(Forschungs)Seminar Web Engineering (WS 2026/2027)

Research Questions

• What practical risks do current AI agents pose? In what form can these risks manifest?
• Where have AI Agents already caused issues? Which concrete incidents already happened, caused by or involving AI Agents?
• How can risks of AI Agents be identified in advance and managed accordingly?

Literature

• National Institute of Standards and Technology, “Artificial Intelligence Risk Management Framework: AI RMF 1.0,” U.S. Department of Commerce, NIST AI 100-1, Jan. 2023. doi: 10.6028/NIST.AI.100-1
• J. Kraprayoon, Z. Williams, and R. Fayyaz, “AI Agent Governance: A Field Guide.” May 2025. doi: 10.48550/arXiv.2505.21808
• S. Z. U. Rashid, I. Montasir, A. Haq, M. T. Ahmmed, and M. M. Alam, “Securing Agentic AI: Threats, Risks, and Mitigation,” in International Conference on Advancement In Cyber Security and Digital Forensics, Springer, 2025, pp. 683–698.
• Own research

Research Questions

• What techniques exist to sandbox autonomous AI agents?
• Which problems are solved through sandboxing, and which are not?
• What overhead and limitations do current sandboxing techniques have?

Literature

• L. Meng, H. Feng, I. Shumailov, and E. Fernandes, “ceLLMate: Sandboxing Browser AI Agents.” Mar 2026. doi: 10.48550/arXiv.2512.12594
• Own research

Research Questions

• How can human users monitor actions of an autonomous AI agent?
• How much detail and insight into the agent’s activities are required for effective oversight and allowing intervention by human users?
• What options for intervention or correction of the behavior of an autonomous AI agent exist?
• How can automation bias and approval fatigue be avoided in these kinds of systems?

Literature

• L. Zhu, Q. Lu, M. Ding, S. U. Lee, and C. Wang, “Designing meaningful human oversight in AI,” AI and Ethics, vol. 6, no. 3, p. 286, May 2026, doi: 10.1007/s43681-026-01147-7
• S. Gaube et al., “Keeping an Eye on AI: A Framework for Effective Human Oversight of AI Systems.” 2026. doi: 10.48550/arXiv.2605.16278
• L. Dong, Q. Lu, and L. Zhu, “AgentOps: Enabling Observability of LLM Agents.” 2024. doi: 10.48550/arXiv.2411.05285
• Own research

Research Questions

• What is the difference between qualitative and quantitative research methodologies?
• Provide an overview of different methodologies, explain how they work, when and in which field they are applied.
• Your use case is the use of data for chatbot research, so which kind of data is used and how it is used.
• Qualitative: Conduct a survey with researchers and professionals. Research on how to formulate survey questions and send them out to experts.
• Quantitative: Analyse existing datasets. Create search queries for databases and analyse the found datasets on specific aspects.
• Describe your methodologies in detail in a research log and note down any meetings, decisions, changes, and outcomes.
• Reflect on the differences in using both methodologies and provide insights into new learnings, pain points, and how you can apply the different methodologies you found in other areas.

Literature

• Own research
• Creswell, J. W. (2014). Research design : Qualitative, quantitative, and mixed methods approaches. (4. ed., in). SAGE. https://katalog.bibliothek.tu-chemnitz.de/Record/0008891954
  Currently only available in printed version
• Trochim, W., Donnelly, J. P., & Arora, K. (2016). Research methods: The essential knowledge base.
• Savin-Baden, M., & Major, C H. (2025). Qualitative research: The essential guide to theory and practice (Second edition.). Routledge. https://katalog.bibliothek.tu-chemnitz.de/Record/0-1939098580 (esp. Chapter 23 – Interviews)
  Alternative access in WebOPAC
• Developing and Validating a Comprehensive Literature Search Strategy https://library.cumc.columbia.edu/kb/developing-comprehensive-search-query

Research Questions

• Which datasets are available for chatbot research / design on public platforms aside from conversational training? Analyse existing chatbot related datasets to answer this question.
• Document every step in a research journal (i.e. what did you work on, what is the result, etc.) and reflect on the current landscape of chatbot datasets for research / design.
• Define inclusion (i.e. English language) and exclusion criteria (i.e. language models, special pre-trained models).
• Create search queries and use them on different platforms (some possible platforms are listed below).
• Collect a meaningful number of datasets and filter them according to your inclusion/exclusion criteria. The goal is to have at least 50 diverse datasets.
• Extract relevant metadata (size, file types, keywords, upload date, license, publications platform, etc.), document it, analyze this information with statistical data and present your analysis results with well-chosen graphs.

Literature

• Own research
• Agarwal S, Agarwal B, Gupta R (2022), Chatbots and virtual assistants: a bibliometric analysis. Library Hi Tech, Vol. 40 No. 4 pp. 1013–1030. https://doi.org/10.1108/LHT-09-2021-0330
• Khan AM, Loan FA, Parray UY, Rashid S (2024), Global overview of research data repositories: an analysis of re3data registry. Information Discovery and Delivery, Vol. 52 No. 1 pp. 53–61. https://doi.org/10.1108/IDD-07-2022-0069
• Marcial, L.H. and Hemminger, B.M. (2010), Scientific data repositories on the Web: An initial survey. J. Am. Soc. Inf. Sci., 61: 2029-2048. https://doi.org/10.1002/asi.21339
• https://ieee-dataport.org/datasets
• https://www.nature.com/search?journal=sdata&q=chatbot
• https://figshare.com/
• https://data.mendeley.com/
• https://dataverse.harvard.edu/
• https://osf.io/
• https://datasetsearch.research.google.com/

Research Questions

• What is Object Detection? What is the role of training data and which impact can it have?
• What is Augmentation and synthetic Generation? How are they used in training or finetuning object detection models?
• Create augmented synthetic data for the detection of UI elements in chatbot interfaces, you can use LLMs for this, all code nevertheless has to be checked by you for being usable and sensible.
• Write scripts which create html code for different chatbot interfaces. These should include different interface elements in different variations (shapes, icons, position, etc.). Make sure that the objects are customizable on these dimensions and are used like building blocks which can be combined in different ways. Chat messages and buttons should contain randomized (but realistic) text to avoid accidental overfitting. The scripts should then create every combination of patterns (keep in mind that these combinations should still make sense).
• Use the created html code for screenshots and use augmentation methods (sharpness, saturation, etc.) to enhance the number of created data.
• Critically evaluate where the results are more and less realistic and how the augmentation changes this.

