arXiv subjects

Anna Squicciarini

Publications and source records attributed to Anna Squicciarini.

18 recordsLinked to original sources

arXiv2026

Losing One's Story: How Vulnerable Users Experience Harm in Online Support Seeking

Online support communities are a critical resource for individuals facing distress, stigma, and limited offline support. However, these spaces are not uniformly supportive, particularly for users with little margin for error. In this paper, we examine how vulnerable users experience harm within online support-seeking interactions. Through 25 semi-structured interviews with Reddit users, we show that support seeking is often a constrained practice shaped by structural vulnerability. We introduce the concept of \emph{narrative harm} to describe how participants experience harm through losses of narrative authority, coherence, and space. Their personal disclosures are questioned, reframed, or displaced by others, reflecting asymmetries in voice, credibility, and interpretive power embedded in platform dynamics and moderation regimes. In response, users engage in defensive strategies, including selective self-disclosure, self-silencing, identity separation, and informal mutual aid, shifting the burden of safety onto those already most vulnerable. Our findings highlight how current platform designs insufficiently account for situated vulnerability and redistribute discursive power away from support seekers. We discuss implications for the design of online support systems that better preserve narrative integrity and reduce the burden of safety labor.

cs.HC↗

arXiv2026

Aggregation-Aware Synthetic Text Generation Against Authorship Re-Identification

Online users often release multiple texts under the same identity, giving attackers an author profile that can reveal more than any single text. Existing authorship obfuscation methods optimize privacy independently for each document, leaving them blind to cross-document correlations that make aggregation dangerous. We propose Aggregation-Aware Synthetic Text Generation (AAST), a framework that addresses this gap by jointly selecting synthetic texts at the bundle level rather than optimizing each text in isolation. AAST targets attribution and verification attacks, including cross-genre settings where attacker references come from a genre not observed during generation or selection. Experiments across same-genre, cross-genre, neural, and independent non-neural stylometric attacks show that AAST lowers account-level linkability as bundle size grows, while preserving semantic quality, linguistic acceptability, and sentiment alignment.

cs.AI↗

arXiv2024

The Task Shield: Enforcing Task Alignment to Defend Against Indirect Prompt Injection in LLM Agents

Large Language Model (LLM) agents are increasingly being deployed as conversational assistants capable of performing complex real-world tasks through tool integration. This enhanced ability to interact with external systems and process various data sources, while powerful, introduces significant security vulnerabilities. In particular, indirect prompt injection attacks pose a critical threat, where malicious instructions embedded within external data sources can manipulate agents to deviate from user intentions. While existing defenses based on rule constraints, source spotlighting, and authentication protocols show promise, they struggle to maintain robust security while preserving task functionality. We propose a novel and orthogonal perspective that reframes agent security from preventing harmful actions to ensuring task alignment, requiring every agent action to serve user objectives. Based on this insight, we develop Task Shield, a test-time defense mechanism that systematically verifies whether each instruction and tool call contributes to user-specified goals. Through experiments on the AgentDojo benchmark, we demonstrate that Task Shield reduces attack success rates (2.07\%) while maintaining high task utility (69.79\%) on GPT-4o.

cs.CR↗

arXiv2024

Time-Efficient Locally Relevant Geo-Location Privacy Protection

Geo-obfuscation serves as a location privacy protection mechanism (LPPM), enabling mobile users to share obfuscated locations with servers, rather than their exact locations. This method can protect users' location privacy when data breaches occur on the server side since the obfuscation process is irreversible. To reduce the utility loss caused by data obfuscation, linear programming (LP) is widely employed, which, however, might suffer from a polynomial explosion of decision variables, rendering it impractical in largescale geo-obfuscation applications. In this paper, we propose a new LPPM, called Locally Relevant Geo-obfuscation (LR-Geo), to optimize geo-obfuscation using LP in a time-efficient manner. This is achieved by confining the geo-obfuscation calculation for each user exclusively to the locally relevant (LR) locations to the user's actual location. Given the potential risk of LR locations disclosing a user's actual whereabouts, we enable users to compute the LP coefficients locally and upload them only to the server, rather than the LR locations. The server then solves the LP problem based on the received coefficients. Furthermore, we refine the LP framework by incorporating an exponential obfuscation mechanism to guarantee the indistinguishability of obfuscation distribution across multiple users. Based on the constraint structure of the LP formulation, we apply Benders' decomposition to further enhance computational efficiency. Our theoretical analysis confirms that, despite the geo-obfuscation being calculated independently for each user, it still meets geo-indistinguishability constraints across multiple users with high probability. Finally, the experimental results based on a real-world dataset demonstrate that LR-Geo outperforms existing geo-obfuscation methods in computational time, data utility, and privacy preservation.

