Twitter has become a major instrument for the rapid dissemination and subsequent debate of news stories, and comprehensive methodologies for systematic research into news discussion on Twitter are beginning to emerge. This paper outlines innovative approaches for large-scale quantitative research into how Twitter is used to discuss and cover the news, focusing especially on #hashtags: brief identifiers which mark a tweet as taking part in an established discussion.

Researching news discussion on Twitter : New methodologies

The widespread dissemination of misinformation in social media has recently received a lot of attention in academia. While the problem of misinformation in social media has been intensively studied, there are seemingly different definitions for the same problem, and inconsistent results in different studies. In this survey, we aim to consolidate the observations, and investigate how an optimal method can be selected given specific conditions and contexts. To this end, we first introduce a definition for misinformation in social media and we examine the difference between misinformation detection and classic supervised learning. Second, we describe the diffusion of misinformation and introduce how spreaders propagate misinformation in social networks. Third, we explain characteristics of individual methods of misinformation detection, and provide commentary on their advantages and pitfalls. By reflecting applicability of different methods, we hope to enable the intensive research in this area to be conveniently reused in real-world applications and open up potential directions for future studies.

Misinformation in Social Media: Definition, Manipulation, and Detection

Graph Neural Networks (GNNs) have been widely applied to fraud detection problems in recent years, revealing the suspiciousness of nodes by aggregating their neighborhood information via different relations However, few prior works have noticed the camouflage behavior of fraudsters, which could hamper the performance of GNN-based fraud detectors during the aggregation process In this paper, we introduce two types of camouflages based on recent empirical studies, ie, the feature camouflage and the relation camouflage Existing GNNs have not addressed these two camouflages, which results in their poor performance in fraud detection problems Alternatively, we propose a new model named CAmouflage-REsistant GNN (CARE-GNN), to enhance the GNN aggregation process with three unique modules against camouflages Concretely, we first devise a label-aware similarity measure to find informative neighboring nodes Then, we leverage reinforcement learning (RL) to find the optimal amounts of neighbors to be selected Finally, the selected neighbors across different relations are aggregated together Comprehensive experiments on two real-world fraud datasets demonstrate the effectiveness of the RL algorithm The proposed CARE-GNN also outperforms state-of-the-art GNNs and GNN-based fraud detectors We integrate all GNN-based fraud detectors as an opensource toolbox: this https URL The CARE-GNN code and datasets are available at this https URL

Enhancing Graph Neural Network-based Fraud Detectors against Camouflaged Fraudsters

We consider the problem of securely computing the Greater Than (GT) predicate and its generalization - securely determining membership in a union of intervals. We approach these problems from the point of view of Q-Conditional Oblivious Transfer (Q-COT), introduced by Di Crescenzo, Ostrovsky and Rajagopalan [4]. Q-COT is an oblivious transfer that occurs iff predicate Q evaluates to true on the parties' inputs. We are working in the semi-honest model with computationally unbounded receiver. In this paper, we propose: (i) a stronger, simple and intuitive definition of COT, which we call strong COT, or Q-SCOT. (ii) A simpler and more efficient one-round protocol for securely computing GT and GT-SCOT. (iii) A simple and efficient modular construction reducing SCOT based on membership in a union of intervals (UI-SCOT) to GT-SCOT, producing an efficient one-round UI-SCOT.

Strong conditional Oblivious Transfer and computing on intervals

Online review system enables users to submit reviews about the products. However, the openness of Internet and monetary rewards for crowdsourcing tasks stimulate a large number of fraudulent users to write fake reviews and post advertisements to interfere the rank of apps. Existing methods for detecting spam reviews have been successful but they usually aims at e-commerce (e.g. Amazon, eBay) and recommendation (e.g. Yelp, Dianping) systems. Since the behaviors of fraudulent users are complexity and varying across different review platforms, existing methods are not suitable for fraudster detection in online app review system. To shed light on this question, we are among the first to analyze the intentions of fraudulent users from different review platforms and categorize them by utilizing characteristics of contents (similarity, special symbols) and behaviors (timestamps, device, login status). With a comprehensive analysis of spamming activities and relationships between normal and malicious users, we design and present FdGars, the first graph convolutional network approach for fraudster detection in online app review system. Then we evaluate FdGars on real-world large-scale dataset (with 82,542 nodes and 42,433,134 edges) from Tencent App Store. The result demonstrates that F1-score of FdGars can achieve 0.938+, which outperforms several baselines and state-of-art fraudsters detecting methods. Moreover, we deploy FdGars on Tencent Beacon Anti-Fraud Platform to show its effectiveness and scalability. To the best of our knowledge, this is the first work to use graph convolutional networks for fraudster detection in the large-scale online app review system. It is worth to mention that FdGars can uncover malicious accounts even the data lack of labels in anti-spam tasks.