Literature

• Own research
• Chen, Y., Li, Y., Kong, T., Qi, L., Chu, R., Li, L., & Jia, J. (2021). Scale-aware Automatic Augmentation for Object Detection. 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 9558-9567. https://doi.org/10.48550/arXiv.2103.17220.
• Traubinger, V., Heil, S., Mithari, A.A., Gaedke, M. (2026). Left, Right, Down, or Round: Exploring the Design of Patterns in Chatbot Interfaces and How Users Like Them. In: Følstad, A., et al. Chatbots and Human-Centered AI. CONVERSATIONS 2025. Lecture Notes in Computer Science, vol 16440. Springer, Cham. https://doi.org/10.1007/978-3-032-26717-7_4
• Mulong Xie, Sidong Feng, Zhenchang Xing, Jieshan Chen, and Chunyang Chen. 2020. UIED: a hybrid tool for GUI element detection. In Proceedings of the 28th ACM Joint Meeting on European Software Engineering Conference and Symposium on the Foundations of Software Engineering (ESEC/FSE 2020). Association for Computing Machinery, New York, NY, USA, 1655–1659. https://doi.org/10.1145/3368089.3417940
• Casado-García, Á., Domínguez, C., García-Domínguez, M. et al. CLoDSA: a tool for augmentation in classification, localization, detection, semantic segmentation and instance segmentation tasks. BMC Bioinformatics 20, 323 (2019). https://doi.org/10.1186/s12859-019-2931-1
• Jieshan Chen, Mulong Xie, Zhenchang Xing, Chunyang Chen, Xiwei Xu, Liming Zhu, and Guoqiang Li. 2020. Object detection for graphical user interface: old fashioned or deep learning or a combination? In Proceedings of the 28th ACM Joint Meeting on European Software Engineering Conference and Symposium on the Foundations of Software Engineering (ESEC/FSE 2020). Association for Computing Machinery, New York, NY, USA, 1202–1214. https://doi.org/10.1145/3368089.3409691

Research Questions

• Introduction: Research outputs are classified by subject in many different ways. Libraries and repositories use the Dewey Decimal Classification (DDC), OpenAlex assigns domains, fields and subfields derived from the Scopus ASJC scheme, the Open Research Knowledge Graph (ORKG) maintains its own taxonomy of research fields, and Wikidata and DBpedia describe academic disciplines as part of their knowledge graphs. When a knowledge graph integrates research outputs from several sources, the same output therefore carries subject classes from different systems. A concordance between these classifications – a set of mappings expressed with SKOS mapping relations – makes it possible to find research outputs by field regardless of the classification the source delivered. The central difficulty is granularity: a DDC class can be broader or narrower than an ORKG research field, and a Wikidata discipline can correspond to either.
• Which subject classifications are used for research outputs (e.g., DDC, the OECD Fields of Research and Development, OpenAlex/ASJC, ORKG Research Fields, ACM CCS, Wikidata and DBpedia disciplines)? How do they differ in purpose, structure, depth, granularity and licensing?
• Which concordances and mappings between these classifications already exist and can be reused (e.g., the coli-conc project, external identifiers in Wikidata, the OpenAlex–ASJC correspondence)? How are they represented, and how do they deal with differing granularity (skos:exactMatch, closeMatch, broadMatch, narrowMatch)?
• Which (semi-)automatic approaches for aligning classifications and taxonomies exist, and what role can Large Language Models play in proposing mapping candidates for the gaps that existing concordances leave?
• Prepare a SKOS concordance that covers all classifications named above – DDC, the OECD Fields of Research and Development, OpenAlex/ASJC, ORKG Research Fields, ACM CCS, and the academic disciplines in Wikidata and DBpedia – as completely as possible. Reuse existing mappings wherever possible, e.g., by mapping every classification to a common hub instead of mapping every pair, and record the relation type and the origin of every mapping (existing concordance, manual, LLM-proposed). Document the coverage achieved for each classification, illustrate the concordance with a few examples and discuss for which kinds of queries it is reliable and where it reaches its limits.

Literature

• Own research
• M. L. Zeng and P. Mayr, “Knowledge Organization Systems (KOS) in the Semantic Web: A multi-dimensional review,” Int. J. Digit. Libr., vol. 20, no. 3, pp. 209–230, Sep. 2019, doi: 10.1007/s00799-018-0241-2.
• U. Balakrishnan, J. Voß, and D. Soergel, “Towards integrated systems for KOS management, mapping, and access: Coli-conc and its collaborative computer-assisted KOS mapping tool Cocoda,” in Proc. 15th Int. ISKO Conf., Porto, Portugal, 2018, pp. 693–701, doi: 10.5771/9783956504211-693.
• J. Neubert, “Wikidata as a linking hub for knowledge organization systems? Integrating an authority mapping into Wikidata and learning lessons for KOS mappings,” in Proc. 17th European Networked Knowledge Organization Systems Workshop (NKOS 2017), Thessaloniki, Greece, 2017, CEUR Workshop Proceedings, vol. 1937, pp. 14–25. [Online]. Available: https://ceur-ws.org/Vol-1937/paper2.pdf
• W. Cui, M. Xiao, L. Wang, X. Wang, Y. Du, and Y. Zhou, “Automated taxonomy alignment via large language models: Bridging the gap between knowledge domains,” Scientometrics, vol. 129, no. 9, pp. 5287–5312, Sep. 2024, doi: 10.1007/s11192-024-05111-2.
• M. Haris, S. Auer, and M. Stocker, “Research information systems and knowledge graphs: A review,” Front. Res. Metr. Anal., vol. 11, Art. no. 1786866, May 2026, doi: 10.3389/frma.2026.1786866.

Research Questions

• Introduction: DBpedia and Wikidata are the two most widely used open knowledge graphs, and both describe academic disciplines such as “computer science” or “sociology”. They do so differently: DBpedia extracts its data from Wikipedia and structures it with the DBpedia ontology and Wikipedia categories, whereas Wikidata is edited collaboratively and builds its hierarchy from statements such as “instance of” (P31) and “subclass of” (P279). Both graphs are connected by equivalence links – owl:equivalentClass or the Wikidata property P1709 at class level, owl:sameAs at instance level.
• How are academic disciplines modelled in DBpedia and in Wikidata? Which classes and relations are used (e.g., dbo:AcademicSubject, Wikipedia categories, P31, P279, P361, and relations such as P2579 “studied by” or P101 “field of work”), and what does the literature report about the structure and quality of both class hierarchies?
• How many DBpedia ontology classes and academic disciplines are linked to Wikidata via owl:equivalentClass, P1709 or owl:sameAs? Are these links correct, and do the hierarchies above them agree?
• Which typical problems occur (e.g., mixing of instances and classes, cycles, very deep subclass chains), and which stopping criteria are needed to traverse the hierarchies in a meaningful way?
• Prepare a manually checked sample of discipline links and a cleaned hierarchy in SKOS that records the origin of every relation (DBpedia, Wikidata, both, or manual), together with a list of incorrect or missing links.

Literature

• Own research
• M. Haris, S. Auer, and M. Stocker, “Research information systems and knowledge graphs: A review,” Front. Res. Metr. Anal., vol. 11, Art. no. 1786866, May 2026, doi: 10.3389/frma.2026.1786866.
• D. Abián, F. Guerra, J. Martínez-Romanos, and R. Trillo-Lado, “Wikidata and DBpedia: A comparative study,” in Proc. 3rd Int. KEYSTONE Conf. (IKC 2017), Gdańsk, Poland, 2017, Lecture Notes in Computer Science, vol. 10546, Cham: Springer, 2018, pp. 142–154, doi: 10.1007/978-3-319-74497-1_14.
• Y. Peng, T. Bonald, and M. Alam, “Refining Wikidata taxonomy using large language models,” in Proc. 33rd ACM Int. Conf. Information and Knowledge Management (CIKM ’24), Boise, ID, USA, 2024, pp. 5395–5399, doi: 10.1145/3627673.3679156.
• A. Ismayilov, D. Kontokostas, S. Auer, J. Lehmann, and S. Hellmann, “Wikidata through the eyes of DBpedia,” Semantic Web, vol. 9, no. 4, pp. 493–503, 2018, doi: 10.3233/SW-170277.
• A. A. Dadalto, J. P. A. Almeida, C. M. Fonseca, and G. Guizzardi, “Evidence of large-scale conceptual disarray in multi-level taxonomies in Wikidata,” Semantic Web, vol. 15, no. 6, pp. 2253–2270, 2024, doi: 10.3233/SW-243562.

