Anett Hoppe
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Anett Hoppe has a total of 76 co-authors. Here are the top co-authors by number of shared publications:
- Ralph Ewerth
TIB – Leibniz Information Centre for Science and Technology
33 shared publications - Arthur Brack
TIB – Leibniz Information Centre for Science and Technology, Hannover, Germany
7 shared publications - Elias Entrup
TIB - Leibniz Information Centre for Science and Technology, Hannover, Germany
6 shared publications - Markos Stamatakis
TIB – Leibniz Information Centre for Science and Technology
6 shared publications - Wolfgang Gritz
TIB – Leibniz Information Centre for Science and Technology
6 shared publications - Ratan Sebastian
Leibniz University Hannover
5 shared publications
Latest Publications
- Open Access
- 6 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 27-37
Robust Chart Information Extraction via an LVLM Ensemble and Pairwise Reasoning
Data visualizations are essential for communicating findings in scientific literature since they convey significant insights such as experimental results, trends, and distributions. However, due to the diverse nature of charts in real-world settings, automatic chart information extraction (CIE) remains challenging. Current state-of-the-art models demonstrate limited generalization across a variety of charts and benchmarks, performing adequately on basic chart types but failing to recognize complex visualizations, particularly when the characteristics (e.g., resolution, axis, complexity) differ from the datasets used to train them. To address these challenges, we make several contributions. (1) We propose an ensemble-based approach that aims to leverage the synergies of models trained for CIE and general-purpose large vision-language models (LVLMs). (2) We propose several inference strategies for our ensemble approach, including a multiple choice and two binary choice approaches based on league- and tournament-style inference to prevent excessive context length and positional bias of generative AI models. (3) For a fair comparison, we suggest a novel evaluation metric that resolves fundamental problems of existing metrics used for CIE task. (4) Experimental results demonstrate that our proposed ensemble approach can efficiently leverage synergies across various models achieving superior results and generalization across four benchmarks in comparison to the state of the art.
- 6 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 27-37
Open AccessRobust Chart Information Extraction via an LVLM Ensemble and Pairwise Reasoning
Data visualizations are essential for communicating findings in scientific literature since they convey significant insights such as experimental results, trends, and distributions. However, due to the diverse nature of charts in real-world settings, automatic chart information extraction (CIE) remains challenging. Current state-of-the-art models demonstrate limited generalization across a variety of charts and benchmarks, performing adequately on basic chart types but failing to recognize complex visualizations, particularly when the characteristics (e.g., resolution, axis, complexity) differ from the datasets used to train them. To address these challenges, we make several contributions. (1) We propose an ensemble-based approach that aims to leverage the synergies of models trained for CIE and general-purpose large vision-language models (LVLMs). (2) We propose several inference strategies for our ensemble approach, including a multiple choice and two binary choice approaches based on league- and tournament-style inference to prevent excessive context length and positional bias of generative AI models. (3) For a fair comparison, we suggest a novel evaluation metric that resolves fundamental problems of existing metrics used for CIE task. (4) Experimental results demonstrate that our proposed ensemble approach can efficiently leverage synergies across various models achieving superior results and generalization across four benchmarks in comparison to the state of the art.
- Open Access
- 6 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 9523-9525
IWILDS'26: The 6th International Workshop on Investigating Learning during Web Search and AI-Mediated Information Access
The emergence of conversational AI, retrieval-augmented generation and adaptive recommendation systems has fundamentally reshaped how learners access, evaluate and construct knowledge from the web. Yet our understanding of how these shifts affect learning processes, and how retrieval systems should be designed to support them still remains limited. The sixth edition of the IWILDS workshop series brings together researchers from information retrieval, knowledge management, learning analytics, user modeling and educational psychology to address this gap. The workshop focuses on three interconnected research questions: how learners formulate and reformulate queries and prompts across search, RAG and conversational paradigms; what evaluation frameworks can capture learning outcomes in AI-mediated information environments; and how to construct datasets that reflect contemporary learning-in-the-wild. IWILDS'26 combines paper-based sessions (paper presentations and invited talk) with interactive phases focused on structured discussions about research questions and a collaboration fair. In its sixth year, the workshop builds on a sustained community anchored in IR and learning sciences, returning to CIKM with a sharpened focus aligned with CIKM's core interests in information retrieval, user modeling and knowledge management.
