Runa Bhaumik

Toward trustworthy recommender systems: An analysis of attack models and algorithm robustness

Publicly accessible adaptive systems such as collaborative recommender systems present a security problem. Attackers, who cannot be readily distinguished from ordinary users, may inject biased profiles in an attempt to force a system to “adapt” in a manner advantageous to them. Such attacks may lead to a degradation of user trust in the objectivity and accuracy of the system. Recent research has begun to examine the vulnerabilities and robustness of different collaborative recommendation techniques in the face of “profile injection” attacks. In this article, we outline some of the major issues in building secure recommender systems, concentrating in particular on the modeling of attacks and their impact on various recommendation algorithms. We introduce several new attack models and perform extensive simulation-based evaluations to show which attacks are most successful and practical against common recommendation techniques. Our study shows that both user-based and item-based algorithms are highly vulnerable to specific attack models, but that hybrid algorithms may provide a higher degree of robustness. Using our formal characterization of attack models, we also introduce a novel classification-based approach for detecting attack profiles and evaluate its effectiveness in neutralizing attacks.

Classification features for attack detection in collaborative recommender systems

Collaborative recommender systems are highly vulnerable to attack. Attackers can use automated means to inject a large number of biased profiles into such a system, resulting in recommendations that favor or disfavor given items. Since collaborative recommender systems must be open to user input, it is difficult to design a system that cannot be so attacked. Researchers studying robust recommendation have therefore begun to identify types of attacks and study mechanisms for recognizing and defeating them. In this paper, we propose and study different attributes derived from user profiles for their utility in attack detection. We show that a machine learning classification approach that includes attributes derived from attack models is more successful than more generalized detection algorithms previously studied.

Analysis and detection of segment-focused attacks against collaborative recommendation

Significant vulnerabilities have recently been identified in collaborative filtering recommender systems. These vulnerabilities mostly emanate from the open nature of such systems and their reliance on user-specified judgments for building profiles. Attackers can easily introduce biased data in an attempt to force the system to “adapt” in a manner advantageous to them. Our research in secure personalization is examining a range of attack models, from the simple to the complex, and a variety of recommendation techniques. In this chapter, we explore an attack model that focuses on a subset of users with similar tastes and show that such an attack can be highly successful against both user-based and item-based collaborative filtering. We also introduce a detection model that can significantly decrease the impact of this attack.

Attacks and Remedies in Collaborative Recommendation

Collaborative-filtering recommender systems are an electronic extension of everyday social recommendation behavior: people share opinions and decide whether or not to act on the basis of what they hear. Collaborative filtering lets you scale such interactions to groups of thousands or even millions. Publicly accessible user-adaptive systems such as collaborative recommender systems introduce security issues that must be solved if users are to perceive these systems as objective, unbiased, and accurate.

Toll-like receptors in the depressed and suicide brain

Abnormalities of the immune function in depression and suicide are based in part on the observation of increased levels of proinflammatory cytokines in the serum and postmortem brain of depressed and suicidal patients. We have examined if abnormalities of the innate immune receptors, known as Toll-like receptors (TLRs), in the brain are associated with depression and suicide, since the activation of these receptors results in production of cytokines. Of all the TLRs shown to be present in humans, TLR3 and TLR4 appear to be unique and important in brain function. We have determined the protein (by ELISA method) and mRNA expression (using qPCR) of TLR3 and TLR4 in the postmortem brain (dorsolateral prefrontal cortex [DLPFC]) of 22 depressed suicide victims, 11 non-depressed suicide victims, 12 depressed non-suicide subjects and 20 normal control subjects. We found that the mRNA expression of TLR3 and TLR4 was significantly increased in DLPFC of depressed suicide victims and in depressed non-suicide subjects, compared with controls. However, the protein expression of TLR3 and TLR4 was significantly increased in depressed suicide victims, but not in depressed non-suicide subjects compared with controls. The observed abnormalities of proinflammatory cytokines in the brain of suicide victims may be related to an abnormality of TLR3 and TLR4 over-expression. To our knowledge, this is the first study of TLRs in the brain of psychiatric subjects.

Detecting Profile Injection Attacks in Collaborative Recommender Systems

Collaborative recommender systems are known to be highly vulnerable to profile injection attacks, attacks that involve the insertion of biased profiles into the ratings database for the purpose of altering the systems recommendation behavior. In prior work, we and others have identified a number of models for such attacks and shown their effectiveness. This paper describes a classification approach to the problem of detecting and responding to profile injection attacks. This technique significantly reduces the effectiveness of the most powerful attack models previously studied

Detecting profile injection attacks in collaborative filtering: a classification-based approach

Collaborative recommender systems have been shown to be vulnerable to profile injection attacks. By injecting a large number of biased profiles into a system, attackers can manipulate the predictions of targeted items. To decrease this risk, researchers have begun to study mechanisms for detecting and preventing profile injection attacks. In prior work, we proposed several attributes for attack detection and have shown that a classifier built with them can be highly successful at identifying attack profiles. In this paper, we extend our work through a more detailed analysis of the information gain associated with these attributes across the dimensions of attack type and profile size. We then evaluate their combined effectiveness at improving the robustness of user based recommender systems.

