Framing Food Online Discourse: Employing Generative AI and Semantic Analysis for Digital Lexicography and Terminology Extraction

Social media platforms provide vast amounts of authentic, user-generated linguistic data that can reveal conceptual structures, terminology, and cultural discourse. This study analyzes Instagram posts related to traditional and local foods in Greece, focusing on semantic frames, key concepts, and lexically salient items that contribute to food consumer identity. Using a bottom-up approach based on hashtag co-occurrence, we identified frames of locality, tradition, and food consumption as culturally embedded structures. Two generative AI tools, ChatGPT (English-based) and Llama-Krikri (Greek-based), were employed for food-relatedness and sentiment classification; ChatGPT achieved very high agreement with human judgments (F1 > 90%), while Llama-Krikri showed more modest performance. Case studies on Mediterranean snails and North Aegean cheeses revealed prototypical associations, with graviera and feta emerging as central references. Importantly, food names and related concepts are treated as terminological units, and selected cases were transformed into structured entries using TBX and OntoLex-Lemon, demonstrating applications to digital terminology and lexicography. The methodology highlights both the opportunities (frame-based vocabulary extraction, resource building) and challenges (Greeklish, spelling variation) of applying large language models to under-resourced languages. This work shows how social-media-based analyses can support the creation of interoperable, frame-based terminological and lexicographical resources.