Research Questions

• Introduction: Wikidata has become one of the largest open sources of bibliographic metadata. Through the WikiCite initiative, it describes more than 40 million scholarly works together with their authors, venues, citations and main subjects, and tools such as Scholia visualise these data. What sets Wikidata apart from other scholarly knowledge graphs is that publications are part of a general-purpose knowledge graph. Other scholarly knowledge graphs follow different approaches. OpenAlex offers a comprehensive, automatically built index of scholarly works, authors, institutions and topics. The Open Research Knowledge Graph (ORKG) describes the research contributions of papers – such as the research problem, the methods and the results – in a structured way and makes them comparable across papers. These graphs differ considerably in scope, data model and the questions they can answer.
• How does the Wikidata scholarly graph work? How are scholarly works modelled, and how are they connected to other entities (e.g., via author, “cites work” or “main subject”, or as references that support statements about other entities)? How do data enter the graph (e.g., bots, WikiCite tools, manual curation), and what changed with the split of the Wikidata Query Service?
• Which statements about publications can be expressed and queried in Wikidata, and which cannot? Where does Wikidata go beyond bibliographic metadata, e.g., regarding topics, citations, retractions, links to datasets and software, or its role as provenance for facts in other domains?
• How does the Wikidata scholarly graph compare to OpenAlex and the ORKG? Develop criteria for a systematic comparison (e.g., scope and coverage, data model, granularity of statements, data acquisition and curation, data quality, provenance, query interfaces, licences, links to other knowledge graphs) and apply them to all three graphs. For which kinds of questions is each graph the most suitable choice?
• Prepare a set of example queries that illustrate the different kinds of connections in the Wikidata scholarly graph, including federated queries across the main and the scholarly graph, and contrast them with corresponding queries on OpenAlex and the ORKG. Discuss how the three graphs could complement each other.

Literature

• Own research
• T. Lubiana, L. Rasberry, and D. Mietchen, “The Wikidata Query Service split and its impact on the scholarly graph,” in Joint Proc. Posters, Demos, Workshops, and Tutorials of the 21st Int. Conf. Semantic Systems (SEMANTiCS-PDWT 2025), Vienna, Austria, 2025, CEUR Workshop Proceedings, vol. 4064. [Online]. Available: https://ceur-ws.org/Vol-4064/PD-paper3.pdf
• A. Waagmeester et al., “Science Forum: Wikidata as a knowledge graph for the life sciences,” eLife, vol. 9, Art. no. e52614, Mar. 2020, doi: 10.7554/eLife.52614.
• M. Haris, S. Auer, and M. Stocker, “Research information systems and knowledge graphs: A review,” Front. Res. Metr. Anal., vol. 11, Art. no. 1786866, May 2026, doi: 10.3389/frma.2026.1786866.
• F. Å. Nielsen, D. Mietchen, and E. Willighagen, “Scholia, scientometrics and Wikidata,” in The Semantic Web: ESWC 2017 Satellite Events, Portorož, Slovenia, 2017, Lecture Notes in Computer Science, vol. 10577, pp. 237–259, doi: 10.1007/978-3-319-70407-4_36.

Research Questions

• Memory Representation: What types of information should a personal LLM agent retain across sessions, and how should user preferences, past interactions, and task experiences be represented?
• Memory Lifecycle Management: How can an agent manage memory creation, updating, conflict resolution, and forgetting while distinguishing explicit user statements from inferred information?
• Context-Aware Memory Retrieval: How can an agent retrieve memories relevant to its current task while preventing outdated, conflicting, or out-of-scope information from influencing its decisions?
• Memory Evaluation: How can long-term memory systems be evaluated beyond recall accuracy, particularly regarding consistency with current user preferences, responsiveness to corrections, and compliance with forgetting requests?

Literature

• Own research
• Packer, C., Wooders, S., Lin, K., Fang, V., Patil, S. G., Stoica, I., & Gonzalez, J. E. (2023). Memgpt: Towards llms as operating systems. arXiv preprint arXiv:2310.08560.
• Xu, W., Liang, Z., Mei, K., Gao, H., Tan, J., & Zhang, Y. (2026). A-mem: Agentic memory for llm agents. Advances in Neural Information Processing Systems, 38, 17577-17604.
• Wu, D., Wang, H., Yu, W., Zhang, Y., Chang, K. W., & Yu, D. (2024). Longmemeval: Benchmarking chat assistants on long-term interactive memory. arXiv preprint arXiv:2410.10813.

Research Questions

• Intent Understanding and Task Formulation: How can a personal AI agent transform ambiguous user requests into actionable tasks and determine when clarification, advice, tool use, or delegation is appropriate?
• Agent Architecture and Task Coordination: Under what conditions does a coordinator–specialist architecture offer advantages over a single agent with multiple tools in terms of task success, coordination overhead, and permission isolation?
• Personalized Task Planning: How can user goals, preferences, and situational context guide task planning without overriding explicit instructions or exceeding the agent’s authorized scope?
• Proactive Assistance: How can a personal AI agent identify opportunities for assistance from recurring user activities, and determine when to offer a suggestion rather than wait for an explicit request?

Literature

• Own research
• Wu, Q., Bansal, G., Zhang, J., Wu, Y., Li, B., Zhu, E., … & Wang, C. (2023). Autogen: Enabling next-gen llm applications via multi-agent conversation. arXiv preprint arXiv:2308.08155.
• Trivedi, H., Khot, T., Hartmann, M., Manku, R., Dong, V., Li, E., … & Balasubramanian, N. (2024, August). Appworld: A controllable world of apps and people for benchmarking interactive coding agents. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) (pp. 16022-16076).
• Lu, Y., Yang, S., Qian, C., Chen, G., Luo, Q., Wu, Y., … & Sun, M. (2025, May). Proactive agent: Shifting llm agents from reactive responses to active assistance. In International Conference on Learning Representations (Vol. 2025, pp. 47431-47457).

Research Questions

• Action Transparency: What information should an agent present before execution so that users can understand the proposed action, its target resources, its data access requirements, and its potential consequences?
• Authorization and Delegation: How should interfaces distinguish authorization to delegate a task from authorization to modify external resources, particularly in workflows involving multiple agents?
• Monitoring, Intervention, and Recovery: How can interfaces help users monitor progress, revise plans, interrupt execution, and recover from failures while clearly communicating the limits of cancellation and reversibility?
• Interaction Evaluation: How do different confirmation and feedback designs affect user understanding, approval errors, interaction effort, and the alignment between user trust and the agent’s actual capabilities?

Literature

• Own research
• Horvitz, E. (1999, May). Principles of mixed-initiative user interfaces. In Proceedings of the SIGCHI conference on Human Factors in Computing Systems (pp. 159-166).
• Amershi, S., Weld, D., Vorvoreanu, M., Fourney, A., Nushi, B., Collisson, P., … & Horvitz, E. (2019, May). Guidelines for human-AI interaction. In Proceedings of the 2019 chi conference on human factors in computing systems (pp. 1-13).
• Mozannar, H., Bansal, G., Tan, C., Fourney, A., Dibia, V., Chen, J., … & Amershi, S. (2025). Magentic-ui: Towards human-in-the-loop agentic systems. arXiv preprint arXiv:2507.22358.

