Augmented Reality (AR) systems that overlay digital instructions onto the physical world can enhance user performance in complex industrial tasks. Procedural instructions can often be derived from existing approved technical documentation, as reuse reduces authoring effort and ensures compliance with established standards. This paper examines the potential of incorporating a Large Language Model (LLM) to add an intelligent, conversational AR assistant to a static AR manual. We conducted an A/B user study (N = 36) evaluating the effects of adding a conversational assistance layer for a hands-on task on a juice mixer laboratory installation using a HoloLens 2. The baseline condition provided a PDF manual requiring manual step navigation, complemented by situated visual anchors. The second variant supplemented this interface with a conversational assistant that could interpret user queries, provide direct verbal guidance, and automatically jump to the relevant page in the manual. We evaluated these interfaces across three distinct scenarios: a linear task, a non-linear task requiring users to jump between pages for troubleshooting, and a task-handoff where users had to identify the current state of a partially completed procedure. Our findings indicate that despite longer total task completion times, participants using the AI assistant spent significantly less time actively working in the linear and non-linear scenarios, indicating improved task efficiency beyond system latency. Eye-tracking analysis supports the efficiency gain observation as the conversational interface allowed users to focus their visual attention on actively understanding and learning the procedural task. This study highlights the potential of LLM-powered agents for AR guidance, suggesting that overcoming system latency is the critical barrier to their practical deployment in industrial fields.

Point-of-interest (POI) recommender systems help users discover relevant locations, but their effectiveness is often compromised by popularity bias, which disadvantages less popular, yet potentially meaningful places. This paper addresses this challenge by evaluating the effectiveness of context-aware models and calibrated popularity techniques as strategies for mitigating popularity bias. Using four real-world POI datasets (Brightkite, Foursquare, Gowalla, and Yelp), we analyze the individual and combined effects of these approaches on recommendation accuracy and popularity bias. Our results reveal that context-aware models cannot be considered a uniform solution, as the models studied exhibit divergent impacts on accuracy and bias. In contrast, calibration techniques can effectively align recommendation popularity with user preferences, provided there is a careful balance between accuracy and bias mitigation. Notably, the combination of calibration and context-awareness yields recommendations that balance accuracy and close alignment with the users' popularity profiles, i.e., popularity calibration.

Many IR systems project harmful societal biases, including gender bias, in their retrieved contents. Uncovering and addressing such biases requires grounded bias measurement principles. However, defining reliable bias metrics for search results is challenging, particularly due to the difficulties in capturing gender-related tendencies in the retrieved documents. In this work, we propose a new framework for search result bias measurement. Within this framework, we first revisit the current metrics for representative search result bias (RepSRB) that are based on the occurrence of gender-specific language in the search results. Addressing their limitations, we additionally propose a metric for comparative search result bias (ComSRB) measurement and integrate it into our framework. ComSRB defines bias as the skew in the set of retrieved documents in response to a non-gendered query toward those for male/female-specific variations of the same query. We evaluate ComSRB against RepSRB on a recent collection of bias-sensitive topics and documents from the MS MARCO collection, using pre-trained bi-encoder and cross-encoder IR models. Our analyses show that, while existing metrics are highly sensitive to the wordings and linguistic formulations, the proposed ComSRB metric mitigates this issue by focusing on the deviations of a retrieval list from its explicitly biased variants, avoiding the need for sub-optimal content analysis processes.

Biases in algorithmic systems have led to discrimination against historically disadvantaged groups, including the reinforcement of outdated gender stereotypes. While a substantial body of research addresses biases in algorithms and underlying data, in this work, we study if and how users themselves reflect these biases in their interactions with systems, which expectedly leads to the further manifestation of biases. More specifically, we investigate the replication of stereotypical gender representations by users in formulating online search queries. Following prototype theory, we define the disproportionate mention of the gender that does not conform to the prototypical representative of a searched domain (e.g., “male nurse”) as an indication of bias. In a pilot study with 224 US participants and a main study with 400 UK participants, we find clear evidence of gender biases in formulating search queries. We also report the effects of an educative text on user behaviour and highlight the wish of users to learn about bias-mitigating strategies in their interactions with search engines.

Popularity bias in recommendation lists refers to over-representation of popular content and is a challenge for many recommendation algorithms. Previous research has suggested several offline metrics to quantify popularity bias, which commonly relate the popularity of items in users' recommendation lists to the popularity of items in their interaction history. Discrepancies between these two factors are referred to as popularity miscalibration. While popularity metrics provide a straightforward and well-defined means to measure popularity bias, it is unknown whether they actually reflect users' perception of popularity bias.