- 6 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 9523-9525
Open AccessIWILDS'26: The 6th International Workshop on Investigating Learning during Web Search and AI-Mediated Information Access
The emergence of conversational AI, retrieval-augmented generation and adaptive recommendation systems has fundamentally reshaped how learners access, evaluate and construct knowledge from the web. Yet our understanding of how these shifts affect learning processes, and how retrieval systems should be designed to support them still remains limited. The sixth edition of the IWILDS workshop series brings together researchers from information retrieval, knowledge management, learning analytics, user modeling and educational psychology to address this gap. The workshop focuses on three interconnected research questions: how learners formulate and reformulate queries and prompts across search, RAG and conversational paradigms; what evaluation frameworks can capture learning outcomes in AI-mediated information environments; and how to construct datasets that reflect contemporary learning-in-the-wild. IWILDS'26 combines paper-based sessions (paper presentations and invited talk) with interactive phases focused on structured discussions about research questions and a collaboration fair. In its sixth year, the workshop builds on a sustained community anchored in IR and learning sciences, returning to CIKM with a sharpened focus aligned with CIKM's core interests in information retrieval, user modeling and knowledge management.
- Open Access
- 13 October 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-5
Real-Time, Context-Specific Educational Metadata Enrichment of Learning Resources on the Web
Web resources rarely carry the educational metadata that teachers use to judge whether a resource fits their lesson, and dedicated educational repositories that do carry it cover only a fraction of what teachers find through web search. Closing this gap in real time, at the point of search, requires classifiers that are fast enough to run per query and that target a vocabulary specific to the teacher’s curriculum and grade structure, for which labelled training data is scarce. We investigate whether Large Language Model (LLM)-generated labels and paraphrases can train small, fast classifiers for this task. Across nine LRMI-aligned educational metadata properties for STEM education in Lower Saxony, Germany, matching model complexity to property difficulty reaches micro F1 > 0.8 for every property at inference latencies under 15ms, with models compact enough for client-side, browser-based deployment. We report this as preliminary, ongoing work: evaluation for six properties currently relies on LLM-generated labels validated against a 50-page human-annotated sample, and train/test splits do not yet separate resources by website, which we flag as open validity questions for the digital libraries community to weigh in on.
- 13 October 2026
- Published by Association for Computing Machinery (ACM)
- p. 1-5
Open AccessReal-Time, Context-Specific Educational Metadata Enrichment of Learning Resources on the Web
Web resources rarely carry the educational metadata that teachers use to judge whether a resource fits their lesson, and dedicated educational repositories that do carry it cover only a fraction of what teachers find through web search. Closing this gap in real time, at the point of search, requires classifiers that are fast enough to run per query and that target a vocabulary specific to the teacher’s curriculum and grade structure, for which labelled training data is scarce. We investigate whether Large Language Model (LLM)-generated labels and paraphrases can train small, fast classifiers for this task. Across nine LRMI-aligned educational metadata properties for STEM education in Lower Saxony, Germany, matching model complexity to property difficulty reaches micro F1 > 0.8 for every property at inference latencies under 15ms, with models compact enough for client-side, browser-based deployment. We report this as preliminary, ongoing work: evaluation for six properties currently relies on LLM-generated labels validated against a 50-page human-annotated sample, and train/test splits do not yet separate resources by website, which we flag as open validity questions for the digital libraries community to weigh in on.
- 7 October 2026
- Published by Springer Nature
- p. 577-592
Are LLM-Based Chatbots Good Enough to Support Computer Science Students in Multiple-Choice Exercises?
- 7 October 2026
- Published by Springer Nature
- p. 577-592
Are LLM-Based Chatbots Good Enough to Support Computer Science Students in Multiple-Choice Exercises?
- Open Access
- 25 May 2026
- Published by Association for Computing Machinery (ACM)
- p. 70-74
Examining Gender and Racial Bias in Vision-language Models within Colonial Contexts
Vision-language models (VLMs) are increasingly used for image retrieval in cultural heritage collections. Yet these models are trained and evaluated on modern web-scraped data which might cause unobserved biases when applied to historical imagery. While prior work has documented biases related to gender and skin tone in VLMs using contemporary datasets, no evaluation datasets exist for historical archives. In this paper, we address this gap by introducing a new dataset for evaluating demographic bias in VLMs on photograph collections from colonial contexts. Our dataset comprises annotated photographs for demographic queries spanning gender and race. Using this dataset, we examine bias in multiple VLMs through retrieval performance across queries, measuring disparities using established information retrieval metrics. Experimental results show that gender biases documented in contemporary datasets persist on historical imagery, with all tested models retrieving certain demographic groups substantially better than others. Notably, models specifically trained for debiasing fail to mitigate bias on historical images and can introduce new disparities across skin tones. These findings underscore the need for domain-specific debiasing approaches and caution against deploying off-the-shelf VLMs in archival contexts without proper evaluation.
