David M. Schmidt

I am incredibly happy to share that our paper "CompoST: A Benchmark for Analyzing the Ability of LLMs To Compositionally Interpret Questions in a QALD Setting" has been accepted as a research track paper at ISWC @iswc-conf.bsky.social! Stay tuned for the paper and see you all in Japan!

- and listened to keynotes of well-known researchers like Frank van Harmelen, Natasha Noy and Enrico Motta
A huge thanks to everyone who made this week such a memorable experience! And, if you are Master's/PhD student or PostDoc, I cannot recommend too much to apply for the next iteration of #ISWS!

- worked in a research task force on building a reliable LLM-based metadata enrichment pipeline for cultural heritage objects (special thanks to our tutor Valentina Presutti and our whole team), as well as writing a corresponding white paper and presenting our results in the final session

During the last week, among many other things, I
- summarized the motivation of my work in a 45s "Minute Madness" session
- presented my work during a poster session, getting helpful feedback from students and tutors (special thanks to Aidan Hogan and Stefano De Giorgis)

The #ISWS2025 experience really managed to combine lots of fun activities, working with leading figures of the Semantic Web field as well as intense networking in a unique, wonderful way! It felt like a month worth of program items had been compressed to one magnificent piece of art.

International Semantic Web Research Summer School 2025 group photo

David M. Schmidt presenting the group's work on AI-based cultural heritage metadata enrichment

David M. Schmidt presenting his poster on NeoDUDES, a compositional Question Answering system using DUDES

David M. Schmidt in front of the beautiful landscape of Bertinoro

What a week! I just had the incredible opportunity to attend the International Semantic Web Research Summer School 2025 @isws-summerschool.bsky.social. I hoped for an intense week filled with inspiring keynotes, people and opportunities to present my work - and I got so much more than "just" that!

Social media is acknowledged as an important source of patient experience data to learn about patients’ unmet needs, priorities, and preferences. The objective of this study was to evaluate to what extent SOTA LLMs can appropriately summarize posts shared by patients in web-based forums.

Summarizing Online Patient Conversations Using Generative Language Models: Experimental and Comparative Study Background: Social media is acknowledged by regulatory bodies (eg, the Food and Drug Administration) as an important source of patient experience data to learn about patients’ unmet needs, priorities,...

🎓 Authors: Rakhi Asokkumar Subjagouri Nair, Matthias Hartung, Philipp Heinisch, Janik Jaskolski, Cornelius Starke-Knäusel, Susana Veríssimo, David M. Schmidt, Philipp Cimiano

🔗 Paper: doi.org/10.2196/62909

Summarizing Online Patient Conversations Using Generative Language Models: Experimental and Comparative Study Background: Social media is acknowledged by regulatory bodies (eg, the Food and Drug Administration) as an important source of patient experience data to learn about patients’ unmet needs, priorities,...

🚀 New paper! 🚀

I am happy to announce our paper "Summarizing Online Patient Conversations Using Generative Language Models: Experimental and Comparative Study," which has just been published in JMIR Medical Informatics!

Research Position - Text Generation/Natural Langua... <div style="text-align: justify;">The Faculty of Engineering at Bielefeld University is looking for a research assistant to work on th...

NLP/Text Generation
EN: uni-bielefeld.hr4you.org/job/view/4054
DE: uni-bielefeld.hr4you.org/job/view/4053

NLP/Information Extraction
EN: uni-bielefeld.hr4you.org/job/view/4059
DE: uni-bielefeld.hr4you.org/job/view/4057

If you have any questions, do not hesitate to contact me or Philipp directly!

We currently have two fully-funded open PhD positions in our group with a focus on #NLProc, #InformationExtraction and #TextGeneration. I can really recommend both the group as well as Philipp Cimiano as a supervisor, so take this opportunity!

🚀 We are #hiring! Are you interested in Natural Language Processing, Text Generation or Information Extraction and want to pursue a PhD? Then you now have the chance to become a part of the Semantic Computing Group at Bielefeld University!

Application Deadline: 20.03.2025

Spring School 2025: Innovating AI Evaluation – Beyond Accuracy and Precision - SAIL Join us for the SAIL Spring School 2025! We are excited to invite you to the SAIL Spring School 2025, taking place March 26-28 at the CITEC lecture hall at Bielefeld University! We have put together a...

📣 Spring School 2025: Innovating AI Evaluation – Beyond Accuracy and Precision

📅 March 26–28
📍 CITEC, Bielefeld University

Join us for an exciting line-up of tutorials, discussions, and networking opportunities! 🎓

➡️More info & program: www.sail.nrw/springschool/

💡 Interested? Try it yourself!

Tool: ag-sc.techfak.uni-bielefeld.de/ctvis/

For selecting clinical trials to be compared in systematic reviews, it is important they measure the same outcomes. Therefore, we developed a tool that provides an overview of the clinical trial information about glaucoma and type 2 diabetes and enables users to group them by outcomes.

