it is necessary to validate the methods being used. Large language models (LLMs) have produced the best results among text-based methods. However, fine-tuning such a model requires a significant amount of training data, which has led to the automatic creation of large-scale misinformation detection datasets. In this paper, we explore the biases present in one such dataset for misinformation detection in English, NELA-GT-2019. We find that models are at least partly learning the stylistic and other features of different news sources rather than the features of unreliable news. Furthermore, we use SHAP to interpret the outputs of a fine-tuned LLM and validate the explanation method using our inherently interpretable baseline. We critically analyze the suitability of SHAP for text applications by comparing the outputs of SHAP to the most important features from our logistic regression models.
DOCUMENT
This applied research project aims to generate a better understanding of the effects of heatwaves on vulnerable population groups in the municipality of The Hague, and suggests ways in which the municipality can help such groups to cope with these heatwaves. The research was performed as a cooperation between The Hague University of Applied Sciences (THUAS), the International Institute of Social Studies (ISS, Erasmus University Rotterdam) and the International Centre for Frugal Innovation (ICFI, Leiden-Delft-Erasmus Universities). Heatwaves constitute an important yet often overlooked part of climate change and their impacts qualify as disasters. According to the World Disasters Report 2020, the three heatwaves affecting Belgium, France, Germany, Italy, the Netherlands, Spain, Switzerland and the UK in the summer of 2019 caused 3,453 deaths.1 2020 was a new record year for the Netherlands because it was the first time that a heatwave included five days in a row during which the temperature reached 35 degrees or more. In addition, 40 degrees was measured for the first time, and periods of tropical days and nights are generally getting longer. Most importantly, this trend is accelerating faster than the climate change models are predicting.2 In addition, the COVID-19 pandemic is compounding the effect of heatwaves, as vulnerable individuals may be reluctant to seek cool spaces out of fear of infection. Already in 2006, the Netherlands ranked near the top of the global disaster index due to the number of excess deaths that could be attributed to the heatwave. In the same year, the EU published the first climate strategy in which heat is recognised as a priority. In 2008, the Netherlands developed its first national heat plan.4 The municipality of The Hague has a municipal climate adaptation strategy and has developed a draft local heat plan in the summer of 2021, which was published in February 2022 . This research was not meant to be and was not set up as an evaluation of the current heat plan, which has not yet been activated. At the level of municipalities and cities, the concept of urban resilience is key. It refers to “the capacity of individuals, communities, institutions, businesses, and systems within a city to survive, adapt, and grow no matter what kinds of chronic stresses and acute shocks they experience”. Heatwaves clearly constitute acute shocks which are rapidly developing into chronic stresses. In turn, heatwaves also exacerbate the chronic stresses that are already there, i.e. existing chronic stresses also lead to greater impact of a heatwave. In other words, there are negative interaction effects. Addressing these effects requires overcoming the silo approach to urban governance, in which different municipal departments as well as other stakeholders (such as the Red Cross, housing corporations, tenants’ associations, care organisations, entrepreneurs etc.) each address different parts of the problem, rather than doing so in an integrated and inclusive manner. The dataset for this study is archived in DANS Easy: https://doi.org/10.17026/dans-xeb-h8uk
MULTIFILE
