The swift proliferation of false information on social media presents significant risks to public confidence and the stability of democracy, which motivates the development of machine learning and deep learning-based fake news detectors. While these systems can effectively analyze news content and user interactions, they remain vulnerable to adversarial attacks. Prior research has focused mainly on modifying article text or retrieving generic user comments, leaving comment-based attack strategies underexplored. Existing comment-based attacks often rely on generating synthetic text that can be unrealistic or retrieving existing comments without strategic guidance that ignores feature importance. In this work, we introduce a novel attack surface that combines model interpretability with generative language models. Our approach uses SHAP (SHapley Additive exPlanations) to identify influential tokens driving fake or real classifications and prompts a large language model (LLM) to generate contextually credible human-like comments that utilize influential tokens. The generated comments are then appended to the article, evaluated against multiple state?of?the?art detectors (dEFEND, TextCNN, and RoBERTa), and compared against existing comment-based attacks such as MALCOM, CopyCat, and retrieval?based methods. Our XAI-guided LLM-based approach is competitive compared to existing generative and retrieval-based attack methods, with higher attack success rates while maintaining naturalness and contextual relevance.

Large language models (LLMs) have dramatically reduced the cost and hardships of producing fluent and persuasive text, a capability that can also be repurposed to generate more or less convincing news. In this paper, we examine how subtle edits infused with LLM can systematically change the perceived believability of a news item. We introduce, for the first time, LLM-crafted targeted, low-visibility transformations to political news stories, preserving the core writing input, while altering credibility cues, to measure how adversarial changes shift human perception of believability. In contrast to prior work on fully synthetic forgeries or blunt paraphrases, our adversarial attack scenarios focus on subtle, context-aware credibility drift, involving small edits that alter causal assertions, remove qualifying facts, or introduce opposing evidence. These minimal modifications not only evade casual human scrutiny as reported in the literature, but also produce measurable shifts in perceived credibility as we show in our experimental evaluation.

This study expands the understanding of linguistic cues in fake news by exploring ambivalent language's role in distinguishing real from fake news and its impact on news spread. Unlike prior research focusing on positive or negative language separately, this work hypothesizes that fake news may exhibit higher ambivalence, aligning with its association with high-arousal emotions and writers' efforts to attract attention. Ambivalence, traditionally viewed as the co-occurrence of conflicting positive and negative elements, is extended here to include diverse dimensions such as textual, temporal, psychological, and content-related ambivalence.

The widespread use of social media has led to an increase in false and misleading information presented as legitimate news, also known as fake news. This poses a threat to societal stability and has led to the development of fake news de-tectors that use machine learning to flag suspicious information. However, existing fake news detection models are vulnerable to attacks by malicious actors who can manipulate data to change predictions. Research on attacks on news comments is limited, and current attack models are easily detectable. We propose two new attack strategies that instead use real, pre-existing comments from the same dataset as the news article to fool fake news detectors. Our experimental results show that fake news detectors are less robust to our proposed attack strategies than existing methods using pre-existing human-written comments, as well as a malicious synthetic comment generator.

The presence of fake news on online social media is overwhelming and is responsible for having impacted several aspects of people’s lives, from health to politics, the economy, and response to natural disasters. Although significant effort has been made to mitigate fake news spread, current research focuses on single aspects of the problem, such as detecting fake news spreaders and classifying stories as either factual or fake. In this article, we propose a new method to exploit inter-relationships between stories, sources, and final users and integrate prior knowledge of these three entities to jointly estimate the credibility degree of each entity involved in the news ecosystem. Specifically, we develop a new graph convolutional network, namely, Role-Relational Graph Convolutional Networks (Role-RGCN), to learn, for each node type (or role), a unique node representation space and jointly connect the different representation spaces with edge relations. To test our proposed approach, we conducted an experimental evaluation on the state-of-the-art FakeNewsNet-Politifact dataset and a new dataset with ground truth on news credibility degrees we collected. Experimental results show a superior performance of our Role-RGCN proposed method at predicting the credibility degree of stories, sources, and users compared to state-of-the-art approaches and other baselines.

In the world of cyber defense, anticipating potential attacks or any increase in risk of attacks is one of the most advantageous pieces of knowledge one can have. However, little research has been done in examining the larger geopolitical environment and using data sources available at the geopolitical level to predict cyberattacks in advance. To this end, we combine the use of a geopolitical conflict dataset, ICEWS, in combination with a cyberattack dataset from the Council on Foreign Relations to determine if we can predict cyberattacks targeting a given nation. We present a novel approach to identify periods of increased likelihood of cyberattacks at the country, regional, and global levels. The approach involves creating a news-based political event graph, generating vectorial representations of the graph using the SIR-GN structural iterative representation learning approach, and applying novelty detection models to predict future nation-initiated attacks. The proposed approach outperforms existing baselines for majority of cases in terms of F1-score, demonstrating its effectiveness in predicting cyberattacks.

