VACANCY Post-doc ENSURE, 2 years –1,0 FTE

This is a post-doctoral position on the EnSuRe (ExplaiNing SeqUences in Recommendations) project. The research agenda involves:
• Gaining an understanding of people's concerns regarding personalisation for sequences of recommended items.
• Gaining an understanding of people's views on the kinds of explanations that alleviate their concerns and help them to make good decisions.
• Producing guidelines for algorithms for constructing explainable recommender sequences.
• Developing algorithms for explaining sequences containing both novelty and trade-offs effectively, and while considering privacy concerns. This includes investigating the role of context and personal characteristics.
• Facilitating a dialogue between policy makers, researchers, and the general public regarding the findings above.

You hold a PhD in computer science or a related discipline. You have a track record of scientific excellence in the field(s) of recommender systems, user-modelling, multi-objective optimisation, and/or human-computer interaction. You must demonstrate either an ability to design algorithms for sequences of items, or extensive experience designing interactions with recommender systems. You will be expected to lead or strongly contribute to academic publications, contribute to grant proposals, and interact with stakeholders outside academia (e.g., end users, business and public policy). Strong verbal and written communication skills are therefore also required.

Details here:

Apply here:

VACANCY Post-doc LDE-BOLDCities, 1 year – 0,8-1,0 FTE

Research theme: The value of data analytics for urban decision-making.

Research question: How can particular methods and tooling for urban big data analytics be used to support urban decision-makers?

Background: In cities, there is a growing volume of data being continuously produced. In search for better-informed decisions by urban stakeholders, there is a natural tendency to collect, process and analyse these data by transforming them into information and actionable insights. However, this process is technology and knowledge intensive – requiring not readily available technology, knowledge and competences – and time consuming and therefore often expensive, especially when considered in relation to the value created. The key question in practice that inspires this research is therefore: Will the value of the insights generated outweigh the investments needed to develop those insights?

Research aim: This research aims to develop a framework to support urban decision-makers to determine the appropriate level or form of data analytics capability on urban decisions. This framework includes the process, the method, the tooling and the skills required from the users to analyse big data and transform into insights and information needed for generating actionable insight

  • Finished PhD in a relevant area of technology
  • Proven experience in relevant fields (urban big data, data analytics, modeling)
  • Willingness to work in a multi-disciplinary environment
  • Excellent communication skills

See - postdoc LDE BOLD Cities.

A few topics for which Postdoc positions might be open soon:

  • Learning Analytics. (With LDE-CEL)
  • Explanations in User Interfaces. (With Nava Tintarev)

For these themes, there might be openings (soon), so good candidates interested can always contact prof. Geert-Jan Houben.

Here are a few examples of projects and topics, we would like to have postdocs on:

  • Complex Search and Learning
    The learning process is more involved than what is currently acknowledged in MOOC platforms. One important aspect of online learning is search (retrieving information) and sensemaking (making sense of the information). Search & sensemaking is an intricate part of the learning process, and for many learners synonymous with accessing and ingesting information through Web search engines.
    In order to support complex search scenarios, we need to design search systems whose effectiveness is optimized over whole search episodes instead of individual search queries. Although existing research in interactive IR has made some headway towards a better understanding of complex search tasks, empirical research approaches are hampered by the fact that they mostly rely on simulated search tasks and a very limited number of lab study participants. In this project, we will design and deploy a Web search system within a number of MOOCs, to collect information on naturally occurring complex information needs and how learners go about solving them. The MOOC setting provides us with explicit information on learners' knowledge and skills, allowing us to explore how search success and search behaviour relate to learning behaviour and knowledge gains - for the first time in a natural and large-scale setting (instead of a small-scale lab setting).
    Contact: Claudia Hauff (
  • Gamifying the MOOC experience through deep learning and crowdsourcing
    MOOC learners are often ill-equipped to excel in this new type of online learning environment for a number of reasons, most importantly a lack of self-regulatory learning skills. Learners receive little information about their learning performance beyond the scores they achieve on assignments. Advances in recent years in the field of natural language processing, in particular its use of deep learning, offer us a way forward through the use of neural language models and their ability to generate language. In this project, we will design and deploy a system that - based on relevant text material such as textbooks and scientific papers - automatically generates an unlimited set of assessment questions on the fly through which learners can self-assess their comprehension of the material, not just at fixed assessment periods but continuously. We will explore crowdsourcing approaches to semi-automatically determine the validity of the generated questions. The system will also include gamification elements (leaderboard, streaks and so on) to entice MOOC learners to use it.
    Contact: Claudia Hauff (
  • Explanations to support media-literacy in learners
    More and more educators are adopting problem-based learning. One common motivation is to motivate students to take ownership for their own learning, and another growing classroom sizes. Learners in these settings, possibly with the support of a mentor, inform their own learning and often independently seek additional resources and references. This is largely a positive development, and unlike previous generations of learners, these learners can take advantage of the rich availability of online resources to support their opinions and arguments when completing course work.
    However, there has also been a surge in the number of articles and statements online that are misleading or false, sometimes with the sheer purpose of profit. These sorts of sensationalized ``fake news’’ are often widely shared, and have received a lot of attention recently. Stopping the proliferation of fake news is not just the responsibility of the platforms used to spread it. Those who consume news also need to find ways of determining if what they are reading is true.
    Toward this end, this project addresses ways of helping learners assess the veracity of online resources. It will use explanation facilities to justify why a source may be reliable, or not. For example, it may be shared on a reliable domain (more reliable) or it may supply strong statements with no quotes (less reliable). The project will help develop a set of signals that can be automatically detected and used as explanations to learners. It will also evaluate whether learners can use these explanations to accurately assess whether online articles are truthful or not.
    Contact: Nava Tintarev (via
  • What am I not seeing? Visualizing Educational blind-spots
    Recommender systems are a familiar part of our everyday online lives, suggesting items to try, and helping us to deal with information overload online. They are also used in education, to help students navigate a plethora of potential learning resources. Previous research has found that showing users their progress using open learner models can help them decide on what to study next. One area that has been under-explored in this regard, is helping learners navigate their ``unknown unknowns’’. These are parts of the learning space the learner is unfamiliar with. These are not just areas where the learner has not made any progress, but also areas that they do not even know exist yet, and are unlikely to explore without external influences.
    Consequently, this project will explore the use of interactive visualisations to help users understand their learner profiles. It will explore how these techniques might be used to highlight learning blind-spots, i.e. regions of the recommendation space that have yet to be exposed to, or explored by, the user. This project will build on previous successful interfaces for novel content discovery, and look at the more sustained effect of the ``nudges’’ given by such explanatory interfaces. It will also study the difference in the effect of these interfaces when these blind-spots are intentional (e.g., the user is uninterested but the system does not know this yet), and unintentional (e.g., the user is simply unaware of this part of the search space but is interested).
    Contact: Nava Tintarev (via

Vacancies connected to data science & Delft Data Science

See Delft Data Science.