Learning Analytics en bias – Learning analytics richt zich op het meten en analyseren van studentgegevens om onderwijs te verbeteren. Bakker onderscheidt hierin verschillende niveaus, zoals student analytics en institutional analytics, en focust op inclusion analytics, waarin gekeken wordt naar kansengelijkheid. Bias – systematische vooroordelen in data – kan vooroordelen in algoritmen versterken en zo kansenongelijkheid veroorzaken. De onderzoeksmethode maakt gebruik van het 4/5-criterium, waarbij fairness in uitkomsten gemeten wordt door te kijken of de kansen voor de beschermde groep minstens 80% zijn van die van de bevoorrechte groep.Onderzoeksaanpak – Bakker gebruikt machine learning om retentie na het eerste studiejaar te voorspellen en onderzoekt vervolgens verschillen tussen groepen studenten, zoals mbo-en vwo-studenten. Hij volgt drie stappen: (1) Data voorbereiden en modellen bouwen: Data worden opgesplitst en opgeschoond om accurate voorspelmodellen te maken. (2) Variabelen analyseren: Invloed van kenmerken op uitkomsten wordt beoordeeld voor verschillende groepen. (3) Fairness berekenen: Het 4/5-criterium wordt toegepast op metrics zoals accuraatheid en statistische gelijkheid om bias en ongelijkheden te identificeren. Resultaten, aanbevelingen en vervolgonderzoek – Uit het onderzoek blijkt dat kansengelijkheid bij veel opleidingen ontbreekt, met name voor mannen en mbo-studenten, die een hogere kans op uitval hebben. Bakker adviseert sensitieve kenmerken zoals migratieachtergrond mee te nemen in analyses op basis van informed consent. Daarnaast pleit hij voor meer flexibiliteit in het beleid, geïnspireerd door maatregelen tijdens de coronacrisis, die een positief effect hadden op studiesucces.Toekomstvisie – Bakker benadrukt dat niet elke ongelijkheid het gevolg is van discriminatie en roept op tot data-informed interventies om sociale rechtvaardigheid in het onderwijs te bevorderen. Zijn methode wordt open access beschikbaar gesteld, zodat ook andere instellingen deze kunnen toepassen en kansengelijkheid systematisch en bewust kunnen onderzoeken.
DOCUMENT
* Poster 1 - Weet wat er leeftDoor data van computer vision, eDNA, geuren en omgevingssensoren te voeden aan een AI systeem, kan de verspreiding van soorten in kassen snel worden bepaald. Met TKI PPS Weet wat er leeft wordt een dergelijk systeem ontwikkeld. De HAS is betrokken bij de werkgroep Vision* Poster 2 - Weet wat er leeft: Hoe meer hoe beter?Het aantal soorten op een vangplaat kan van invloed zijn op de performance van een Vision model. Hier testten we in hoeverre het mogelijk is om automatisch verschillende plaatsoorten op plakplaten te herkennen* Poster 3 - Weet wat er leeft: Bestrijders in the pictureHet aantal soorten op een foto kan van invloed zijn op de performance van een Vision model. Hier testten we in hoeverre het mogelijk is om automatisch verschillende biologische bestrijders te herkennen* Poster 4 - Weet wat er leeft: Een licht schijnen op telefoonsOm automatische beeldherkenning goed te kunnen gebruiken moet een model goed zijn afgestemd op het uiteindelijke gebruik. Hier testen we het effect van verschillende telefoons en lichtcondities op de performance* Poster 5 - Weet wat er leeft: Combineren van Vision en eDNAZowel computer vision als eDNA technieken worden steeds meer gebruikt om soorten te monitoren. Ze kennen beiden hun voordelen en beperkingen. In deze studie onderzoeken we hoe ze elkaar kunnen aanvullen* Poster 6 - Weet wat er leeft: Optimaliseren van modelEen model is zo goed als de data waar het op getraind is. Door grote verschillen in aantallen insecten en locaties, kan bias plaatsvinden. In dit project is trainingsdata van een Custom Vision model aangepast en is bepaald of het model daar beter van wordt* Poster 7 - Weet wat er leeft: Combinatie van techniekenPlaatmonitoring in de kas is arbeids- en kennisintensief. In dit onderzoek is onderzocht op welke manier automatische beeldherkenning en eDNA elkaar aanvullen in zowel een gecontroleerde al sin een praktijkomgeving als alternatieve monitoringstechnieken* Poster 8 - Weet wat er leeft: verschillende platenHet Custom Vision model (CV2) dat gebruikt wordt binnen het project is getraind op gele droge platen. Hier onderzoeken we of dat model ook geschikt is voor andere platen, of dat het nodig is nieuwe modellen te trainen* Poster 9 - Weet wat er leeft: plakplaten door de tijdIn de praktijk blijven plakplaten om te monitoren meerdere weken in de kas hangen. Dit kan invloed hebben op de performance van een model, omdat insecten verouderen en de dichtheid toeneemt
DOCUMENT
