The open positions for our project "BehAIve: Behaviour Monitoring and Support of Older Adults" are coming out, so keep an eye on our website for new announcements. https://datascience.uni-greifswald.de/forschung/projects/behaive/
Kristina Yordanova
- 18 Followers
- 13 Following
- 44 Posts
Our work in progress paper "Loc-DemCare: Supporting Informal Dementia
Caregivers with a Local, Intelligent Chatbot" has been accepted to IEEE PerCom! The paper proposes a GraphRAG approach for the support of family members and informal caregivers in coping with agitated behaviour in persons with dementia. It investigates the usage of small large language models that can be run locally and shows that they perform comparably to their larger counterparts when combined with our agitation ontology.
Caregivers with a Local, Intelligent Chatbot" has been accepted to IEEE PerCom! The paper proposes a GraphRAG approach for the support of family members and informal caregivers in coping with agitated behaviour in persons with dementia. It investigates the usage of small large language models that can be run locally and shows that they perform comparably to their larger counterparts when combined with our agitation ontology.
Our demo paper "KITCHAT: Kitchen Inference and Tracking Using Clue-Based Home Assistive Talk" has been accepted to IEEE PerCom! It showcases a system architecture targeted at elderly individuals and people with cognitive impairments who struggle with the completion of daily living tasks. It utilises hybrid machine learning methods for state estimation and error detection, local LLMs for state translation into natural language, and a speech generation module for communication with the users.
Universität GreifswaldRückblick auf den MV-Tag in Greifswald: Wir sind begeistert von der Resonanz, führten intensive Gespräche und konnten die kleinen und großen Besucher*innen von unserer Forschung begeistern! Auch Bettina Martin, Ministerium für Wissenschaft, Kultur, Bundes- und Europaangelegenheiten Mecklenburg-Vorpommern des Landes Mecklenburg-Vorpommern, kam am Stand vorbei, hielt ein Stück Erdkruste aus 30 Kilometer Tiefe in der Hand und erfuhr, wie Roboter im Studiengang Data Science eingesetzt werden.
Digitale Reise ins Heilige Land! Die KI-App 📲✨„Tin Gustaf – Your Holy Land AI Pocket Museum“ des Gustaf-Dalman-Instituts an der Uni Greifswald ermöglicht neue Einblicke in die Kultur Palästinas um 1900. Es können mehr als 14 000 Datensätze der Institutssammlung ausgewertet werden. Beim Klick auf ein einzelnes Foto werden weitere Informationen wie Bildmotiv, Alter und Fotograf sichtbar, die in der Deutschen Digitalen Bibliothek hinterlegt wurden. 👉 https://ugreif.de/7dgvk
We are happy to announce that this year the ARDUOUS workshop will be affiliated to ECAI'25! https://arduous.eu/ https://ecai2025.org/ @emmatonkin @GregTourte
We are looking for a highly motivated candidate to develop the data management infrastructure of the collaborative research centre WETSCAPES2.0 and to investigate new automatic methods for quality assurance of heterogenous data. https://shorturl.at/yI47b
I am happy to announce that together with @emmatonkin we are serving as guest editors for the special collection "Clinical Informatics" at Nature's "Scientific Data". https://www.nature.com/collections/aecgjaiccf
Frank KrügerWe just published the first dataset created within our @Textplus cooperation project. For this dataset, we manually transcribed the tables of arriving spa and bathing guests from the Swinemünder Badeanzeiger, published between 1910 and 1932. The data is available under CC0 from https://zenodo.org/records/14603757
OCR Groundtruth for Swinemünder Badeanzeiger
This dataset contains the ground truth annotation for extracting and structuring information from the old newspaper "Swinemünder Badeanzeiger" tables. The newspaper was obtained from Digitale Bibliothek Mecklenburg Vorpommern https://www.digitale-bibliothek-mv.de/viewer/toc/PPN636776093/ The data was obtained by selecting one "Swinemünder Badeanzeiger" image per year and manually transcribing the content. The dataset is structured based on the newspaper's publication year. One folder for each year contains a folder named according to the original image ID and includes the following data table_[running_number].jpg image with the segmented table table_[running_number]_annotation.json data extracted and structured from the segmented image by manual transcription table_[running_number]_index_connected.json list that connected the entry with the corresponding table rows to maintain multi-row entries For each entry, a JSON entry was created and added to table_[running_number]_annotation.json, which consists of the following fields: input: Transcription of the original row, including markers for columns Nummer: The sequence number of the row as extracted from the input field Vorname: The first name, if it exists otherwise null Nachname: The last name, if it exists; otherwise null Titel: The (academic) title, if it exists, otherwise null Beruf: The profession, if it exists; otherwise null Sozialer Stand: The social status, if it exists, otherwise null Begleitung: Any companion, such as family members or servants, if exists, otherwise null Wohnort: The city, where the person(s) arrived from, if it exists, otherwise null Wohnung: The local residence, such as a hotel, pension, or vacation home, if it exists, otherwise null Personenanzahl: The overall number of persons that are represented by this entry In addition to the separate annotation files, the file swinebad_groundtruth.json has a complete list of all entries to facilitate more straightforward data analysis. To this end, each entry was completed with the following data. date: The publication date of the newspaper where the entry was published The following example lists an entry which was obtained from the fourth line of the table as published at https://www.digitale-bibliothek-mv.de/viewer/image/PPN636776093_1910/1/LOG_0003/ { "input": "973 | Dr. Auerbach, Richard, Journalist, mit Frau | „ | Villa Kaiser Wilhelm | 2", "Nummer": "973", "Vorname": "Richard", "Nachname": "Auerbach", "Titel": "Dr.", "Beruf": "Journalist", "Sozialer Stand": null, "Begleitung": "mit Frau", "Wohnort": "Berlin", "Wohnung": "Villa Kaiser Wilhelm", "Personenanzahl": "2", "date": "1910-06-06" }, You are welcome to cite the 'Digitale Bibliothek MV / Universität Greifswald' (+ URN for digital publications or the shelfmark for printed publications) as the source for the images.
I'm excited to be part of the new SFB Transregio WETSCAPES2.0, where I will contribute to the research data management in the domain of rewetted fen landscapes https://www.dfg.de/de/service/presse/pressemitteilungen/2024/pressemitteilung-nr-47
