Bawa, VanishikaBaroud, IbrahimSchaffer, StefanKlein, MaikeKrupka, DanielWinter, CorneliaWohlgemuth, Volker2023-11-292023-11-292023978-3-88579-731-9https://dl.gi.de/handle/20.500.12116/43090As part of the “Stadt-Land-Fluss” project, we investigate the usage of an AI-powered chatbot to support three key players of the regional food supply chain: producers, food processors, and buyers, in selling or buying their products digitally. A Minimal Viable Prototype was developed with input from domain specialists and includes two scenarios with concrete use cases; namely, “see crop plannings”, “create sales item”, “looking for a food-processor” in the first and, “create delivery series” or “see offers” in the second. Additionally, the chatbot provides personalized follow-up questions in the case where an offer or crop planning is due. We identify 17 intents and 9 entities and evaluate the performance of NLU components using 5-fold cross validation. This work contributes to the field by curating domain-specific use cases and data based on expert insights. We also extend the data using large language models. Additionally, we show that using DIET with spaCy outperforms using DIET with BERT-based tokenizers and featurizers in Intent Detection (ID) and Entity Recognition (ER) on our data when keeping other parts of the chatbot constant.enChatbotConversational DesignPre-trained Large Language ModelsPrototypingFood Supply MarketRegional FoodDeep LearningText/Conference Paper10.18420/inf2023_1641617-5468