Research Questions

• What is agentic provenance, and how does it differ from traditional data/workflow provenance?
• What should be recorded during an agent’s execution?
• How can W3C PROV be extended to represent an agent’s execution?
• What is the appropriate granularity of provenance: every token, message, tool call, decision, or workflow step?
• Can an agent execution be reproduced from its provenance trace? How can provenance be used to debug agent failures and hallucinations? What information should not be recorded because of privacy, security, or cost? How can we evaluate the completeness and usefulness of an agent provenance trace?

Literature

• Own research
• Wang et al. (2026), From Agent Traces to Trust: A Survey of Evidence Tracing and Execution Provenance in LLM Agents.
• Souza et al. (2025), PROV-AGENT: Unified Provenance for Tracking AI Agent Interactions in Agentic Workflows.

Research Questions

• Why might an agent need structured knowledge in addition to an LLM?
• How can an agent query a KG during planning and reasoning?
• What is the difference between RAG over documents and KG-based retrieval?
• Can an agent construct or update a KG autonomously?
• How can provenance be attached to KG facts generated by an agent?
• How can an agent detect contradictory KG information?
• Can knowledge graphs reduce hallucinations?
• How should an agent decide whether to trust information retrieved from a KG?

Literature

• Own research
• Hogan et al., Knowledge Graphs, ACM Computing Surveys (2021)
• Pan et al., Unifying Large Language Models and Knowledge Graphs: A Roadmap (2024), IEEE TKDE.

Research Questions

• Can AI agents make computational research more reproducible?
• What makes a computational experiment reproducible?
• What can an AI agent automate in a reproducibility workflow?
• Can an agent automatically detect missing dependencies, data, software versions, or execution instructions?
• How can an agent reproduce a research repository/notebook?
• Can agents diagnose and repair reproducibility failures?
• What information should an agent record to make its own actions reproducible?
• How should we evaluate a ReproAgent?
• Does agent automation actually improve reproducibility, or can it introduce new sources of non-reproducibility?

Literature

• Own research

Research Questions

• What is an ontology, and which steps does a typical ontology engineering methodology contain?
• At which of these steps can an LLM help, and how (competency questions, suggesting classes and properties, proposing axioms, mapping to existing vocabularies)?
• How do current approaches differ in what they let the LLM do and what they leave to the human?
• What kinds of errors do LLM-generated ontologies contain, and how can they be detected?
• How should the quality of an LLM-generated ontology be evaluated: against a gold-standard ontology, with a reasoner, with competency questions, or by domain experts?
• Does LLM assistance actually save effort, or does it move the effort to checking and correcting?
• What should the human expert still decide, and how should these decisions be recorded?

Literature

• Own research
• A. Hogan et al., “Knowledge Graphs,” ACM Comput. Surv., vol. 54, no. 4, pp. 1–37, 2022, doi: 10.1145/3447772.
• H. Babaei Giglou, J. D’Souza, and S. Auer, “LLMs4OL: Large Language Models for Ontology Learning,” in The Semantic Web – ISWC 2023, LNCS vol. 14265, Cham: Springer, 2023, pp. 408–427, doi: 10.1007/978-3-031-47240-4_22.
• M. J. Saeedizade and E. Blomqvist, “Navigating Ontology Development with Large Language Models,” in The Semantic Web – ESWC 2024, LNCS vol. 14664, Cham: Springer, 2024, pp. 143–161, doi: 10.1007/978-3-031-60626-7_8.
• A. Lo, A. Q. Jiang, W. Li, and M. Jamnik, “End-to-End Ontology Learning with Large Language Models,” in Advances in Neural Information Processing Systems 37 (NeurIPS 2024), 2024.

Research Questions

• What is SHACL, and why is validation needed when a knowledge graph already has an ontology?
• Which constraints can SHACL express, and which can it not?
• How does SHACL validation differ from OWL reasoning (closed world vs. open world)?
• How are SHACL shapes created in practice: written by hand, generated from an ontology, or learned from data?
• How is SHACL used for data quality assessment, and which quality dimensions does it cover?
• What are the limits of SHACL for data quality, and what would be needed beyond it?
• How can validation results be presented so that a non-expert can act on them?

Literature

• Own research
• H. Knublauch and D. Kontokostas, Eds., “Shapes Constraint Language (SHACL),” W3C Recommendation, Jul. 20, 2017. [Online]. Available: https://www.w3.org/TR/shacl/
• P. Pareti and G. Konstantinidis, “A Review of SHACL: From Data Validation to Schema Reasoning for RDF Graphs,” in Reasoning Web. Declarative Artificial Intelligence, LNCS vol. 13100, Cham: Springer, 2022, pp. 115–144, doi: 10.1007/978-3-030-95481-9_6.
• K. Rabbani, M. Lissandrini, and K. Hose, “SHACL and ShEx in the Wild: A Community Survey on Validating Shapes Generation and Adoption,” in Companion Proc. Web Conf. 2022 (WWW ’22 Companion), 2022, pp. 260–263, doi: 10.1145/3487553.3524253.
• C. Cortés, L. Ehrlinger, L. Etcheverry, and F. Naumann, “Is SHACL Suitable for Data Quality Assessment?,” in Joint Proc. WOP 2025 and HAIBRIDGE 2025, CEUR Workshop Proceedings, vol. 4093, 2025, pp. 1–17.

Research Questions

• What is SPARQL, and why is writing it hard for non-experts?
• How can an LLM turn a natural-language question into a SPARQL query, and what must it know about the graph (schema, vocabulary, identifiers)?
• How good are current LLMs at writing SPARQL, and how is this measured (benchmarks, metrics)?
• What kinds of errors do generated queries contain (wrong identifiers, wrong structure, invented properties)?
• How can a generated query be checked, corrected or retried before it is executed?
• When is generating a query the better choice, and when is it better to give the LLM graph data directly?
• How can a user see where an answer came from, and does that make the answer more trustworthy?

Literature

• Own research
• S. Harris and A. Seaborne, Eds., “SPARQL 1.1 Query Language,” W3C Recommendation, Mar. 21, 2013. [Online]. Available: https://www.w3.org/TR/sparql11-query/
• D. Banerjee, P. A. Nair, J. N. Kaur, R. Usbeck, and C. Biemann, “Modern Baselines for SPARQL Semantic Parsing,” in Proc. 45th Int. ACM SIGIR Conf. Research and Development in Information Retrieval, 2022, pp. 2260–2265, doi: 10.1145/3477495.3531841.
• L.-P. Meyer, J. Frey, F. Brei, and N. Arndt, “Assessing SPARQL Capabilities of Large Language Models,” in Proc. 3rd Int. Workshop on Natural Language Processing for Knowledge Graph Creation (NLP4KGc 2024), CEUR Workshop Proceedings, vol. 3874, 2024, pp. 35–53.
• S. Pan, L. Luo, Y. Wang, C. Chen, J. Wang, and X. Wu, “Unifying Large Language Models and Knowledge Graphs: A Roadmap,” IEEE Trans. Knowl. Data Eng., vol. 36, no. 7, pp. 3580–3599, 2024, doi: 10.1109/TKDE.2024.3352100.

Research Questions

• Can AI-agents form structures on the web that are similar to societies, in terms of societal hierachies, culture, language and interchange and transformation of ideas and finally transformation of those societies themselves?
• Can this happen spontaneously or is (partial) pre-configuration necessary?
• Which foundational requirements need to be fulfilled in order for such behavior to emerge?
• What examples for such societies are there?
• To what extent do these phenomena constitute genuine emergence through continuous interaction, in contrast to just reproduction of human social patterns from training data or prompt adherence?