cs.CR↗

arXiv2023

RoCourseNet: Distributionally Robust Training of a Prediction Aware Recourse Model

Counterfactual (CF) explanations for machine learning (ML) models are preferred by end-users, as they explain the predictions of ML models by providing a recourse (or contrastive) case to individuals who are adversely impacted by predicted outcomes. Existing CF explanation methods generate recourses under the assumption that the underlying target ML model remains stationary over time. However, due to commonly occurring distributional shifts in training data, ML models constantly get updated in practice, which might render previously generated recourses invalid and diminish end-users trust in our algorithmic framework. To address this problem, we propose RoCourseNet, a training framework that jointly optimizes predictions and recourses that are robust to future data shifts. This work contains four key contributions: (1) We formulate the robust recourse generation problem as a tri-level optimization problem which consists of two sub-problems: (i) a bi-level problem that finds the worst-case adversarial shift in the training data, and (ii) an outer minimization problem to generate robust recourses against this worst-case shift. (2) We leverage adversarial training to solve this tri-level optimization problem by: (i) proposing a novel virtual data shift (VDS) algorithm to find worst-case shifted ML models via explicitly considering the worst-case data shift in the training dataset, and (ii) a block-wise coordinate descent procedure to optimize for prediction and corresponding robust recourses. (3) We evaluate RoCourseNet's performance on three real-world datasets, and show that RoCourseNet consistently achieves more than 96% robust validity and outperforms state-of-the-art baselines by at least 10% in generating robust CF explanations. (4) Finally, we generalize the RoCourseNet framework to accommodate any parametric post-hoc methods for improving robust validity.

cs.LG↗

arXiv2023

Consensus in Complex Networks with Noisy Agents and Peer Pressure

In this paper we study a discrete time consensus model on a connected graph with monotonically increasing peer-pressure and noise perturbed outputs masking a hidden state. We assume that each agent maintains a constant hidden state and a presents a dynamic output that is perturbed by random noise drawn from a mean-zero distribution. We show consensus is ensured in the limit as time goes to infinity under certain assumptions on the increasing peer-pressure term and also show that the hidden state cannot be exactly recovered even when model dynamics and outputs are known. The exact nature of the distribution is computed for a simple two vertex graph and results found are shown to generalize (empirically) to more complex graph structures.

physics.soc-ph↗

arXiv2022

User Customizable and Robust Geo-Indistinguishability for Location Privacy

Location obfuscation functions generated by existing systems for ensuring location privacy are monolithic and do not allow users to customize their obfuscation range. This can lead to the user being mapped in undesirable locations (e.g., shady neighborhoods) to the location-requesting services. Modifying the obfuscation function generated by a centralized server on the user side can result in poor privacy as the original function is not robust against such updates. Users themselves might find it challenging to understand the parameters involved in obfuscation mechanisms (e.g., obfuscation range and granularity of location representation) and therefore struggle to set realistic trade-offs between privacy, utility, and customization. In this paper, we propose a new framework called, CORGI, i.e., CustOmizable Robust Geo-Indistinguishability, which generates location obfuscation functions that are robust against user customization while providing strong privacy guarantees based on the Geo-Indistinguishability paradigm. CORGI utilizes a tree representation of a given region to assist users in specifying their privacy and customization requirements. The server side of CORGI takes these requirements as inputs and generates an obfuscation function that satisfies Geo-Indistinguishability requirements and is robust against customization on the user side. The obfuscation function is returned to the user who can then choose to update the obfuscation function (e.g., obfuscation range, granularity of location representation). The experimental results on a real dataset demonstrate that CORGI can efficiently generate obfuscation matrices that are more robust to the customization by users.