- 25 May 2026
- Published by Association for Computing Machinery (ACM)
- p. 70-74
Open AccessExamining Gender and Racial Bias in Vision-language Models within Colonial Contexts
Vision-language models (VLMs) are increasingly used for image retrieval in cultural heritage collections. Yet these models are trained and evaluated on modern web-scraped data which might cause unobserved biases when applied to historical imagery. While prior work has documented biases related to gender and skin tone in VLMs using contemporary datasets, no evaluation datasets exist for historical archives. In this paper, we address this gap by introducing a new dataset for evaluating demographic bias in VLMs on photograph collections from colonial contexts. Our dataset comprises annotated photographs for demographic queries spanning gender and race. Using this dataset, we examine bias in multiple VLMs through retrieval performance across queries, measuring disparities using established information retrieval metrics. Experimental results show that gender biases documented in contemporary datasets persist on historical imagery, with all tested models retrieving certain demographic groups substantially better than others. Notably, models specifically trained for debiasing fail to mitigate bias on historical images and can introduce new disparities across skin tones. These findings underscore the need for domain-specific debiasing approaches and caution against deploying off-the-shelf VLMs in archival contexts without proper evaluation.
- Open Access
- 17 September 2025
- Published by Royal Society of Chemistry (RSC) in Chemistry Education Research and Practice
- Vol. 27 (1) , 351-367
Towards defining, assessing and modelling competency levels in stoichiometry
Stoichiometry is a significant yet challenging topic in chemistry education. While extensive research has explored students’ conceptions, difficulties, and learning approaches, this study adopts a competency-based approach to introduce a new model defining three competency levels in stoichiometry. The stoichiometry competency level model (StoiCoLe model) offers a framework for evaluating students’ performance in algorithmic stoichiometry. To test the assumptions of the StoiCoLe model, a 40-item test was developed to measure and categorise student's competencies according to the model's levels. Using data from 289 students enrolled in an introductory chemistry course across three semesters, psychometric properties and model assumptions were analysed through Rasch-analysis and item processing times. The results indicate that there is sufficient psychometric reliability in the categorization of students according to the StoiCoLe model. However, both the item difficulty and the processing times are only partially consistent with the assumptions of the model and indicate an adaptation of the model. In line with prior studies, the majority of students exhibited lower competency levels. These findings are discussed in terms of how the competency-based approach can enhance relevant competencies and contribute to literature on chemistry education in stoichiometry.
- 17 September 2025
- Published by Royal Society of Chemistry (RSC) in Chemistry Education Research and Practice
- Vol. 27 (1) , 351-367
Open AccessTowards defining, assessing and modelling competency levels in stoichiometry
Stoichiometry is a significant yet challenging topic in chemistry education. While extensive research has explored students’ conceptions, difficulties, and learning approaches, this study adopts a competency-based approach to introduce a new model defining three competency levels in stoichiometry. The stoichiometry competency level model (StoiCoLe model) offers a framework for evaluating students’ performance in algorithmic stoichiometry. To test the assumptions of the StoiCoLe model, a 40-item test was developed to measure and categorise student's competencies according to the model's levels. Using data from 289 students enrolled in an introductory chemistry course across three semesters, psychometric properties and model assumptions were analysed through Rasch-analysis and item processing times. The results indicate that there is sufficient psychometric reliability in the categorization of students according to the StoiCoLe model. However, both the item difficulty and the processing times are only partially consistent with the assumptions of the model and indicate an adaptation of the model. In line with prior studies, the majority of students exhibited lower competency levels. These findings are discussed in terms of how the competency-based approach can enhance relevant competencies and contribute to literature on chemistry education in stoichiometry.
- Open Access
- 24 August 2025
- Published by Taylor & Francis in International Journal of Science Education
- Vol. 48 (15) , 1-23
Investigating evidence-oriented generation of synthetic text data with a generative large language model in science education
- 24 August 2025
- Published by Taylor & Francis in International Journal of Science Education
- Vol. 48 (15) , 1-23
Open AccessInvestigating evidence-oriented generation of synthetic text data with a generative large language model in science education
- 15 July 2025
- Published by Springer Nature
- p. 236-250
Supporting Teachers Through AI-Augmented Web Search for Educational Resource Discovery
- 15 July 2025
- Published by Springer Nature
- p. 236-250
Supporting Teachers Through AI-Augmented Web Search for Educational Resource Discovery
- 15 July 2025
- Published by Springer Nature
- p. 453-467
From Formulas to Figures: How Visual Elements Impact User Interactions in Educational Videos
- 15 July 2025
- Published by Springer Nature
- p. 453-467
From Formulas to Figures: How Visual Elements Impact User Interactions in Educational Videos
- 15 July 2025
- Published by Springer Nature
- p. 305-319
Enhancing the Learning Experience: Using Vision-Language Models to Generate Questions for Educational Videos
- 15 July 2025
- Published by Springer Nature
- p. 305-319
Enhancing the Learning Experience: Using Vision-Language Models to Generate Questions for Educational Videos