Background

The way we interact, make decisions, work, study, and socialize is rapidly evolving. Much of our leisure as well as work activities are social. Whether we are planning to watch a movie with our family or work on a joint task, adaptive and personalized systems for groups are becoming more and more prevalent. From simple chat-based apps to more sophisticated collaborative platforms, it has become almost impossible to imagine our daily lives without relying on such systems. Furthermore, adaptive and personalized systems for groups, can already help existing permanent or ephemeral groups to select a restaurant for a dinner, or a destination to travel to for a vacation, a Points-of-Interest tour in a city, a movie to watch, and so on. In this case the system can make recommendations, or a sequence of recommendations (as in a tour) for the group, having the goal to keep all the group members as satisfied as possible with the suggested items [11].

The first, and mostly researched, task of such a system is to combine individual preferences into a group model, based on which items of interest for the whole group could be found. Secondly, as recognized in the more recent works [3, 12, 13], another task is to support the group in their decision-making process, hence not just suggesting an item or a rigid ranked list of items, but rather to truly help the group to reach a joint decision. To increase the effectiveness of these functionalities, the system can make use of various individual, group, interrelationship or even decision-process features, such as, individual personalities, group type (for example family, close friends, or colleagues), close, highly emotional or loosely coupled relationships within the group, having a well defined, structured decision making-process, or a casual open conversation [6].

People are increasingly participating in new forms of work, such as crowdsourcing platforms and the gig economy, while companies are increasingly posting their jobs online to attract the creativity of heterogeneous contributors from all over the world [15]. From individuals completing mini jobs on Amazon Mechanical Turk to ad-hoc groups organised in “crowd farms” to uptake and complete complex software projects, large-scale online work is increasingly gaining track [20, 21]. The pandemic has hastened this transition to intelligent work automation, resulting in a flood of new workers who have turned to digital labor platforms to repurpose their talents and increase their employability [2, 7]. Due to the scale of the involved users and tasks, online platform work is typically algorithmically mediated [1, 17], in order to manage the hiring, filtering, and placement of the online workers to tasks and, for more complex tasks, to groups. Unfortunately, the majority of these algorithms are designed to supervise work in a top-down and controlling manner, and they rely only on external reinforcement without involving the users in any of the work management processes [14].

For example, most team formation algorithms rely on a fixed, pre-decided, and non-adaptive user model to decide which person should work with whom, and to suggest tasks to the formed groups. The algorithms rarely offer group decision-making support, provide explanations, or adapt their matching decisions based on user input.

With very few and mostly research exceptions [8, 18, 19], existing online team formation algorithms are designed to function according to the Taylorism work management style, micro-managing the online workers and distrusting their initiative [10, 22]. Such algorithm design approaches may be appropriate for group tasks that are well-defined and with known knowledge boundaries, such as micro-tasks. They are however highly ineffective for group tasks that are complex, open-ended, and have multiple knowledge interdependencies [5]. Examples of such tasks, which require a different and more user-centered approach, include radical innovation, “wicked” problems, large-scale research, and group tasks within the creative industries like software and game creation, advertising, or art generation.

Research shows that these tasks benefit the most from groups that are formed in a way that affords the involved users the freedom to affect the team formation outcome, based on their own experience, creativity or intuition, and to be able to adapt and re-plan the recommended group structures according to the changing needs of the task at-hand. Performance aside, allowing people to affect the decision of who they will be recommended to work with has also been found to increase one’s sense of control over their own work and ideas, which in turn promotes creativity, intrinsic motivation, team cohesion, and user well-being [4, 9, 16].

In this workshop, we want to cover both lines of research related to adaptive personalized systems for groups. The transversal topic that we are emphasizing is the importance of putting the user “back in the loop”, that is, empowering the user to enjoy a level of control in these systems. Hence, we aim to discuss (i) radical new ways to design team recommendation systems and algorithms that afford users the freedom to decide with whom to collaborate with and how, to solve the open-ended problem at hand, all while guaranteeing the quality of the creative work, and, (ii) models, algorithms and systems that suggest items for groups to experience, through approaches that support joint and incremental group decision-making.