Literature

• A. F. Ashery, L. M. Aiello, and A. Baronchelli, “Emergent social conventions and collective bias in LLM populations,” Science Advances, vol. 11, no. 20, Art. no. eadu9368, May 2025, doi: 10.1126/sciadv.adu9368.
• W. Hou and Z. Ji, “Structural divergence between the Moltbook AI-agent network and human social networks,” Advanced Science, September 2026, doi: 10.1002/advs.77665.
• Own research

Research Questions

• How do web applications differ in correctness, security, and maintainability when built through pure vibe coding (code never inspected) versus AI-assisted development with code review?
• Which programming competencies make that review effective, and which have become dispensable?
• For which kinds of web engineering tasks is pure vibe coding sufficient, and where does it break down?

Literature

• A. Fawzy, A. Tahir, and K. Blincoe, “Vibe coding in practice: Motivations, challenges, and a future outlook – a grey literature review,” in Proceedings of the IEEE/ACM 48th International Conference on Software Engineering: Software Engineering in Practice (ICSE-SEIP ’26), 2026, doi: 10.1145/3786583.3786866.
• S. Thorgeirsson, T. B. Weidmann, and Z. Su, “Computer science achievement and writing skills predict vibe coding proficiency,” in Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems (CHI ’26), Barcelona, Spain, April 2026, doi: 10.1145/3772318.3791666.
• Own research

Research Questions

• To what extent do native Web Component standards (Custom Elements, Shadow DOM, HTML Templates) cover the capabilities of modern frameworks such as React, Vue, and Svelte, and where do gaps remain (reactivity, state management, SSR)?
• How interoperable are Web Components across frameworks in practice, and which problems arise (property/event binding, styling across Shadow DOM boundaries, hydration)?
• How do Web Components and framework-based components compare in longevity and maintenance cost, given that frameworks go through breaking changes while web standards remain stable?

Literature

• J. Vepsäläinen, A. Hellas, and P. Vuorimaa, “The rise of disappearing frameworks in web development,” in Web Engineering: 23rd International Conference, ICWE 2023, Alicante, Spain, Lecture Notes in Computer Science, vol. 13893. Cham, Switzerland: Springer, 2023, pp. 319–326, doi: 10.1007/978-3-031-34444-2_23.
• J. Vepsäläinen, M. Hevery, and P. Vuorimaa, “Resumability: A new primitive for developing web applications,” IEEE Access, vol. 12, pp. 9038–9046, 2024.
• Own research

Research Questions

• How does a Systemization of Knowledge work? Prepare a guideline for computer science students explaining the main aspects.
• How does it compare to a Systematic Literature Review? How does it compare to a Systematic Mapping Study? When to apply which one?
• What are risks and limitations of the method?
• Which SoK artifacts should be provided to allow for reproducibility and replicability?
• What tools exist to support SoKs? Demonstrate a suitable tool.

Literature

• https://www.jsys.org/type_SoK/
• https://oaklandsok.github.io/
• https://dl.acm.org/doi/pdf/10.1145/3613904.3642315
• Kitchenham, B. (2004). Procedures for Undertaking Systematic Reviews. https://www.inf.ufsc.br/~aldo.vw/kitchenham.pdf
• Kitchenham, B., Pearl Brereton, O., Budgen, D., Turner, M., Bailey, J., & Linkman, S. (2009). Systematic literature reviews in software engineering – A systematic literature review. Information and Software Technology, 51(1), 7–15.
• Brereton, P., Kitchenham, B. a., Budgen, D., Turner, M., & Khalil, M. (2007). Lessons from applying the systematic literature review process within the software engineering domain. Journal of Systems and Software, 80(4), 571–583.
• Petersen, K., Vakkalanka, S., & Kuzniarz, L. (2015). Guidelines for conducting systematic mapping studies in software engineering: An update. Information and Software Technology, 64, 1–18.
• Díaz, O., Medina, H., & Anfurrutia, F. I. (2019). Coding-Data Portability in Systematic Literature Reviews. Proceedings of the Evaluation and Assessment on Software Engineering – EASE ’19, 178–187.
• Khadka, R., Saeidi, A. M., Idu, A., Hage, J., & Jansen, S. (2013). Legacy to SOA Evolution: A Systematic Literature Review. In A. D. Ionita, M. Litoiu, & G. Lewis (Eds.), Migrating Legacy Applications: Challenges in Service Oriented Architecture and Cloud Computing Environments (pp. 40–71). IGI Global.
• Jamshidi, P., Ahmad, A., & Pahl, C. (2013). Cloud Migration Research: A Systematic Review. IEEE Transactions on Cloud Computing, 1(2), 142–157.
• Rai, R., Sahoo, G., & Mehfuz, S. (2015). Exploring the factors influencing the cloud computing adoption: a systematic study on cloud migration. SpringerPlus, 4(1), 197.
• A. Hinderks, F. José, D. Mayo, J. Thomaschewski and M. J. Escalona, “An SLR-Tool: Search Process in Practice : A tool to conduct and manage Systematic Literature Review (SLR),” 2020 IEEE/ACM 42nd International Conference on Software Engineering: Companion Proceedings (ICSE-Companion), 2020, pp. 81-84.
• PRISMA 2020 http://www.prisma-statement.org/
• Own reasearch

Research Questions

• How is a scientific work, especially a thesis in computer science, structured? What sections should a thesis contain and what purpose do they have? Give an overview.
• What is the importance of an introduction? What should it contain? How long should it be? How is it related to the other sections in a scientific work?
• What makes a “good” motivation for your scientific work? Why is it important for the readers? What are methods to write it so the reader can relate to the writer?
• What are current and well known best practices/guidelines/schemes/principles/advices etc.? Which evidence base (e.g. experimental studies) are supporting them? Present them.
• Choose a sufficient scientific work and work out a way to visually represent its whole structure. Show how the introduction, motivation and scope relate to the other parts.

Literature

• Peat, J., Elliott, E., Baur, L., & Keena, V. (2013). Scientific writing: easy when you know how. John Wiley & Sons. DOI:10.1002/9781118708019
• X Barbara Minto: The Pyramid Principle. Pearson Education, 2009.
• Mensh, B., & Kording, K. (2017). Ten simple rules for structuring papers. PLoS computational biology, 13(9), e1005619. DOI: https://doi.org/10.1371/journal.pcbi.1005619
• J. M. Setchell, “Writing a Scientific Report,” in Studying Primates: How to Design, Conduct and Report Primatological Research, Cambridge: Cambridge University Press, 2019, pp. 271–298.
• Blackwell, J., & Martin, J. (2011). A scientific approach to scientific writing. Springer Science & Business Media.
• Williams, J. M., & Bizup, J. (2014). Lessons in clarity and grace. Pearson.
• Oguduvwe, J. I. P. (2013). Nature, Scope and Role of Research Proposal in Scientific Investigations. IOSR Journal Of Humanities And Social Science (IOSR-JHSS), 17(2), 83-87. https://www.iosrjournals.org/iosr-jhss/papers/Vol17-issue2/L01728387.pdf
• Own research

Research Questions

• What guidelines and principles exist to safeguard Good Research Practice?
• How can these guidelines and principles be integrated into the research process?
• What is scientific misconduct/scientific malpractice?
• What is considered “high-quality research”? What are indicators thereof?