Societal biases resonate in the retrieved contents of information retrieval (IR) systems, resulting in reinforcing existing stereotypes. Approaching this issue requires established measures of fairness in respect to the representation of various social groups in retrieval results, as well as methods to mitigate such biases, particularly in the light of the advances in deep ranking models. In this work, we first provide a novel framework to measure the fairness in the retrieved text contents of ranking models. Introducing a ranker-agnostic measurement, the framework also enables the disentanglement of the effect on fairness of collection from that of rankers. To mitigate these biases, we propose AdvBert, a ranking model achieved by adapting adversarial bias mitigation for IR, which jointly learns to predict relevance and remove protected attributes. We conduct experiments on two passage retrieval collections (MSMARCO Passage Re-ranking and TREC Deep Learning 2019 Passage Re-ranking), which we extend by fairness annotations of a selected subset of queries regarding gender attributes. Our results on the MSMARCO benchmark show that, (1) all ranking models are less fair in comparison with ranker-agnostic baselines, and (2) the fairness of Bert rankers significantly improves when using the proposed AdvBert models. Lastly, we investigate the trade-off between fairness and utility, showing that we can maintain the significant improvements in fairness without any significant loss in utility.

In online social learning environments, tagging has demonstrated its potential to facilitate search, to improve recommendations and to foster reflection and learning.Studies have shown that shared understanding needs to be established in the group as a prerequisite for learning. We hypothesise that this can be fostered through tag recommendation strategies that contribute to semantic stabilization. In this study, we investigate the application of two tag recommenders that are inspired by models of human memory: (i) the base-level learning equation BLL and (ii) Minerva. BLL models the frequency and recency of tag use while Minerva is based on frequency of tag use and semantic context. We test the impact of both tag recommenders on semantic stabilization in an online study with 56 students completing a group-based inquiry learning project in school. We find that displaying tags from other group members contributes significantly to semantic stabilization in the group, as compared to a strategy where tags from the students' individual vocabularies are used. Testing for the accuracy of the different recommenders revealed that algorithms using frequency counts such as BLL performed better when individual tags were recommended. When group tags were recommended, the Minerva algorithm performed better. We conclude that tag recommenders, exposing learners to each other's tag choices by simulating search processes on learners' semantic memory structures, show potential to support semantic stabilization and thus, inquiry-based learning in groups.

The European research project weSPOT aims at supporting science learning in secondary and higher education. The underlying pedagogical approach, inquiry-based learning, is often criticized for the lack in teaching learning content and for overburden novice learners. To fill this gap, we developed the Formal Concept Analysis (FCA) tool which is used by teachers to define a knowledge domain, i.e. The objects, attributes and their relations to each other. Learning resources can be assigned to subsets of objects and attributes. By navigating through the concept lattice students get an overview of the topic. They learn by consuming learning resources, either in a self-regulated way, by interacting with the nodes of the lattice, or when following a recommender system which suggests learning resources based on the domain model and defined pedagogical rules. The paper describes the FCA tool and how it is used by teachers and students, and the recommendation strategy that supports students when browsing the knowledge domain.

Adaptive learning systems aim to address a learner's specific needs, considering factors such as prior knowledge, learning efficiency, learning goals and motivation. Especially in distance education, often directed to adult learners with full-time jobs, it is very important to provide assistance to counteract high dropout rates. This paper describes an approach on how to support adult learners through the adoption of personalisation and guidance in Moodle. The implementation grounds on the combination of two pedagogical theories, competence-based learning and Self-Regulated Learning (SRL). Three-phases were used to roughly frame the design of the SRL learning flow, where the individual phases are supported by competence-based guidance. In this way Moodle is extended from a teacher and course management to a learner-centric system. This work has been implemented and evaluated in the course of a European project that targets vocational training of heat pump installers.

Learner models form the basis of adaptive learning systems, representing what the system knows about a learner. Knowledge about a learner’s expertise within a learning domain plays an essential role when recommending learning objects as well as when supporting a learner’s reflection and awareness. In this paper we present the Competence Analyser tool to construct knowledge structures as a foundation for learner models in accordance with a psychologically sound framework, the Competence based Knowledge Space Theory (CbKST), which supports performance assessment in user-adaptive learning systems. The tool enables the learner to actively influence the learning profile creation by setting learning goals and defining prior knowledge in the context of learning domains. A first prototype has been developed and evaluated within the frame of the EC-funded TARGET project that provides a game-based learning environment with adaptive story recommendation. The paper outlines how to foster awareness and self-regulation while building learner and domain models for adaptive learning systems using a graph-based GUI tool.

This paper presents a novel approach how learning paths consisting of game units can be created and adapted to learners based on their behavior during the game play. Non-invasive assessment procedures interpret the behavior and calculate information about the competences of the learners. A user model holds probabilistic information on the competence profile. Based on this competence profile game units/stories are recommended fitting to the actual competence state of the learner. This approach is part of the EC-funded TARGET project which provides the technical infrastructure regarding the 3D virtual game environment. The innovative part of this paper is the adaptive learning strategy and how it can be included in a game-based environment. The user perspective is demonstrated on a concrete scenario where the learner has to solve a task in the game-based environment.

Emerging e-learning systems cover a wide range of application fields satisfying user expectations such as rapid information access or flexibility in time and location. Still a challenge is to automatically respond to individual learner’s perceptions. This paper presents a glimpse on concepts, evolution and practical implementation of the Artificial Mentor, a highly flexible, adaptive software component. The Artificial Mentor selects and triggers micro adaptations within a learning environment, intended to enhance the learning process. It was initially developed in the setting of a 3D virtual learning environment to address modern learner’s individual learning needs.