Social networks are a platform for individuals and organizations to connect with each other and inform, advertise, spread ideas, and ultimately influence opinions. These platforms have been known to propel misinformation. We argue that this could be compounded by the recommender algorithms that these platforms use to suggest items potentially of interest to their users, given the known biases and filter bubbles issues affecting recommender systems. While much has been studied about misinformation on social networks, the potential exacerbation that could result from recommender algorithms in this environment is in its infancy. In this manuscript, we present the result of an in-depth analysis conducted on two datasets ( Politifact FakeNewsNet dataset and HealthStory FakeHealth dataset ) in order to deepen our understanding of the interconnection between recommender algorithms and misinformation spread on Twitter. In particular, we explore the degree to which well-known recommendation algorithms are prone to be impacted by misinformation. Via simulation, we also study misinformation diffusion on social networks, as triggered by suggestions produced by these recommendation algorithms. Outcomes from this work evidence that misinformation does not equally affect all recommendation algorithms. Popularity-based and network-based recommender algorithms contribute the most to misinformation diffusion. Users who are known to be superspreaders are known to directly impact algorithmic performance and misinformation spread in specific scenarios. Findings emerging from our exploration result in a number of implications for researchers and practitioners to consider when designing and deploying recommender algorithms in social networks.

This paper addresses the problem of identifying sockpuppet accounts on Wikipedia. We formulate the problem as a binary classification task and propose a set of features based on user activity and the semantics of their contributions to separate sockpuppets from benign users. We tested our system on a dataset we built (and released to the research community) containing 17K accounts validated as sockpuppets. Experimental results show that our approach achieves an F1-score of 0.82 and outperforms other systems proposed in the literature. Moreover, our proposed approach is able to achieve an F1-score of 0.73 at detecting sockpuppet accounts by just considering their first edit.

We propose an approach inspired by the diffusion of innovations theory to model and characterize fake news sharing in social media through the lens of the different levels of influential factors (users, networks, and news). We address the problem of predicting fake news sharing as a classification task and demonstrate the potentials of the proposed features by achieving an AUROC of 0.97 and an average precision of 0.88, consistently outperforming baseline models with a higher margin (about 30% of AUROC). Also, we show that news-based features are the most effective at predicting real and fake news sharing, followed by the user- and network-based features.

In this paper, we share preliminary results from our research work focused on understanding how people assess news items as fake or real and improve their ability to identify fake news. Using existing real/fake news samples and best practices in qualitative, inductive data analysis, we identify factors that appear to impede ability of individuals to identify fake news. Based on this work we suggest one approach to improve human ability to identify fake news, and sketch a process for systematic development of means for supporting people in identifying fake news.

The Panama Papers network maintained by the International Consortium of Investigative Journalists (ICIJ) represents a large set of relationships between people, companies, and organizations involved in the creation of offshore companies in tax-haven territories, mainly for hiding their assets. The Panama Papers network includes people or companies that had affairs with the Panamanian offshore law firm Mossack Fonseca, often with the purpose of laundering money. In our previous work, we proposed a ranking algorithm, namely the Suspiciousness Rank Back and Forth (SRBF) algorithm, that, given the Panama Papers network, leverages a blacklist of known bad entities to assign a degree of suspiciousness to each entity in the network. This algorithm proved to be efficient in detecting known bad entities in the Panama Papers, but we were not able to verify the accuracy of the produced entity ranking for non-blacklisted entities. In this paper, we propose to use the open-source intelligence (OSINT) methodology as a modern derivative of classical ethnographic and archaeological research methods that help us in validating with external open source data the ranking result of the Suspiciousness Rank Back and Forth algorithm. More specifically, we conduct a parallel, but independent, investigation using OSINT to assess the claims of SRBF algorithm. We identify positive outcomes from this study, describe current gaps in our process, and propose solutions to the gaps in order to better integrate the OSINT methodology with the SRBF ranking approach.

Wikipedia, the free and open-collaboration based online encyclopedia, has millions of pages that are maintained by thousands of volunteer editors. As per Wikipedia’s fundamental principles, pages on Wikipedia are written with a neutral point of view and maintained by volunteer editors for free with well-defined guidelines in order to avoid or disclose any conflict of interest. However, there have been several known incidents where editors intentionally violate such guidelines in order to get paid (or even extort money) for maintaining promotional spam articles without disclosing such.