Literature

• The European Code of Conduct for Research Integrity http://www.allea.org/wp-content/uploads/2017/03/ALLEA-European-Code-of-Conduct-for-Research-Integrity-2017-1.pdf
• Guidelines for Safeguarding Good Research Practice https://www.dfg.de/download/pdf/foerderung/rechtliche_rahmenbedingungen/gute_wissenschaftliche_praxis/kodex_gwp_en.pdf
• Open Research Data and Data Management Plans https://erc.europa.eu/sites/default/files/document/file/ERC_info_document-Open_Research_Data_and_Data_Management_Plans.pdf
• Own research

Research Questions

• What are guardrails for AI agents?
• What are the limitations of agent guardrails? What can be enforced with guardrails and what cannot?
• How can guardrails be implemented in AI agents? Identify and compare different approaches and technologies for guardrails in AI agents. Present your findings.

Literature

• S. A. Akheel, “Guardrails for large language models: A review of techniques and challenges,” J Artif Intell Mach Learn & Data Sci, vol. 3, no. 1, pp. 2504–2512, 2025.
• Own research

Research Questions

• Introduction: Research outputs are classified by subject in many different ways. Libraries and repositories use the Dewey Decimal Classification (DDC), OpenAlex assigns domains, fields and subfields derived from the Scopus ASJC scheme, the Open Research Knowledge Graph (ORKG) maintains its own taxonomy of research fields, and Wikidata and DBpedia describe academic disciplines as part of their knowledge graphs. When a knowledge graph integrates research outputs from several sources, the same output therefore carries subject classes from different systems. A concordance between these classifications – a set of mappings expressed with SKOS mapping relations – makes it possible to find research outputs by field regardless of the classification the source delivered. The central difficulty is granularity: a DDC class can be broader or narrower than an ORKG research field, and a Wikidata discipline can correspond to either.
• Which subject classifications are used for research outputs (e.g., DDC, the OECD Fields of Research and Development, OpenAlex/ASJC, ORKG Research Fields, ACM CCS, Wikidata and DBpedia disciplines)? How do they differ in purpose, structure, depth, granularity and licensing?
• Which concordances and mappings between these classifications already exist and can be reused (e.g., the coli-conc project, external identifiers in Wikidata, the OpenAlex–ASJC correspondence)? How are they represented, and how do they deal with differing granularity (skos:exactMatch, closeMatch, broadMatch, narrowMatch)?
• Which (semi-)automatic approaches for aligning classifications and taxonomies exist, and what role can Large Language Models play in proposing mapping candidates for the gaps that existing concordances leave?
• Prepare a SKOS concordance that covers all classifications named above – DDC, the OECD Fields of Research and Development, OpenAlex/ASJC, ORKG Research Fields, ACM CCS, and the academic disciplines in Wikidata and DBpedia – as completely as possible. Reuse existing mappings wherever possible, e.g., by mapping every classification to a common hub instead of mapping every pair, and record the relation type and the origin of every mapping (existing concordance, manual, LLM-proposed). Document the coverage achieved for each classification, illustrate the concordance with a few examples and discuss for which kinds of queries it is reliable and where it reaches its limits.

Literature

• Own research
• M. L. Zeng and P. Mayr, “Knowledge Organization Systems (KOS) in the Semantic Web: A multi-dimensional review,” Int. J. Digit. Libr., vol. 20, no. 3, pp. 209–230, Sep. 2019, doi: 10.1007/s00799-018-0241-2.
• U. Balakrishnan, J. Voß, and D. Soergel, “Towards integrated systems for KOS management, mapping, and access: Coli-conc and its collaborative computer-assisted KOS mapping tool Cocoda,” in Proc. 15th Int. ISKO Conf., Porto, Portugal, 2018, pp. 693–701, doi: 10.5771/9783956504211-693.
• J. Neubert, “Wikidata as a linking hub for knowledge organization systems? Integrating an authority mapping into Wikidata and learning lessons for KOS mappings,” in Proc. 17th European Networked Knowledge Organization Systems Workshop (NKOS 2017), Thessaloniki, Greece, 2017, CEUR Workshop Proceedings, vol. 1937, pp. 14–25. [Online]. Available: https://ceur-ws.org/Vol-1937/paper2.pdf
• W. Cui, M. Xiao, L. Wang, X. Wang, Y. Du, and Y. Zhou, “Automated taxonomy alignment via large language models: Bridging the gap between knowledge domains,” Scientometrics, vol. 129, no. 9, pp. 5287–5312, Sep. 2024, doi: 10.1007/s11192-024-05111-2.
• M. Haris, S. Auer, and M. Stocker, “Research information systems and knowledge graphs: A review,” Front. Res. Metr. Anal., vol. 11, Art. no. 1786866, May 2026, doi: 10.3389/frma.2026.1786866.

Research Questions

• Introduction: DBpedia and Wikidata are the two most widely used open knowledge graphs, and both describe academic disciplines such as “computer science” or “sociology”. They do so differently: DBpedia extracts its data from Wikipedia and structures it with the DBpedia ontology and Wikipedia categories, whereas Wikidata is edited collaboratively and builds its hierarchy from statements such as “instance of” (P31) and “subclass of” (P279). Both graphs are connected by equivalence links – owl:equivalentClass or the Wikidata property P1709 at class level, owl:sameAs at instance level.
• How are academic disciplines modelled in DBpedia and in Wikidata? Which classes and relations are used (e.g., dbo:AcademicSubject, Wikipedia categories, P31, P279, P361, and relations such as P2579 “studied by” or P101 “field of work”), and what does the literature report about the structure and quality of both class hierarchies?
• How many DBpedia ontology classes and academic disciplines are linked to Wikidata via owl:equivalentClass, P1709 or owl:sameAs? Are these links correct, and do the hierarchies above them agree?
• Which typical problems occur (e.g., mixing of instances and classes, cycles, very deep subclass chains), and which stopping criteria are needed to traverse the hierarchies in a meaningful way?
• Prepare a manually checked sample of discipline links and a cleaned hierarchy in SKOS that records the origin of every relation (DBpedia, Wikidata, both, or manual), together with a list of incorrect or missing links.

Literature

• Own research
• M. Haris, S. Auer, and M. Stocker, “Research information systems and knowledge graphs: A review,” Front. Res. Metr. Anal., vol. 11, Art. no. 1786866, May 2026, doi: 10.3389/frma.2026.1786866.
• D. Abián, F. Guerra, J. Martínez-Romanos, and R. Trillo-Lado, “Wikidata and DBpedia: A comparative study,” in Proc. 3rd Int. KEYSTONE Conf. (IKC 2017), Gdańsk, Poland, 2017, Lecture Notes in Computer Science, vol. 10546, Cham: Springer, 2018, pp. 142–154, doi: 10.1007/978-3-319-74497-1_14.
• Y. Peng, T. Bonald, and M. Alam, “Refining Wikidata taxonomy using large language models,” in Proc. 33rd ACM Int. Conf. Information and Knowledge Management (CIKM ’24), Boise, ID, USA, 2024, pp. 5395–5399, doi: 10.1145/3627673.3679156.
• A. Ismayilov, D. Kontokostas, S. Auer, J. Lehmann, and S. Hellmann, “Wikidata through the eyes of DBpedia,” Semantic Web, vol. 9, no. 4, pp. 493–503, 2018, doi: 10.3233/SW-170277.
• A. A. Dadalto, J. P. A. Almeida, C. M. Fonseca, and G. Guizzardi, “Evidence of large-scale conceptual disarray in multi-level taxonomies in Wikidata,” Semantic Web, vol. 15, no. 6, pp. 2253–2270, 2024, doi: 10.3233/SW-243562.

