Keynote Speakers

About the workshop

Data mining algorithms allow us to make sense of large quantities of data and obtain insights from it. This includes mining from different kinds of data, including tabular, sequence, time series, graph, etc.  

Classic algorithms are often designed for a specific, often narrow, task (i.e., task-centered), yet do not take into account the specific goals of the users (which may well differ from user to user). For instance, when looking for outputs useful for automating a decision process, the data mining algorithm user may look for patterns that do not go against their prior knowledge. Oppositely, another user may look for ‘surprising’ outputs when their goal is to understand previously unknown phenomena (knowledge discovery).

However, most existing data mining algorithms can only be customised to the extent allowed by their pre-defined parameters, which are not guaranteed to be able to align with the users' needs. Furthermore, specifying preferences via parameters often requires extensive data mining expertise. This is often not the case in the application domains, where users tend to be domain experts with knowledge about the source of the data. 

Recently, researchers in the data mining field have turned their attention towards integrating users in the mining process, trying to take into account both their priors and their goals. However, current approaches tend to model users in ways that are closely tied to the specific task requirements, limiting their generalisability. As such, there is not one recipe that can be applied to data mining tasks in general. At a higher level, it is unclear how user needs and domain expertise can be modelled so as to improve data mining approaches.

The goal of this workshop is to bring together researchers and practitioners from (interactive) data mining, interpretable machine learning, data visualization, human-computer interaction, active learning, machine teaching, and cognitive science, and to facilitate the discussion between experts to identify and formalise the problem of integrating human preferences and situation-specificness in data mining approaches. We aim to foster discussions about the key research questions/challenges, including but not limited to: 

• How can we rigorously define the goal of incorporating the human (in an interactive manner) in the data mining process?

• How can we categorize existing algorithms and use cases with respect to how human feedback may be incorporated?

• How to evaluate algorithms that rely on human feedback? Can the effect of human guidance be measured reliably in a benchmark setting?

• How can we empower users to guide algorithm outputs toward their goals or preferences in a user-friendly way? Is there a principled way to incorporate user preferences into the algorithm? 

We specifically encourage submissions/discussions on the topic of issues encountered when applying (interactive) data mining to real-world use cases, as well as innovations towards solving those issues. We hope that this will inspire future collaborations, both to solve specific research problems, and to steer the community on the subject of human-centered data mining. [Check our call for the papers here!]

Location

The HuMINE workshop will be held as part of the ECML-PKDD 2025 conference, in Porto, Portugal. 

Information about the precise location, dates, and time will be added later.