Large Language Models (LLMs) are transforming personalized services by enabling adaptive, context-aware recommendations and interactions. However, deploying these models at scale raises significant concerns about environmental impact, fairness, privacy, and trustworthiness, including high energy consumption, biased outputs, privacy breaches, and hallucinations. The LLM4Good workshop was already hosted at UMAP’25 and is a half-day workshop that addresses these challenges by fostering dialogue on sustainable and ethical approaches to LLM-based personalization. Participants will explore energy-efficient techniques, bias mitigation, privacy-preserving methods, and responsible deployment strategies. The workshop aligns with Sustainable Development Goals and Digital Humanism principles. It aims to guide the development of trustworthy, human-centric LLM systems that positively impact education, healthcare, and other domains.

In the evolving realm of recommender systems, our study contributes to the understanding of potential improvements in news recommendation beyond accuracy.Central to our research is the integration of insights from news industry experts and prospective readers, compared with automated news recommendations.We conducted a labeling study with 168 articles, using Best-Worst Scaling (BWS) for ranking and topic modeling.This approach enabled a thorough examination of stakeholder expectations for ideal reading recommendations, specifically by investigating the gap between stated and revealed preferences.Our findings show alignment in ranking behavior among journalists, prospective readers, and the BM-25 algorithm.However, preferences for different beyondaccuracy measures varied.Accompanying this work, a corpus of news articles and the labeled rankings have been made available.

Populism is a phenomenon that is noticeably present in the political landscape of various countries over the past decades. While populism expressed by politicians has been thoroughly examined in the literature, populism expressed by citizens is still underresearched, especially when it comes to its automated detection in text. This work presents the PopAut corpus, which is the first annotated corpus of news comments for populism in the German language. It features 1,200 comments collected between 2019-2021 that are annotated for populist motives anti-elitism, people-centrism and people-sovereignty. Following the definition of Cas Mudde, populism is seen as a thin ideology. This work shows that annotators reach a high agreement when labeling news comments for these motives. The data set is collected to serve as the basis for automated populism detection using machine-learning methods. By using transformer-based models, we can outperform existing dictionaries tailored for automated populism detection in German social media content. Therefore our work provides a rich resource for future work on the classification of populist user comments in the German language.

The research domain of recommender systems is rapidly evolving. Initially, optimization efforts focused primarily on accuracy. However, recent research has highlighted the importance of addressing bias and beyond-accuracy measures such as novelty, diversity, and serendipity. With the rise of multi-domain recommender systems, the need to re-examine bias and beyond-accuracy measures in cross-domain settings has become crucial. Traditional methods face challenges such as cold-start problems, which can potentially be mitigated by leveraging LLMs. This proposed work investigates how LLM-based recommendation methods can enhance cross-domain recommender systems, focusing on identifying, measuring, and mitigating bias while evaluating the impact of beyond-accuracy measures. We aim to provide new insights by comparing traditional and LLM-based systems within a real-world environment encompassing the domains of news, books, and various lifestyle areas. Our research seeks to address the outlined gaps and develop effective evaluation strategies for the unique challenges posed by LLMs in cross-domain recommender systems.

This paper introduces the Austrian German sentiment dictionary ALPIN to account for the lack of resources for dictionary-based sentiment analysis in this specific variety of German, which is characterized by lexical idiosyncrasies that also affect word sentiment. The proposed language resource is based on Austrian news media in the field of politics, an austriacism list based on different resources and a posting data set based on a popular Austrian news media. Different resources are used to increase the diversity of the resulting language resource. Extensive crowd-sourcing is performed followed by evaluation and automatic conversion into sentiment scores. We show that crowd-sourcing enables the creation of a sentiment dictionary for the Austrian German domain. Additionally, the different parts of the sentiment dictionary are evaluated to show their impact on the resulting resource. Furthermore, the proposed dictionary is utilized in a web application and available for future research and free to use for anyone.

The domain of German polarity dictionaries is heterogeneous with many small dictionaries created for different purposes and using different methods. This paper aims to map out the landscape of freely available German polarity dictionaries by clustering them to uncover similarities and shared features. We find that, although most dictionaries seem to agree in their assessment of a word’s sentiment, subsets of them form groups of interrelated dictionaries. These dependencies are in most cases an immediate reflex of how these dictionaries were designed and compiled. As a consequence, we argue that sentiment evaluation should be based on multiple and diverse sentiment resources in order to avoid error propagation and amplification of potential biases.