Search

In today’s digital age, online platforms serve as a primary conduit for individuals seeking health-related information. While the web provides a vast repository of health knowledge, it has simultaneously birthed a daunting challenge: the proliferation of online health misinformation. This malady, when unchecked, poses serious repercussions for public health, as individuals, often untrained in medical nuances, make health decisions based on misleading or outright false information. Addressing this pressing concern, my thesis delves deep into understanding and mitigating the challenge of Online Health Misinformation, exploring avenues of detection, retrieval, and explainability. Our research journey began with a focus on the detection of health misinformation, by utilizing structural, content, and context-aware strategies. This new model was uniquely poised to assess the truthfulness of online health content. By exploiting a specialized medical lexicon, the model crafted embedded representations of web pages, thereby comprehending subtle nuances associated with health misinformation. The innovation lay in the model’s capability to also consider URLs embedded within these pages, which proved instrumental in the classification effort. Moreover, the strategic inclusion of a domain-specific pre-trained representation considerably amplified the model’s efficiency. In the subsequent phase, we built upon these foundational findings to birth the Vec4Cred model - an advanced approach tailored explicitly for detecting health misinformation online. Vec4Cred was underscored by a multi-layered framework, focusing on embedding representations of various web page attributes. The model’s prowess lay in its capacity to seamlessly integrate embedding representations from parts-of-speech tags and keywords from linked pages. Forward-looking, the model beckons enhancement through advanced contextual embedding methodologies, thereby continually refining its accuracy in misinformation detec, Nell’odierna epoca digitale, le piattaforme online costituiscono uno dei mezzi principali utilizzati dalle persone per cercare informazioni relative alla propria salute. Nonostante il web sia un vasto repository di conoscenze in tale ambito, è affetto dal problema dalla proliferazione e diffusione di disinformazione sanitaria (health misinformation). Questo problema può portare a gravi ripercussioni per la salute pubblica, poiché gli individui possono assumere decisioni basate su informazioni fuorvianti o palesemente false. Questa tesi di dottorato propone quindi tecniche per la mitigazione della disinformazione sanitaria online, affrontando in particolare i problemi della sua identificazione, del reperimento di informazione affidabile e della spiegabilità (explainability) degli algoritmi proposti. Il lavoro di ricerca è iniziato concentrandosi sull’idendificazione della dinsinformazione sanitaria, definendo un modello basato su aspetti topologici, contenutistici e contestuali delle pagine web. Sfruttando un lessico medico specializzato, il modello ha creato rappresentazioni embedded delle pagine. Il modello ha quindi la capacità di considerare, oltre ad altre feature lessicali, anche gli URL incorporati, che si sono rivelati strumentali nello task di classificazione. Le valutazioni comparative hanno dimostrato la superiorità del modello rispetto alle tecniche tradizionali di machine learning. Inoltre, l'inclusione di una rappresentazione pre-addestrata specifica del dominio ha notevolmente amplificato l'efficacia del modello. Si è poi sviluppato il modello Vec4Cred, la cui caratteristica principale è quella di integrare rappresentazioni embedded da parti del discorso e parole chiave delle pagine collegate. I risultati sperimentali hanno confermato la bontà del modello nell'identificare la disinformazione sulla salute online, sottolineando la sua adattabilità ed efficacia. Il reperimento di informazioni sanitarie veritiere costituisce un’altra fase di que

In this paper, we propose a novel approach to consider multiple dimensions of relevance in cross-encoder re-ranking. On the one hand, cross-encoders constitute an effective solution for re-ranking when considering a single relevance dimension such as topicality, but are not designed to straightforwardly account for additional relevance dimensions. On the other hand, the majority of re-ranking models accounting for multdimensional relevance are often based on the aggregation of multiple relevance scores at the re-ranking stage, leading to potential compensatory effects. To address these issues, in the proposed solution we enhance the candidate documents retrieved by a first-stage lexical retrieval model with suitable relevance statements related to distinct relevance dimensions, and then perform a re-ranking on them with cross-encoders. In this work we focus, in particular, on an extra dimension of relevance beyond topicality, namely, credibility, to address health misinformation in the Consumer Health Search task. Experimental evaluations are performed by considering publicly available datasets; our results show that the proposed approach statistically outperforms state-of-the-art aggregation-based and cross-encoder re-rankers.

Introduction: People are today increasingly relying on health information they find online to make decisions that may impact both their physical and mental wellbeing. Therefore, there is a growing need for systems that can assess the truthfulness of such health information. Most of the current literature solutions use machine learning or knowledge-based approaches treating the problem as a binary classification task, discriminating between correct information and misinformation. Such solutions present several problems with regard to user decision making, among which: (i) the binary classification task provides users with just two predetermined possibilities with respect to the truthfulness of the information, which users should take for granted; indeed, (ii) the processes by which the results were obtained are often opaque and the results themselves have little or no interpretation. Methods: To address these issues, we approach the problem as an ad hoc retrieval task rather than a classification task, with reference, in particular, to the Consumer Health Search task. To do this, a previously proposed Information Retrieval model, which considers information truthfulness as a dimension of relevance, is used to obtain a ranked list of both topically-relevant and truthful documents. The novelty of this work concerns the extension of such a model with a solution for the explainability of the results obtained, by relying on a knowledge base consisting of scientific evidence in the form of medical journal articles. Results and discussion: We evaluate the proposed solution both quantitatively, as a standard classification task, and qualitatively, through a user study to examine the “explained” ranked list of documents. The results obtained illustrate the solution's effectiveness and usefulness in making the retrieved results more interpretable by Consumer Health Searchers, both with respect to topical relevance and truthfulness.

Research aimed at finding solutions to the problem of the diffusion of distinct forms of non-genuine information online across multiple domains has attracted growing interest in recent years, from opinion spam to fake news detection. Currently, partly due to the COVID-19 virus outbreak and the subsequent proliferation of unfounded claims and highly biased content, attention has focused on developing solutions that can automatically assess the genuineness of health information. Most of these approaches, applied both to Web pages and social media content, rely primarily on the use of handcrafted features in conjunction with Machine Learning. In this article, instead, we propose a health misinformation detection model that exploits as features the embedded representations of some structural and content characteristics of Web pages, which are obtained using an embedding model pre-trained on medical data. Such features are employed within a deep learning classification model, which categorizes genuine health information versus health misinformation. The purpose of this article is therefore to evaluate the effectiveness of the proposed model, namely Vec4Cred, with respect to the problem considered. This model represents an evolution of a previous one, with respect to which new features and architectural choices have been considered and illustrated in this work.