Media framing refers to the emphasis on specific aspects of perceived reality to shape how an issue is defined and understood. Its primary purpose is to shape public perceptions often in alignment with the authors' opinions and stances. However, the interaction between stance and media frame remains largely unexplored. In this work, we apply an interdisciplinary approach to conceptualize and computationally explore this interaction with internet memes on climate change. We curate CLIMATEMEMES, the first dataset of climate-change memes annotated with both stance and media frames, inspired by research in communication science. CLIMATEMEMES includes 1,184 memes sourced from 47 subreddits, enabling analysis of frame prominence over time and communities, and sheds light on the framing preferences of different stance holders. We propose two meme understanding tasks: stance detection and media frame detection. We evaluate LLaVA-NeXT and Molmo in various setups, and report the corresponding results on their LLM backbone. Human captions consistently enhance performance. Synthetic captions and human-corrected OCR also help occasionally. Our findings highlight that VLMs perform well on stance, but struggle on frames, where LLMs outperform VLMs. Finally, we analyze VLMs' limitations in handling nuanced frames and stance expressions on climate change internet memes.

Recent advancements in Natural Language Processing have significantly improved the extraction of structured semantic representations from unstructured text, especially through Frame Semantic Role Labeling (FSRL). Despite this progress, the potential of Retrieval-Augmented Generation (RAG) models for frame detection remains under-explored. In this paper, we present the first RAG-based approach for frame detection called RCIF (Retrieve Candidates and Identify Frames). RCIF is also the first approach to operate without the need for explicit target span and comprises three main stages: (1) generation of frame embeddings from various representations ; (2) retrieval of candidate frames given an input text; and (3) identification of the most suitable frames. We conducted extensive experiments across multiple configurations, including zero-shot, few-shot, and fine-tuning settings. Our results show that our retrieval component significantly reduces the complexity of the task by narrowing the search space thus allowing the frame identifier to refine and complete the set of candidates. Our approach achieves state-of-the-art performance on FrameNet 1.5 and 1.7, demonstrating its robustness in scenarios where only raw text is provided. Furthermore, we leverage the structured representation obtained through this method as a proxy to enhance generalization across lexical variations in the task of translating natural language questions into SPARQL queries.

Defined by Entmann (1993) as a process of selecting and making more salient the specific aspects of social reality, framing is among the most extensively used concepts in the field of communication science (Olsson & Ihlen, 2018). The abundant body of research utilising the concept of framing highlights the versatility of the concept: it has been used for examining the representation of armed conflict (Tschirky & Makhortykh, 2024), climate change (Vu et al., 2021), politics (Ogan et al., 2018), and racial injustice (Lane et al., 2020). The diversity of areas in which the concept of framing is applied and the vagueness of its operationalisation are, however, occasionally viewed as the concept's weakness: Cacciatore et al. (2016) note that it results in the unnecessarily broad understanding of framing that overlaps with other concepts, such as agenda-setting, and diminishes its explanatory potential. Despite the above-mentioned criticism, we suggest that framing remains an essential tool for understanding how certain interpretations of important societal issues become more visible and in which ways individuals are exposed to these interpretations. The importance of such an understanding increases under the conditions of the high-choice media environment (van Aelst et al., 2017) in which we are consuming information. With more available information sources and, consequently, more possibilities for being exposed to them -- both selectively (Messing & Westwood, 2014) and incidentally (Lee & Kim, 2014) -- it is crucial to be able to distinguish between frames coming from these sources, especially regarding the salience of epistemically contested issues which can easily amplify polarisation in the society. The ability to detect the presence or absence of specific frames in this context also becomes paramount for detecting attempts to manipulate public opinion. Another reason why frame detection is highly relevant is the growing reliance on artificial intelligence (AI)-powered systems for organising and generating information regarding societally relevant issues. The adoption of systems such as search engines and recommendations systems and, recently, generative AI-powered chatbots has profound implications for how individuals are exposed to information as these systems decide what information sources and interpretations are prioritised in response to the user input (e.g.