This paper presents our on-going research on studying the actors responsible for misinformation spread and identifying potential victims. Preliminary results show that (i) there is a correlation between fake news publisher bias and its credibility and (ii) social network properties help in identifying active fake news spreaders. Moreover, we discuss the most vulnerable victims of fake news and report on our experience in educating seniors about online misinformation.

Depression is the most common mental illness in the U.S., with 6.7% of all adults who have experienced a major depressive episode. Unfortunately, depression extends to teens and young users as well, and researchers observed an increasing rate in the recent years (from 8.7% in 2005 to 11.3% in 2014 in adolescents and from 8.8% to 9.6% in young adults), especially among girls and women. People themselves are a barrier to fight this disease as they tend to hide their symptoms and do not receive treatments. However, protected by anonymity, they share their sentiments on the Web, looking for help.

Graph databases such as chemical databases, protein databases, and RNA motif databases, are simply a collection of graphs. Querying a graph database involves the computation of a subgraph isomorphism problem (which is NP-complete) for each graph in the database. Therefore, an index is required to filter out false positives and reduce the number of subgraph isomorphisms to compute.

Link prediction is the problem of inferring new relationships among nodes in a network that are likely to occur in the near future. Classical approaches mainly consider neighborhood structure similarity when linking nodes. However, we may also want to take into account if the two nodes are already indirectly interacting and if they will benefit from the link by having an active interaction over the time. For instance, it is better to link two nodes u and v if we know that these two nodes will interact in the social network even in the future, rather than suggesting v', who will never interact with u. In this paper, we deal with a new variant of the link prediction problem: given a pair of indirectly interacting nodes, predict whether or not they will form a link in the future. We propose a solution to this problem that leverages the predicted duration of their interaction and propose two supervised learning approaches to predict how long will two nodes interact in a network. Given a set of network-based predictors, the basic approach consists of learning a binary classifier to predict whether or not an observed indirect interaction will last in the future. The second and more fine-grained approach consists of estimating how long the interaction will last by modeling the problem via survival analysis or as a regression task. Once the duration is estimated, new links are predicted according to their descending order. Experimental results on the Facebook Network and Wall Interaction dataset show that our more fine-grained approach performs the best with an AUROC of 0.85 and clearly beats a link prediction model that does not consider the interaction duration and is based only on network properties.

The Panama Papers represent a large set of relationships between people, companies, and organizations that had affairs with the Panamanian offshore law firm Mossack Fonseca, often due to money laundering. In this paper, we address for the first time the problem of searching the Panama Papers for people and companies that may be involved in illegal acts. We use a collection of international blacklists of sanctioned people and organizations as ground truth for bad entities. We propose a new ranking algorithm, named Suspiciousness Rank Back and Forth (SRBF), that leverages this ground truth to assign a degree of suspiciousness to each entity in the Panama Papers. We experimentally show that our algorithm achieves an AUROC of 0.85 and an Area Under the Recall Curve of 0.87 and outperforms existing techniques.

In this paper, we focus on English Wikipedia, one of the main user-contributed content systems, and study the problem of predicting which users will become inactive and stop contributing to the encyclopedia. We propose a predictive model leveraging frequent patterns appearing in user's editing behavior as features to predict active vs. inactive Wikipedia users. Our experiments show that our method can effectively predict inactive users with an AUROC of 0.97 and significantly beats competitors in the task of early prediction of inactive users.

Effective friend classification in Online Social Networks (OSN) has many benefits in privacy. Anything posted by a user in social networks like Facebook is distributed among all their friends. Although the user can select the manual option for their post-dissemination, it is not feasible every time. Since not all friends are the same in social networks, the visibility access for the post should be different for different strengths of friendship for privacy. Previous works in finding friendship strength in social networks have used interaction and similarity based features but none of them has considered using sentiment-based features as the driving factor to determine the strength.

A graph database D is a collection of graphs. To speed up subgraph query answering on graph databases, indexes are commonly used. State-of-the-art graph database indexes do not adapt or scale well to dynamic graph database use; they are static, and their ability to prune possible search responses to meet user needs worsens over time as databases change and grow. Users can re-mine indexes to gain some improvement, but it is time consuming. Users must also tune numerous parameters on an ongoing basis to optimize performance and can inadvertently worsen the query response time if they do not choose parameters wisely. Recently, a one-pass algorithm has been developed to enhance the performance of frequent subgraphs based indexes by using the algorithm to update them regularly. However, there are some drawbacks, most notably the need to make updates as the query workload changes.