Multidimensional prediction of treatment response to antidepressants with cognitive control and functional MRI.

Predicting treatment response for major depressive disorder can provide a tremendous benefit for our overstretched health care system by reducing number of treatments and time to remission, thereby decreasing morbidity. The present study used neural and performance predictors during a cognitive control task to predict treatment response (% change in Hamilton Depression Rating Scale pre- to post-treatment). Forty-nine individuals diagnosed with major depressive disorder were enrolled with intent to treat in the open-label study; 36 completed treatment, had useable data, and were included in most data analyses. Participants included in the data analysis sample received treatment with escitalopram (n = 22) or duloxetine (n = 14) for 10 weeks. Functional MRI and performance during a Parametric Go/No-go test were used to predict per cent reduction in Hamilton Depression Rating Scale scores after treatment. Haemodynamic response function-based contrasts and task-related independent components analysis (subset of sample: n = 29) were predictors. Independent components analysis component beta weights and haemodynamic response function modelling activation during Commission errors in the rostral and dorsal anterior cingulate, mid-cingulate, dorsomedial prefrontal cortex, and lateral orbital frontal cortex predicted treatment response. In addition, more commission errors on the task predicted better treatment response. Together in a regression model, independent component analysis, haemodynamic response function-modelled, and performance measures predicted treatment response with 90% accuracy (compared to 74% accuracy with clinical features alone), with 84% accuracy in 5-fold, leave-one-out cross-validation. Convergence between performance markers and functional magnetic resonance imaging, including novel independent component analysis techniques, achieved high accuracy in prediction of treatment response for major depressive disorder. The strong link to a task paradigm provided by use of independent component analysis is a potential breakthrough that can inform ways in which prediction models can be integrated for use in clinical and experimental medicine studies.

Abnormal gene expression of proinflammatory cytokines and their membrane-bound receptors in the lymphocytes of depressed patients

Abnormalities of protein levels of proinflammatory cytokines and their soluble receptors have been reported in plasma of depressed patients. In this study, we examined the role of cytokines and their membrane-bound receptors in major depressive disorder (MDD). We determined the protein and mRNA expression of proinflammatory cytokines, interleukin (IL)-1β, IL-6, tumor necrosis factor (TNF)-α, and mRNA expression of their membrane-bound receptors in the lymphocytes from 31 hospitalized MDD patients and 30 non-hospitalized normal control (NC) subjects. The subjects were diagnosed according to DSM-IV criteria. Protein levels of cytokines were determined by ELISA, and mRNA levels in lymphocytes were determined by the qPCR method. We found that the mean mRNA levels of the proinflammatory cytokines IL-1β, IL-6, TNF-α, their receptors, TNFR1, TNFR2, IL-1R1 and the antagonist IL-1RA were significantly increased in the lymphocytes of MDD patients compared with NC. No significant differences in the lymphocyte mRNA levels of IL-1R2, IL-6R, and Gp130 were observed between MDD patients and NC. These studies suggest abnormal gene expression of these cytokines and their membrane-bound receptors in the lymphocytes of MDD patients, and that their mRNA expression levels in the lymphocytes could be a useful biomarker for depression.

Redesigning community mental health services for urban children: Supporting schooling to promote mental health

OBJECTIVE This study examined a school- and home-based mental health service model, Links to Learning, focused on empirical predictors of learning as primary goals for services in high-poverty urban communities. METHOD Teacher key opinion leaders were identified through sociometric surveys and trained, with mental health providers and parent advocates, on evidence-based practices to enhance childrens learning. Teacher key opinion leaders and mental health providers cofacilitated professional development sessions for classroom teachers to disseminate 2 universal (Good Behavior Game, peer-assisted learning) and 2 targeted (Good News Notes, Daily Report Card) interventions. Group-based and home-based family education and support were delivered by mental health providers and parent advocates for children in kindergarten through 4th grade diagnosed with 1 or more disruptive behavior disorders. Services were Medicaid-funded through 4 social service agencies (N = 17 providers) in 7 schools (N = 136 teachers, 171 children) in a 2 (Links to Learning vs. services as usual) × 6 (pre- and posttests for 3 years) longitudinal design with random assignment of schools to conditions. Services as usual consisted of supported referral to a nearby social service agency. RESULTS Mixed effects regression models indicated significant positive effects of Links to Learning on mental health service use, classroom observations of academic engagement, teacher report of academic competence and social skills, and parent report of social skills. Nonsignificant between-groups effects were found on teacher and parent report of problem behaviors, daily hassles, and curriculum-based measures. Effects were strongest for young children, girls, and children with fewer symptoms. CONCLUSION Community mental health services targeting empirical predictors of learning can improve school and home behavior for children living in high-poverty urban communities.