Research Questions

• Which datasets are available for chatbot research / design on public platforms aside from conversational training? Analyse existing chatbot related datasets to answer this question.
• Document every step in a research journal (i.e. what did you work on, what is the result, etc.) and reflect on the current landscape of chatbot datasets for research / design.
• Define inclusion (i.e. English language) and exclusion criteria (i.e. language models, special pre-trained models).
• Create search queries and use them on different platforms (some possible platforms are listed below).
• Collect a meaningful number of datasets and filter them according to your inclusion/exclusion criteria. The goal is to have at least 50 diverse datasets.
• Extract relevant metadata (size, file types, keywords, upload date, license, publications platform, etc.), document it, analyze this information with statistical data and present your analysis results with well-chosen graphs.

Literature

• Own research
• Agarwal S, Agarwal B, Gupta R (2022), Chatbots and virtual assistants: a bibliometric analysis. Library Hi Tech, Vol. 40 No. 4 pp. 1013–1030. https://doi.org/10.1108/LHT-09-2021-0330
• Khan AM, Loan FA, Parray UY, Rashid S (2024), Global overview of research data repositories: an analysis of re3data registry. Information Discovery and Delivery, Vol. 52 No. 1 pp. 53–61. https://doi.org/10.1108/IDD-07-2022-0069
• Marcial, L.H. and Hemminger, B.M. (2010), Scientific data repositories on the Web: An initial survey. J. Am. Soc. Inf. Sci., 61: 2029-2048. https://doi.org/10.1002/asi.21339
• https://ieee-dataport.org/datasets
• https://www.nature.com/search?journal=sdata&q=chatbot
• https://figshare.com/
• https://data.mendeley.com/
• https://dataverse.harvard.edu/
• https://osf.io/
• https://datasetsearch.research.google.com/

Research Questions

• What is the difference between qualitative and quantitative research methodologies?
• Provide an overview of different methodologies, explain how they work, when and in which field they are applied.
• Your use case is the use of data for chatbot research, so which kind of data is used and how it is used.
• Qualitative: Conduct a survey with researchers and professionals. Research on how to formulate survey questions and send them out to experts.
• Quantitative: Analyse existing datasets. Create search queries for databases and analyse the found datasets on specific aspects.
• Describe your methodologies in detail in a research log and note down any meetings, decisions, changes, and outcomes.
• Reflect on the differences in using both methodologies and provide insights into new learnings, pain points, and how you can apply the different methodologies you found in other areas.

Literature

• Own research
• Creswell, J. W. (2014). Research design : Qualitative, quantitative, and mixed methods approaches. (4. ed., in). SAGE. https://katalog.bibliothek.tu-chemnitz.de/Record/0008891954
  Currently only available in printed version
• Trochim, W., Donnelly, J. P., & Arora, K. (2016). Research methods: The essential knowledge base.
• Savin-Baden, M., & Major, C H. (2025). Qualitative research: The essential guide to theory and practice (Second edition.). Routledge. https://katalog.bibliothek.tu-chemnitz.de/Record/0-1939098580 (esp. Chapter 23 – Interviews)
  Alternative access in WebOPAC
• Developing and Validating a Comprehensive Literature Search Strategy https://library.cumc.columbia.edu/kb/developing-comprehensive-search-query

Research Questions

• Which tools are available for assessing FAIRness, and which digital objects do they support (e.g., datasets, software, metadata, repositories)?
• Which FAIR principles and individual FAIR indicators does each tool assess? Which indicators are missing or only partially assessed?
• How do the tools differ in their use of automated tests, manual questionnaires, metadata inspection, and external services?
• When multiple tools assess the same digital object, how similar are their results? What explains differences in their scores?
• What aspects of FAIRness cannot be reliably assessed automatically? Do the tools evaluate actual reusability or primarily technical metadata and accessibility?
• What are the most important gaps in current tools, and how could future FAIR assessment approaches address them?

Literature

• Krans et al. (2022), FAIR assessment tools: evaluating use and performance
• An automated solution for measuring the progress toward FAIR research data: 10.1016/j.patter.2021.100370
• Wilkinson et al. (2016), The FAIR Guiding Principles for scientific data management and stewardship: https://doi.org/10.1038/sdata.2016.18
  

Research Questions

• To what extent do selected ontologies satisfy the FAIR principles according to FOOPS!?
• How does FAIRness vary across ontology domains, sizes, maturity levels, or publication repositories?
• Which FAIR principles or indicators are not adequately assessed by FOOPS!, and what complementary tools or methods could address these gaps?
• What are the most common FAIRness shortcomings, and which improvements could ontology developers make to address them?

Literature

• Garijo, D., Corcho, O., & Poveda-Villalón, M. (2021). FOOPS!: An Ontology Pitfall Scanner for the FAIR Principles.
• Amdouni, E., Bouazzouni, S., & Jonquet, C. (2022). O’FAIRe makes you an offer: Metadata-based Automatic FAIRness Assessment for Ontologies and Semantic Resources.
• Wilkinson et al. (2016), The FAIR Guiding Principles for scientific data management and stewardship: https://doi.org/10.1038/sdata.2016.18

Research Questions

• Can AI-agents form structures on the web that are similar to societies, in terms of societal hierachies, culture, language and interchange and transformation of ideas and finally transformation of those societies themselves?
• Can this happen spontaneously or is (partial) pre-configuration necessary?
• Which foundational requirements need to be fulfilled in order for such behavior to emerge?
• What examples for such societies are there?
• To what extent do these phenomena constitute genuine emergence through continuous interaction, in contrast to just reproduction of human social patterns from training data or prompt adherence?

Literature

• A. F. Ashery, L. M. Aiello, and A. Baronchelli, “Emergent social conventions and collective bias in LLM populations,” Science Advances, vol. 11, no. 20, Art. no. eadu9368, May 2025, doi: 10.1126/sciadv.adu9368.
• W. Hou and Z. Ji, “Structural divergence between the Moltbook AI-agent network and human social networks,” Advanced Science, September 2026, doi: 10.1002/advs.77665.
• Own research

Research Questions

• How do web applications differ in correctness, security, and maintainability when built through pure vibe coding (code never inspected) versus AI-assisted development with code review?
• Which programming competencies make that review effective, and which have become dispensable?
• For which kinds of web engineering tasks is pure vibe coding sufficient, and where does it break down?

Literature

• A. Fawzy, A. Tahir, and K. Blincoe, “Vibe coding in practice: Motivations, challenges, and a future outlook – a grey literature review,” in Proceedings of the IEEE/ACM 48th International Conference on Software Engineering: Software Engineering in Practice (ICSE-SEIP ’26), 2026, doi: 10.1145/3786583.3786866.
• S. Thorgeirsson, T. B. Weidmann, and Z. Su, “Computer science achievement and writing skills predict vibe coding proficiency,” in Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems (CHI ’26), Barcelona, Spain, April 2026, doi: 10.1145/3772318.3791666.
• Own research

Research Questions

• Memory Representation: What types of information should a personal LLM agent retain across sessions, and how should user preferences, past interactions, and task experiences be represented?
• Memory Lifecycle Management: How can an agent manage memory creation, updating, conflict resolution, and forgetting while distinguishing explicit user statements from inferred information?
• Context-Aware Memory Retrieval: How can an agent retrieve memories relevant to its current task while preventing outdated, conflicting, or out-of-scope information from influencing its decisions?
• Memory Evaluation: How can long-term memory systems be evaluated beyond recall accuracy, particularly regarding consistency with current user preferences, responsiveness to corrections, and compliance with forgetting requests?