cs.CR↗

arXiv2022

A Longitudinal Dataset of Twitter ISIS Users

We present a large longitudinal dataset of tweets from two sets of users that are suspected to be affiliated with ISIS. These sets of users are identified based on a prior study and a campaign aimed at shutting down ISIS Twitter accounts. These users have engaged with known ISIS accounts at least once during 2014-2015 and are still active as of 2021. Some of them have directly supported the ISIS users and their tweets by retweeting them, and some of the users that have quoted tweets of ISIS, have uncertain connections to ISIS seed accounts. This study and the dataset represent a unique approach to analyzing ISIS data. Although much research exists on ISIS online activities, few studies have focused on individual accounts. Our approach to validating accounts as well as developing a framework for differentiating accounts' functionality (e.g., propaganda versus operational planning) offers a foundation for future research. We perform some descriptive statistics and preliminary analyses on our collected data to provide deeper insight and highlight the significance and practicality of such analyses. We further discuss several cross-disciplinary potential use cases and research directions.

cs.SI↗

arXiv2022

Automated Detection of Doxing on Twitter

Doxing refers to the practice of disclosing sensitive personal information about a person without their consent. This form of cyberbullying is an unpleasant and sometimes dangerous phenomenon for online social networks. Although prior work exists on automated identification of other types of cyberbullying, a need exists for methods capable of detecting doxing on Twitter specifically. We propose and evaluate a set of approaches for automatically detecting second- and third-party disclosures on Twitter of sensitive private information, a subset of which constitutes doxing. We summarize our findings of common intentions behind doxing episodes and compare nine different approaches for automated detection based on string-matching and one-hot encoded heuristics, as well as word and contextualized string embedding representations of tweets. We identify an approach providing 96.86% accuracy and 97.37% recall using contextualized string embeddings and conclude by discussing the practicality of our proposed methods.

cs.SI↗

arXiv2021

A Synthetic Prediction Market for Estimating Confidence in Published Work

Explainably estimating confidence in published scholarly work offers opportunity for faster and more robust scientific progress. We develop a synthetic prediction market to assess the credibility of published claims in the social and behavioral sciences literature. We demonstrate our system and detail our findings using a collection of known replication projects. We suggest that this work lays the foundation for a research agenda that creatively uses AI for peer review.

cs.CY↗

arXiv2020

Privacy in Crisis: A study of self-disclosure during the Coronavirus pandemic

We study observed incidence of self-disclosure in a large dataset of Tweets representing user-led English-language conversation about the Coronavirus pandemic. Using an unsupervised approach to detect voluntary disclosure of personal information, we provide early evidence that situational factors surrounding the Coronavirus pandemic may impact individuals' privacy calculus. Text analyses reveal topical shift toward supportiveness and support-seeking in self-disclosing conversation on Twitter. We run a comparable analysis of Tweets from Hurricane Harvey to provide context for observed effects and suggest opportunities for further study.

cs.SI↗

arXiv2019

Power Law Public Goods Game for Personal Information Sharing in News Commentaries

We propose a public goods game model of user sharing in an online commenting forum. In particular, we assume that users who share personal information incur an information cost but reap the benefits of a more extensive social interaction. Freeloaders benefit from the same social interaction but do not share personal information. The resulting public goods structure is analyzed both theoretically and empirically. In particular, we show that the proposed game always possesses equilibria and we give sufficient conditions for pure strategy equilibria to emerge. These correspond to users who always behave the same way, either sharing or hiding personal information. We present an empirical analysis of a relevant data set, showing that our model parameters can be fit and that the proposed model has better explanatory power than a corresponding null (linear) model of behavior.

cs.GT↗

arXiv2019

Consensus and Information Cascades in Game-Theoretic Imitation Dynamics with Static and Dynamic Network Topologies

We construct a model of strategic imitation in an arbitrary network of players who interact through an additive game. Assuming a discrete time update, we show a condition under which the resulting difference equations converge to consensus. Two conjectures on general convergence are also discussed. We then consider the case where players not only may choose their strategies, but also affect their local topology. We show that for prisoner's dilemma, the graph structure converges to a set of disconnected cliques and strategic consensus occurs in each clique. Several examples from various matrix games are provided. A variation of the model is then used to create a simple model for the spreading of trends, or information cascades in (e.g., social) networks. We provide theoretical and empirical results on the trend-spreading model.

