Chair Data Science at CLEF 2026
12 October 2026
From the 21st to 24th of September, the Conference and Labs of the Evaluation Forum (CLEF) 2026 took place at Friedrich-Schiller-University in Jena, Germany. During four conference days, researchers from the fields of computer science, mathematics, biology and digital humanities discussed results from peer-reviewed publications and evaluation labs. The topics that were discussed in evaluation labs involved problems in natural language processing such as biomedical information extraction (BioASQ), sexism and humor detection (EXIST and JOKER) or the augmentation of historical news articles with geolocatable information on mentions of persons (HIPE), among other topics.
The goal of the HIPE shared task was the development of multilingual systems capable of determining whether a person that is mentioned in a historical news article has been at a place which is also mentioned within the same text. Furthermore, the classification is split into two subtasks: classifying whether the person was ever at said place and whether the person was at the mentioned place within the immediate timeframe of the publication of the news article. The organizers shared digitized 19th and 20th century texts in English, French and German, as well as French literary as a surprise test. Systems were developed using training data shared prior to official test data release and were then evaluated within two days, once the testing data was shared.
Figure: Moritz Hennen with his poster.
Moritz Hennen represented the chair for Data Science and presented the publication ”FI-CODE@HIPE-2026: Efficient Person-Place Relation Classification from Distance Heuristics to Prompted Language Models“ both as a poster and orally. The team Data Science (Lena Griesbeck, Moritz Hennen, Florian Babl and Michaela Geierhos) developed three distinct approaches to the HIPE 2026 shared task. These systems tackle the classification problems at different layers, with varying complexity and compute demand. The most efficient approach aims to refrain from using machine learning alltogether and instead leverage insights from the training data to solve the HIPE classification task. In detail, the approach revolved around the idea that persons predominantly have been to those places, that are mentioned in close proximity to each other. Using this simple method, the Data Science team was able to score second in the accuracy-efficiency evaluation profile. Divided by languages, we scored 1st in French, 2nd in English and 5th in German. This data driven method allowed the team to gain further insight into the inherent structure of (historical) news articles and thus the area of information extraction and could be used as a baseline comparison method in the future.
Sources: Moritz Hennen (RI CODE)