FdGars: Fraudster Detection via Graph Convolutional Networks in Online App Review System

We propose a system called TwoFace to uncover crowdsourced review manipulators who target online review systems. A unique feature of TwoFace is its three-phase framework:(i) in the first phase, we intelligently sample actual evidence of manipulation(e.g., review manipulators) by exploiting low moderation crowdsourcing platforms that reveal evidence of strategic manipulation;(ii) we then propagate the suspiciousness of these seed users to identify similar users through a random walk over a "suspiciousness»» graph; and(iii) finally, we uncover(hidden) distant users who serve structurally similar roles by mapping users into a low-dimensional embedding space that captures community structure. Altogether, the TwoFace system recovers 83% to 93% of all manipulators in a sample from Amazon of 38,590 reviewers, even when the system is seeded with only a few samples from malicious crowdsourcing sites.

https://dl.acm.org/doi/pdf/10.1145/3159652.3159726

Combating Crowdsourced Review Manipulators: A Neighborhood-Based Approach

Privacy issues in intrusion detection systems: A taxonomy, survey and future directions

We exploit the prevalence of malicious review writers on crowdsourcing platforms like RapidWorkers to identify actual fraud reviews on Amazon. Complementary to previous efforts which often rely on proxies for fraud reviews, we present a long-term study of actual fraudulent behaviors in online review manipulation. We find that these malicious reviewers – though often providing seemingly legitimate opinions – do exhibit significant differences from normal reviewers in terms of ratings distribution, length of the reviews, and the burstiness of the reviews themselves. We additionally study the evolution of these reviews, and find striking temporal changes that could support future discovery of these reviewers who may be “hiding in plain sight”.

Behavioral Analysis of Review Fraud: Linking Malicious Crowdsourcing to Amazon and Beyond.

The insider threat remains one of the most serious challenges to computer security. An insider attack occurs when an authorized user misuses his privileges and causes damages to the organization. Deception techniques have served as a common solution to insider threat detection, and several techniques, such as approaches based on honey entities, have been proposed. On the other hand, access control systems lack the ability to detect insider threats. In this paper, we focus on integrating deception into the role-based access control (RBAC) model, which is one of the most widely used access control models. We introduce the notion of “honey permission” and use it to extend RBAC to help in insider threat detection. We define honey permissions as permissions that exceed the authorized access, and are assigned to a subset of roles known as “candidate roles”. Objects included in honey permissions are fake versions of sensitive objects that are enticing for malicious users. In this way, an attempt to access sensitive resources by unauthorized users would be detected. We extend the RBAC model by adding honey permissions, indicating candidate roles, and adding a monitoring unit which monitors the sessions in which the owners of the sessions activate a subset of candidate roles and have access to an object through a honey permission. We propose an algorithm to select candidate roles and assign honey permissions to them. Furthermore, we provide security analysis and consider the overhead that would be added to the RBAC system for evaluation.

Toward an Insider Threat Detection Framework Using Honey Permissions.

User reviews have become a cornerstone of how we make decisions. However, this user-based feedback is susceptible to manipulation as recent research has shown the feasibility of automatically generating fake reviews. Previous investigations, however, have focused on generative fake review approaches that are (i) domain dependent and not extendable to other domains without replicating the whole process from scratch; and (ii) character-level based known to generate reviews of poor quality that are easily detectable by anti-spam detectors and by end users. In this work, we propose and evaluate a new class of attacks on online review platforms based on neural language models at word-level granularity in an inductive transfer-learning framework wherein a universal model is refined to handle domain shift, leading to potentially wide-ranging attacks on review systems. Through extensive evaluation, we show that such model-generated reviews can bypass powerful anti-spam detectors and fool end users. Paired with this troubling attack vector, we propose a new defense mechanism that exploits the distributed representation of these reviews to detect model-generated reviews. We conclude that despite the success of neural models in generating realistic reviews, our proposed RNN-based discriminator can combat this type of attack effectively (90% accuracy).

https://dl.acm.org/doi/pdf/10.1145/3357384.3358034

Parisa Kaghazgaran

Papers

Combating Crowdsourced Review Manipulators: A Neighborhood-Based Approach

Privacy issues in intrusion detection systems: A taxonomy, survey and future directions

Behavioral Analysis of Review Fraud: Linking Malicious Crowdsourcing to Amazon and Beyond.

Toward an Insider Threat Detection Framework Using Honey Permissions.

Wide-Ranging Review Manipulation Attacks: Model, Empirical Study, and Countermeasures