math.DS↗

arXiv2018

Backdoor Embedding in Convolutional Neural Network Models via Invisible Perturbation

Deep learning models have consistently outperformed traditional machine learning models in various classification tasks, including image classification. As such, they have become increasingly prevalent in many real world applications including those where security is of great concern. Such popularity, however, may attract attackers to exploit the vulnerabilities of the deployed deep learning models and launch attacks against security-sensitive applications. In this paper, we focus on a specific type of data poisoning attack, which we refer to as a {\em backdoor injection attack}. The main goal of the adversary performing such attack is to generate and inject a backdoor into a deep learning model that can be triggered to recognize certain embedded patterns with a target label of the attacker's choice. Additionally, a backdoor injection attack should occur in a stealthy manner, without undermining the efficacy of the victim model. Specifically, we propose two approaches for generating a backdoor that is hardly perceptible yet effective in poisoning the model. We consider two attack settings, with backdoor injection carried out either before model training or during model updating. We carry out extensive experimental evaluations under various assumptions on the adversary model, and demonstrate that such attacks can be effective and achieve a high attack success rate (above $90\%$) at a small cost of model accuracy loss (below $1\%$) with a small injection rate (around $1\%$), even under the weakest assumption wherein the adversary has no knowledge either of the original training data or the classifier model.

cs.CR↗

arXiv2017

Increasing Peer Pressure on any Connected Graph Leads to Consensus

In this paper, we study a model of opinion dynamics in a social network in the presence increasing interpersonal influence, i.e., increasing peer pressure. Each agent in the social network has a distinct social stress function given by a weighted sum of internal and external behavioral pressures. We assume a weighted average update rule and prove conditions under which a connected group of agents converge to a fixed opinion distribution, and under which conditions the group reaches consensus. We show that the update rule is a gradient descent and explain its transient and asymptotic convergence properties. Through simulation, we study the rate of convergence on a scale-free network and then validate the assumption of increasing peer pressure in a simple empirical model.

cs.SI↗

arXiv2014

A cooperate-defect model for the spread of deviant behavior in social networks

We present a game-theoretic model for the spread of deviant behavior in online social networks. We utilize a two-strategy framework wherein each player's behavior is classified as normal or deviant and evolves according to the cooperate-defect payoff scheme of the classic prisoner's dilemma game. We demonstrate convergence of individual behavior over time to a final strategy vector and indicate counterexamples to this convergence outside the context of prisoner's dilemma. Theoretical results are validated on a real-world dataset collected from a popular online forum.

cs.GT↗

arXiv2012

Two Species Evolutionary Game Model of User and Moderator Dynamics

We construct a two species evolutionary game model of an online society consisting of ordinary users and behavior enforcers (moderators). Among themselves, moderators play a coordination game choosing between being "positive" or "negative" (or harsh) while ordinary users play prisoner's dilemma. When interacting, moderators motivate good behavior (cooperation) among the users through punitive actions while the moderators themselves are encouraged or discouraged in their strategic choice by these interactions. We show the following results: (i) We show that the $ω$-limit set of the proposed system is sensitive both to the degree of punishment and the proportion of moderators in closed form. (ii) We demonstrate that the basin of attraction for the Pareto optimal strategy $(\text{Cooperate},\text{Positive})$ can be computed exactly. (iii) We demonstrate that for certain initial conditions the system is self-regulating. These results partially explain the stability of many online users communities such as Reddit. We illustrate our results with examples from this online system.

cs.GT↗

arXiv2012

An Informed Model of Personal Information Release in Social Networking Sites

The emergence of online social networks and the growing popularity of digital communication has resulted in an increasingly amount of information about individuals available on the Internet. Social network users are given the freedom to create complex digital identities, and enrich them with truthful or even fake personal information. However, this freedom has led to serious security and privacy incidents, due to the role users' identities play in establishing social and privacy settings. In this paper, we take a step toward a better understanding of online information exposure. Based on the detailed analysis of a sample of real-world data, we develop a deception model for online users. The model uses a game theoretic approach to characterizing a user's willingness to release, withhold or lie about information depending on the behavior of individuals within the user's circle of friends. In the model, we take into account both the heterogeneous nature of users and their different attitudes, as well as the different types of information they may expose online.

cs.SI↗