Nowadays, detecting health-violating restaurants is a serious problem due to the limited number of health inspectors in a city as compared to the number of restaurants. Rarely inspectors are helped by formal complains, but many complaints are reported as reviews on social media such as Yelp. In this paper we propose new predictors to detect health-violating restaurants based on restaurant sub-area location, previous inspections history, Yelp reviews content, and Yelp users behavior. The resulting method outperforms past work, with a percentage of improvement in Cohen's kappa and Matthews correlation coefficient of at least 16%. In addition, we define a new method that directly evaluates the benefit of a classifier on the ability of an inspector in detecting health-violating restaurants. We show that our classification method really improves the ability of the inspector and outperforms previous solutions.

Wikipedia is based on the idea that anyone can make edits to the website in order to create reliable and crowd-sourced content. Yet with the cover of internet anonymity, some users make changes to the website that do not align with Wikipedia's intended uses. For this reason, Wikipedia allows for some pages of the website to become protected, where only certain users can make revisions to the page. This allows administrators to protect pages from vandalism, libel, and edit wars. However, with over five million pages on Wikipedia, it is impossible for administrators to monitor all pages and manually enforce page protection. In this paper we consider for the first time the problem of deciding whether a page should be protected or not in a collaborative environment such as Wikipedia. We formulate the problem as a binary classification task and propose a novel set of features to decide which pages to protect based on (i) users page revision behavior and (ii) page categories. We tested our system, called DePP, on a new dataset we built consisting of 13.6K pages (half protected and half unprotected) and 1.9M edits. Experimental results show that DePP reaches 93.24% classification accuracy and significantly improves over baselines.

Weighted signed networks (WSNs) are networks in which edges are labeled with positive and negative weights. WSNs can capture like/dislike, trust/distrust, and other social relationships between people. In this paper, we consider the problem of predicting the weights of edges in such networks. We propose two novel measures of node behavior: the goodness of a node intuitively captures how much this node is liked/trusted by other nodes, while the fairness of a node captures how fair the node is in rating other nodes' likeability or trust level. We provide axioms that these two notions need to satisfy and show that past work does not meet these requirements for WSNs. We provide a mutually recursive definition of these two concepts and prove that they converge to a unique solution in linear time. We use the two measures to predict the edge weight in WSNs. Furthermore, we show that when compared against several individual algorithms from both the signed and unsigned social network literature, our fairness and goodness metrics almost always have the best predictive power. We then use these as features in different multiple regression models and show that we can predict edge weights on 2 Bitcoin WSNs, an Epinions WSN, 2 WSNs derived from Wikipedia, and a WSN derived from Twitter with more accurate results than past work. Moreover, fairness and goodness metrics form the most significant feature for prediction in most (but not all) cases.

We study the problem of detecting vandals on Wikipedia before any human or known vandalism detection system reports flagging potential vandals so that such users can be presented early to Wikipedia administrators. We leverage multiple classical ML approaches, but develop 3 novel sets of features. Our Wikipedia Vandal Behavior (WVB) approach uses a novel set of user editing patterns as features to classify some users as vandals. Our Wikipedia Transition Probability Matrix (WTPM) approach uses a set of features derived from a transition probability matrix and then reduces it via a neural net auto-encoder to classify some users as vandals. The VEWS approach merges the previous two approaches. Without using any information (e.g. reverts) provided by other users, these algorithms each have over 85% classification accuracy. Moreover, when temporal recency is considered, accuracy goes to almost 90%. We carry out detailed experiments on a new data set we have created consisting of about 33K Wikipedia users (including both a black list and a white list of editors) and containing 770K edits. We describe specific behaviors that distinguish between vandals and non-vandals. We show that VEWS beats ClueBot NG and STiki, the best known algorithms today for vandalism detection. Moreover, VEWS detects far more vandals than ClueBot NG and on average, detects them 2.39 edits before ClueBot NG when both detect the vandal. However, we show that the combination of VEWS and ClueBot NG can give a fully automated vandal early warning system with even higher accuracy.

The aim of this paper is to present more general criteria and techniques for chase termination. We first present extensions of the well-known stratification criterion and introduce a new criterion, called local stratification, which generalizes both super-weak acyclicity and stratification-based criteria (including the class of constraints which are inductively restricted). Next, the paper presents a rewriting algorithm transforming the original set of constraints Σ into an “equivalent” set Σαand verifying the structural properties for chase termination on Σα. The rewriting of constraints allows us to recognize larger classes of constraints for which chase termination is guaranteed. In particular, we show that if Σ satisfies chase termination conditions T, then the rewritten set Σαsatisfies T as well, but the vice versa is not true, that is there are significant classes of constraints for which Σαsatisfies T and Σ does not. A more general rewriting algorithm producing as output an equivalent set of dependencies and a Boolean value stating whether a sort of cyclicity has been detected is also proposed. The new rewriting technique and the checking of acyclicity allow us to introduce the class of acyclic constraints, which generalizes local stratification and guarantees that all chase sequences are finite with a length polynomial in the size of the input database.

