Introduction

In the July 1983 issue of The New York Times, American historian Daniel J. Boorstin remarked on the computerization of libraries as follows:

Technology is so much fun but we can drown in our technology. The fog of information can drive out knowledge.

Approximately four decades later, its relevance may arguably surpass its initial significance. Indeed, the advent of Web 2.0 technologies has ushered in a process of disintermediation in ROMCIR 2024: The 4th Workshop on Reducing Online Misinformation through Credible Information Retrieval (held as part of ECIR 2024: the 46th European Conference on Information Retrieval), March 24, 2024, Glasgow, UK * Corresponding author. Envelope marinella.petrocchi@iit.cnr.it (M. Petrocchi); marco.viviani@unimib.it (M. Viviani) GLOBE https://www.iit.cnr.it/en/marinella.petrocchi (M. Petrocchi); http://www.ir.disco.unimib.it/people/marco-viviani/ (M. Viviani) Orcid 0000-0003-0591-877X (M. Petrocchi); 0000-0002-2274-9050 (M. Viviani) the creation and dissemination of online content within the Social Web, resulting in welldocumented challenges such as information overload [1,2] and the proliferation of misinformation [3]. These issues hamper users' ability to access truly valuable information for their needs [4,5]. Furthermore, recent advances in generative models such as Large Language Models (LLMs) pose a new threat, as they can generate text that mimics human writing but may lack accuracy and truthfulness [6,7,8].

Therefore, the ROMCIR Workshop focuses on studying and developing Information Retrieval (IR) solutions aimed at providing users with access to relevant and truthful information, while also addressing the phenomenon of information disorder across various domains. Information disorder encompasses a spectrum of issues, from unintentional misinformation rooted in ignorance or bias to deliberate dissemination of false content, both manually and through automated means [9,10]. This challenge is exacerbated by filter bubbles and echo chambers prevalent in the digital ecosystem [11,12,13,14,15].

The resolution of the information disorder issue is inherently complex, involving various types of content, web platforms, and user objectives. Furthermore, emerging AI-related concerns such as the explainability of search results [16,17], assessment of truthfulness in user-generated content [5,18], and the use of generative models to support IR systems require attention [19,20]. In addition, ensuring data confidentiality, especially in unstructured data, is paramount [21,22]. In this context, the development of appropriate experimental evaluation paradigms for IR systems is crucial [23,24].

Aim and topics of interest

The Workshop seeks to explore the challenges surrounding online information disorder within the realm of Information Retrieval, while also delving into associated domains of Artificial Intelligence such as Natural Language Processing (NLP), Natural Language Understanding (NLU), Computer Vision, Machine Learning, and Deep Learning. Therefore, the focal points of interest for ROMCIR 2024 encompass, yet are not confined to: . His current research interests include a wide range of Information Retrieval (IR) and related areas such as: IR probabilistic models, summarization, novelty detection, sentence retrieval, patent search and opinion mining. He is an active member of the IR community. He participated in the Programme Committee of prestigious international conferences such as SIGIR or ECIR. He also led several R&D projects and contracts in the area of search technologies. In 2011 he was recognized with an ACM senior member award. Website: https://citius.gal/team/david-enrique-losada-carril/ Health misinformation detection: search challenges, annotation issues and reliability of LLMs. In this presentation, we will share insights from our work at CiTIUS (University of Santiago de Compostela, Spain) on the development of technological and scientific solutions for detecting health misinformation. We will delve into the complexities of developing a multifaceted retrieval system for misinformation detection that integrates multiple content-based features. The challenges of creating robust credibility benchmarks, given the subjective nature of credibility, will also be discussed. Lastly, we will share our recent efforts to evaluate the quality of LLMs' responses to health-related queries.

Submissions

The ROMCIR 2024 workshop received 14 submissions, of which 6 were accepted, resulting in an acceptance rate of approximately 43%. The submissions came from five different countries, including Germany (2), Italy (1), The Netherlands (1), Poland (1), and Spain (1). This year, submissions have particularly focused on the issues of health misinformation, fact-checking (including multimedia and multimodal approaches), and information filtering and misinformation.

The work by Fernández Pichel, Bink, Losada, and Elsweiler focuses on assessing the credibility of information in the context of health information seeking. The authors acknowledge the subjectivity and bias susceptibility of this process and emphasize the importance of defining robust guidelines for credibility assessment. Through a study involving 1,000 participants, they demonstrate a correlation between participants' judgments and the reference values established following such guidelines. Further data analyses reveal concerning insights into people's ability to evaluate the credibility of online medical content, posing the risk of personal harm.

Mongelli, Maiano, and Amerini's work focuses on enhancing deepfake detection by simultaneously analyzing audio and visual cues, proposing the Convolutional Multimodal deepfake detection model (CMDD). This novel approach improves detection accuracy by leveraging the power of Convolutional Neural Networks (CNNs) to extract spatial and temporal features concurrently. Frick and Steinebach address the challenge of combating false information on social media by proposing a method to assess the check-worthiness of tweets. Their approach incorporates analysis of image content, captions, and text obtained from optical character recognition to outperform existing recognition techniques. Vogel, Möhle, Meghana, and Steinebach's work focuses on detecting check-worthy statements to prioritize claims for fact-checking. They propose an adapter fusion model combining task and Named Entity Recognition (NER) adapters, achieving state-of-the-art results in check-worthiness benchmarks.

In relation to the last research topic, Hornig, Pera, and Scholtes delve into the issue of propagation of misinformation in the domain of video recommendation. They evaluate a range of top-N recommendation algorithms to assess their effectiveness in minimizing misinformation while optimizing overall performance. Their empirical exploration highlights the potential of certain algorithms, including neighborhood-based, neural, and advanced collaborative filtering approaches, in combating misinformation and promoting responsible recommender systems. Hasimi and Poniszewska-Maranda focus on the broader implications of fake news and disinformation on human rights, particularly freedom of speech. They explore the use of Artificial Intelligence (AI) in detecting and filtering disinformation, highlighting the risks to freedom of expression posed by censorship and the suppression of critical thinking.

Past Editions