Open challenges for the automatic synthesis of clinical trials - BMC Research Notes Objective An important criterion for selecting clinical trials to be compared in systematic reviews and meta-analyses is that they measure the same outcomes. However, this represents a challenge as th...

🚀 New month, new paper! 🚀

Our paper "Open challenges for the automatic synthesis of clinical trials" has been published at BMC Research Notes!

🎓 Authors: Olivia Sánchez Graillet, David M. Schmidt, Christian Kullik and Philipp Cimiano

🔗 Paper: doi.org/10.1186/s131...

💡 Interested? Try it yourself!

Zenodo artifact: doi.org/10.5281/zeno...

GitHub repository: github.com/ag-sc/clinic...

In this work, we investigate the influence of grammar-constrained decoding (GCD) as well as pointer generators (PG) on the performance of a domain-specific information extraction (IE) system. We investigate whether the addition of GCD and PG improve IE results of fine-tuned encoder-decoder models.

Illustration of the baseline model as well as the two adjustments added to that baseline, grammar-constrained decoding and pointer generator-like behavior. Words in boxes represent single tokens, numbers below those boxes symbolize outputs from the decoder, where higher values stand for a higher probability that this is the best next token as estimated by the model. For greedy decoding, the token with the highest value is chosen. For GCD, a filter is applied before, visualized as gray, crossed-out boxes for tokens that are filtered out. Red boxes show the selected token. (A) Greedy decoding (baseline, basic). (B) Grammar-constrained decoding (GCD). (C) Pointer generators + grammar-constrained decoding (ptr).

🚀 New year, new paper! 🚀

Proud to share our paper "Grammar-constrained decoding for structured information extraction with fine-tuned generative models applied to clinical trial abstracts" has been published at Frontiers in Artificial Intelligence!

🔗 doi.org/10.3389/frai...

🧑‍💻 Additionally, you can find the code and data if our approach on Zenodo, GitHub and DockerHub:
Zenodo artifact: doi.org/10.5281/zeno...
GitHub repository: github.com/ag-sc/neodud...
DockerHub image: hub.docker.com/r/dvs23/neod...

Lexicalization Is All You Need: Examining the Impact of Lexical Knowledge in a Compositional QALD System In this paper, we examine the impact of lexicalization on Question Answering over Linked Data (QALD). It is well known that one of the key challenges in interpreting natural language questions with re...

💡Missed the talk or want to know more? You can find our paper here:
doi.org/10.1007/978-...
Preprint: doi.org/10.48550/arX...

At the main conference, I presented our paper "Lexicalization Is All You Need: Examining the Impact of Lexical Knowledge in a Compositional QALD System" as well as an accompanying poster and demo illustrating the strengths of our lexicon-based, compositional question answering approach.

David M. Schmidt giving a talk on the paper "Lexicalization Is All You Need: Examining the Impact of Lexical Knowledge in a Compositional QALD System" at the 24th International Conference on Knowledge Engineering and Knowledge Management in Amsterdam.

David M. Schmidt presenting a poster and a demo on "Lexicalization Is All You Need: Examining the Impact of Lexical Knowledge in a Compositional QALD System" at the 24th International Conference on Knowledge Engineering and Knowledge Management in Amsterdam.

It has been an exciting week at EKAW 2024 in Amsterdam! Lots of interesting talks, inspiring discussions and entertaining social events! #ekaw2024

Time for a starter pack on information retrieval: go.bsky.app/MXPJoTn

A starter pack for #NLP #NLProc researchers! 🎉

go.bsky.app/SngwGeS

Hey all! I started a second starter pack with people who didn't make the first one, please let me know if you'd like to be added:

go.bsky.app/JgneRQk

💬 Additionally to the paper presentation at the EKAW - International Conference on Knowledge Engineering and Knowledge Management, we will also take part in the poster session. So drop by if you want to discuss future avenues of question answering research!

💡 Interested? Try it yourself!

Zenodo artifact: doi.org/10.5281/zeno...
GitHub repository: github.com/ag-sc/neodud...
DockerHub image: hub.docker.com/r/dvs23/neod...

Lexicalization Is All You Need: Examining the Impact of Lexical Knowledge in a Compositional QALD System In this paper, we examine the impact of lexicalization on Question Answering over Linked Data (QALD). It is well known that one of the key challenges in interpreting natural language questions with re...

🚨 Thrilled to announce a new preprint!

📝 Title: Lexicalization Is All You Need: Examining the Impact of Lexical Knowledge in a Compositional QALD System
👩‍🎓👨‍🎓 Authors: David M. Schmidt, Mohammad Fazleh Elahi, Philipp Cimiano
🔗 Preprint: arxiv.org/abs/2411.03906