In recent years, probabilistic data management has received a lot of attention due to several applications that deal with uncertain data: RFID systems, sensor networks, data cleaning, scientific and biomedical data management, and approximate schema mappings. Query evaluation is a challenging problem in probabilistic databases, proved to be #P-hard. A general method for query evaluation is based on the lineage of the query and reduces the query evaluation problem to computing the probability of a propositional formula. The main approaches proposed in the literature to approximate probabilistic queries confidence computation are based on Monte Carlo simulation, or formula compilation into decision diagrams (e.g., d-trees). The former executes a polynomial, but with too many, iterations, while the latter is polynomial for easy queries, but may be exponential in the worst case. We designed a new optimized Monte Carlo algorithm that drastically reduces the number of iterations and proposed an efficient parallel version that we implemented on GPU. Thanks to the elevated degree of parallelism provided by the GPU, combined with the linear speedup of our algorithm, we managed to reduce significantly the long running time required by a sequential Monte Carlo algorithm. Experimental results show that our algorithm is so efficient as to be comparable with the formula compilation approach, but with the significant advantage of avoiding exponential behavior.

Online social networks like Slashdot bring valuable information to millions of users - but their accuracy is based on the integrity of their user base. Unfortunately, there are many “trolls” on Slashdot who post misinformation and compromise system integrity. In this paper, we develop a general algorithm called TIA (short for Troll Identification Algorithm) to classify users of an online “signed” social network as malicious (e.g. trolls on Slashdot) or benign (i.e. normal honest users). Though applicable to many signed social networks, TIA has been tested on troll detection on Slashdot Zoo under a wide variety of parameter settings. Its running time is faster than many past algorithms and it is significantly more accurate than existing methods.

The South Asian Violent Extremist Group (SAVE) dataset describes 500 interactions between 30 South Asian terrorist groups over a 20-year period. We analyze 4 types of interactions between these groups (financial, logistical, operational, and political) via 4 network-theoretic techniques: spectral decomposition and Simmelian ties in an undirected version of the network, and PageRank and betweenness centrality on directed versions of the network. We identify the major players in both the provision and the flows through this network for each type of support. Our analysis shows that the five entities most involved in providing and facilitating all forms of support are Al Qaeda, the Pakistani Inter-Services Intelligence Agency, the Taliban, Lashkar-e-Taiba, and Tehreek-e-Taliban Pakistan.

This paper focuses primarily on the Person Successor Problem (PSP): when a terrorist is removed from a terrorist network, who is most likely to take his place? We leverage the solution to PSP to predict a new terrorist network after removal of a set of terrorists and to answer the question: which set of k (k > 0) terrorists should be removed in order to minimize the lethality of the terrorist network? We propose a theoretical model to study these questions taking into account the fact that terrorists may have different individual capabilities. We develop an algorithm for PSP in which analysts can specify the conditions an individual needs to satisfy in order to replace another person. We test the correctness of our algorithm on a real-world partial network dataset for two terrorist groups: Al-Qaeda and Lashkar-e-Taiba where we have ground truth about who replaced who, as well as a synthetic dataset where experts estimate who replaced who. Building on the solution to PSP, we develop an algorithm to identify which set of k people to remove from a terrorist network to minimize the organization's efficiency (formalized as an objective function in some different ways).

Several database areas such as data exchange and integration share the problem of fixing database instance violations with respect to a set of constraints. The chase algorithm solves such violations by inserting tuples and setting the value of nulls. Unfortunately, the chase algorithm may not terminate and the problem of deciding whether the chase process terminates is undecidable. Recently there has been an increasing interest in the identification of sufficient structural properties of constraints which guarantee that the chase algorithm terminates [8, 10, 14, 15]. In this paper we propose an original technique which allows to improve current conditions detecting chase termination. Our proposal consists in rewriting the original set of constraints Σ into an 'equivalent' set Σ α and verifying the structural properties for chase termination on Σ α . The rewriting of constraints allows to recognize larger classes of constraints for which chase termination is guaranteed. In particular, we show that if Σ satisfies chase termination conditions T, then the rewritten set Σ α satisfies T as well, but the vice versa is not true, that is there are significant classes of constraints for which Σ α satisfies T and Σ does not.