Literature

• Own research
• Packer, C., Wooders, S., Lin, K., Fang, V., Patil, S. G., Stoica, I., & Gonzalez, J. E. (2023). Memgpt: Towards llms as operating systems. arXiv preprint arXiv:2310.08560.
• Xu, W., Liang, Z., Mei, K., Gao, H., Tan, J., & Zhang, Y. (2026). A-mem: Agentic memory for llm agents. Advances in Neural Information Processing Systems, 38, 17577-17604.
• Wu, D., Wang, H., Yu, W., Zhang, Y., Chang, K. W., & Yu, D. (2024). Longmemeval: Benchmarking chat assistants on long-term interactive memory. arXiv preprint arXiv:2410.10813.

Research Questions

• Intent Understanding and Task Formulation: How can a personal AI agent transform ambiguous user requests into actionable tasks and determine when clarification, advice, tool use, or delegation is appropriate?
• Agent Architecture and Task Coordination: Under what conditions does a coordinator–specialist architecture offer advantages over a single agent with multiple tools in terms of task success, coordination overhead, and permission isolation?
• Personalized Task Planning: How can user goals, preferences, and situational context guide task planning without overriding explicit instructions or exceeding the agent’s authorized scope?
• Proactive Assistance: How can a personal AI agent identify opportunities for assistance from recurring user activities, and determine when to offer a suggestion rather than wait for an explicit request?

Literature

• Own research
• Wu, Q., Bansal, G., Zhang, J., Wu, Y., Li, B., Zhu, E., … & Wang, C. (2023). Autogen: Enabling next-gen llm applications via multi-agent conversation. arXiv preprint arXiv:2308.08155.
• Trivedi, H., Khot, T., Hartmann, M., Manku, R., Dong, V., Li, E., … & Balasubramanian, N. (2024, August). Appworld: A controllable world of apps and people for benchmarking interactive coding agents. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) (pp. 16022-16076).
• Lu, Y., Yang, S., Qian, C., Chen, G., Luo, Q., Wu, Y., … & Sun, M. (2025, May). Proactive agent: Shifting llm agents from reactive responses to active assistance. In International Conference on Learning Representations (Vol. 2025, pp. 47431-47457).

Research Questions

• What is Trustworthy AI? Why is Trustworthy AI important in system architecture?
• Which trustworthiness aspects can be represented in architecture models, such as fairness, explainability, privacy, or robustness?
• How can modeling languages such as UML, SysML, or WAM help describe Trustworthy AI systems?
• Show a small example of a system architecture model with Trustworthy AI aspects
• Literature

Literature

• Own research

Research Questions

• What is Green AI, and how is it different from traditional AI system design?
• How can system architecture models represent sustainability aspects such as energy efficiency, carbon footprint, and resource consumption?
• How can architectural modeling languages such as UML, SysML, or WAM be extended to include Green AI concerns?
• Show a small demo or conceptual model of a system architecture that includes Green AI metrics
• Literature

Literature

• Stephan Sehestedt, Chih-Hong Cheng, and Eric Bouwers. 2014. Towards quantitative metrics for architecture models. In Proceedings of the WICSA 2014 Companion Volume (WICSA ’14 Companion). Association for Computing Machinery, New York, NY, USA, Article 5, 1–4. https://doi.org/10.1145/2578128.2578226
• Verónica Bolón-Canedo, Laura Morán-Fernández, Brais Cancela, Amparo Alonso-Betanzos,A review of green artificial intelligence: Towards a more sustainable future, Neurocomputing, Volume 599, 2024, 128096, ISSN 0925-2312, https://doi.org/10.1016/j.neucom.2024.128096.
• Own research

Research Questions

• What is an ontology, and which steps does a typical ontology engineering methodology contain?
• At which of these steps can an LLM help, and how (competency questions, suggesting classes and properties, proposing axioms, mapping to existing vocabularies)?
• How do current approaches differ in what they let the LLM do and what they leave to the human?
• What kinds of errors do LLM-generated ontologies contain, and how can they be detected?
• How should the quality of an LLM-generated ontology be evaluated: against a gold-standard ontology, with a reasoner, with competency questions, or by domain experts?
• Does LLM assistance actually save effort, or does it move the effort to checking and correcting?
• What should the human expert still decide, and how should these decisions be recorded?

Literature

• Own research
• A. Hogan et al., “Knowledge Graphs,” ACM Comput. Surv., vol. 54, no. 4, pp. 1–37, 2022, doi: 10.1145/3447772.
• H. Babaei Giglou, J. D’Souza, and S. Auer, “LLMs4OL: Large Language Models for Ontology Learning,” in The Semantic Web – ISWC 2023, LNCS vol. 14265, Cham: Springer, 2023, pp. 408–427, doi: 10.1007/978-3-031-47240-4_22.
• M. J. Saeedizade and E. Blomqvist, “Navigating Ontology Development with Large Language Models,” in The Semantic Web – ESWC 2024, LNCS vol. 14664, Cham: Springer, 2024, pp. 143–161, doi: 10.1007/978-3-031-60626-7_8.
• A. Lo, A. Q. Jiang, W. Li, and M. Jamnik, “End-to-End Ontology Learning with Large Language Models,” in Advances in Neural Information Processing Systems 37 (NeurIPS 2024), 2024.

Research Questions

• What is SHACL, and why is validation needed when a knowledge graph already has an ontology?
• Which constraints can SHACL express, and which can it not?
• How does SHACL validation differ from OWL reasoning (closed world vs. open world)?
• How are SHACL shapes created in practice: written by hand, generated from an ontology, or learned from data?
• How is SHACL used for data quality assessment, and which quality dimensions does it cover?
• What are the limits of SHACL for data quality, and what would be needed beyond it?
• How can validation results be presented so that a non-expert can act on them?

Literature

• Own research
• H. Knublauch and D. Kontokostas, Eds., “Shapes Constraint Language (SHACL),” W3C Recommendation, Jul. 20, 2017. [Online]. Available: https://www.w3.org/TR/shacl/
• P. Pareti and G. Konstantinidis, “A Review of SHACL: From Data Validation to Schema Reasoning for RDF Graphs,” in Reasoning Web. Declarative Artificial Intelligence, LNCS vol. 13100, Cham: Springer, 2022, pp. 115–144, doi: 10.1007/978-3-030-95481-9_6.
• K. Rabbani, M. Lissandrini, and K. Hose, “SHACL and ShEx in the Wild: A Community Survey on Validating Shapes Generation and Adoption,” in Companion Proc. Web Conf. 2022 (WWW ’22 Companion), 2022, pp. 260–263, doi: 10.1145/3487553.3524253.
• C. Cortés, L. Ehrlinger, L. Etcheverry, and F. Naumann, “Is SHACL Suitable for Data Quality Assessment?,” in Joint Proc. WOP 2025 and HAIBRIDGE 2025, CEUR Workshop Proceedings, vol. 4093, 2025, pp. 1–17.

Research Questions

• What techniques exist to sandbox autonomous AI agents?
• Which problems are solved through sandboxing, and which are not?
• What overhead and limitations do current sandboxing techniques have?

Literature

• L. Meng, H. Feng, I. Shumailov, and E. Fernandes, “ceLLMate: Sandboxing Browser AI Agents.” Mar 2026. doi: 10.48550/arXiv.2512.12594
• Own research