The increasing frequency of foodborne illnesses, safety hazards, and disease outbreaks in the food supply chain demands urgent attention to protect public health. These incidents, ranging from contamination to intentional adulteration of food and feed, pose serious risks to consumers, leading to poisoning, and disease outbreaks that lead to product recalls. Identifying and tracking the sources and pathways of contamination is essential for timely intervention and prevention. This paper explores the use of social media and regulatory news reports to detect food safety issues and disease outbreaks. We present an automated approach leveraging a multi-task sequence labeling and sequence classification model that uses a liquid time-constant neural network augmented with a graph convolution network to extract and analyze relevant information from social media posts and official reports. Our methodology includes the creation of annotated datasets of social media content and regulatory documents, enabling the model to identify foodborne infections and safety hazards in real-time. Preliminary results demonstrate that our model outperforms baseline models, including advanced large language models like LLAMA-3 and Mistral-7B, in terms of accuracy and efficiency. The integration of liquid neural networks significantly reduces computational and memory requirements, achieving superior performance with just 1.2 × e⁶ bytes of memory, compared to the 20.3 GB of GPU memory needed by traditional transformer-based models. This approach offers a promising solution for leveraging social media data in monitoring and mitigating food safety risks and public health threats.

Corporate environmental claims frequently appear in news articles, providing an external view of sustainability-related communications. However, extracting such claims remains challenging as they are often implicit, distributed across multiple sentences, and embedded within narrative contexts. In this work, we propose an end-to-end system for automated environmental claim extraction enriched with contextual metadata. Our pipeline combines multi-label topic classification, topic-conditioned candidate claim-evidence extraction using large language models (LLMs), and candidate filtering with ASE-BERT, a lightweight aspect- and sentiment-aware classifier for claim detection. To support evaluation, we construct a curated dataset of corporate environmental news, annotated with claims, supporting evidence, and contextual attributes. Experimental results demonstrate that for claim classification, ASE-BERT attains the highest F1 (0.87) across the four combined environmental claim datasets, exceeding the best baseline without task-specific fine-tuning by 4 points. For end-to-end extraction, topic conditioning alone improves LLM F1 by an average of 15 points, and our full pipeline adds a further 5 points over the best topic-conditioned setting. These findings demonstrate that incorporating topic awareness and claim filtering substantially improves environmental claim extraction from corporate news, supporting downstream sustainability intelligence and decision-making tasks.

Corporate sustainability reports communicate environmental disclosures through text, tables, charts, and infographics, making assessment against sustainability reporting standards challenging. Existing disclosure extraction systems focus on information extraction but do not determine whether disclosed evidence satisfies reporting requirements. We study requirement-level multimodal environmental disclosure alignment, a task that evaluates whether sustainability report evidence satisfies GRI 305 and ESRS E1-4 and E1-6 requirements. To support this task, we curate an expert-annotated benchmark comprising 731 disclosure-alignment instances from 43 sustainability reports. Using this benchmark, we evaluate three Vision Language Models (VLMs): GPT-5.5, Gemini, and LLaMA Maverick under three conditioning strategies: open-ended, framework-specific, and knowledge-augmented. Results show that although frontier VLMs possess substantial implicit knowledge of sustainability reporting standards, they frequently produce over generalized and optimistic alignment judgments. Requirement level grounding improves alignment accuracy by 12% and yields more accurate disclosure gap identification. Our findings demonstrate the value of knowledge grounded multimodal reasoning for scalable, interpretable, and auditable sustainability disclosure assessment.

Sustainability metrics have increasingly become a crucial non-financial criterion in investment decision-making. Organizations worldwide are recognizing the importance of sustainability and are proactively highlighting their efforts through specialized sustainability reports. Unlike traditional annual reports, these sustainability disclosures are typically text-heavy and are often expressed as infographics, complex tables, and charts. The non-machine-readable nature of these reports presents a significant challenge for efficient information extraction. The rapid advancement of Vision Language Models (VLMs) has raised the question whether these VLMs can address such challenges in domain specific task. In this study, we demonstrate the application of VLMs for extracting sustainability information from dedicated sustainability reports. Our experiments highlight the limitations in the performance of several open-source VLMs in extracting information about sustainability disclosures from different type of pages.

Advanced neural network architectures have provided several opportunities to develop systems to automatically capture information from domain-specific unstructured text sources. The FinSim4-ESG shared task, collocated with the FinNLP workshop, proposed two sub-tasks. In sub-task1, the challenge was to design systems that could utilize contextual word embeddings along with sustainability resources to elaborate an ESG taxonomy. In the second sub-task, participants were asked to design a system that could classify sentences into sustainable or unsustainable sentences. In this paper, we utilize semantic similarity features along with BERT embeddings to segregate domain terms into a fixed number of class labels. The proposed model not only considers the contextual BERT embeddings but also incorporates Word2Vec, cosine, and Jaccard similarity which gives word-level importance to the model. For sentence classification, several linguistic elements along with BERT embeddings were used as classification features. We have shown a detailed ablation study for the proposed models.