Shilling Attack Detection Research Articles

Enhancing Recommendation Systems with Skew Deviation Bias for Shilling Attack Detection

Introduction: Recommender systems serve as a powerful tool to address the challenges of information overload by delivering personalized recommendations. However, their susceptibility to profile injection or shilling attacks poses a significant threat. Malicious entities can introduce fabricated profiles into the database of users to manipulate the popularity of specific items, subsequently influencing prediction outcomes. Method: Detecting and mitigating the impact of such attacks is critical for preserving recommendation accuracy and user trust. The primary objective of this study is to develop an integrated framework for robust shilling attack detection and data sparsity mitigation in recommendation systems. This approach aims to make the system more resistant to manipulative attacks and improve recommendation quality, especially when dealing with limited data. In this paper, Skew Deviation Bias (SDB), is a novel metric that gauges the skewness within rating distributions, enabling the identification of both fabricated shilling profiles and the anomalous rating behaviors exhibited by attackers. Building upon this foundation, SDB is integrated with other statistical metrics like Rating deviation from the mean agreement (RDMA), Weighted deviation from the mean agreement (WDMA), Weighted degree of agreement (WDA), and length variance. This research investigates the impact of incorporating SDB alongside existing attributes in countering various attack scenarios, including random, average, and bandwagon attacks. Result: Extensive experiments are conducted to compare the effectiveness of SDB when integrated with existing attributes against scenarios employing only existing attributes. These experiments cover a range of attack sizes while maintaining a fixed 50% filler size. The results of thorough comparative analyses demonstrate the consistent superiority of the SDB-integrated approach, resulting in higher accuracy across all attack types compared to scenarios using only existing attributes. Notably, the random attack scenario shows the most significant accuracy improvement among the evaluated scenarios. Conclusion: The approach achieves a detection accuracy of 97.08% for random shilling attacks, affirming its robustness. Furthermore, in the context of data sparsity, the approach notably enhances recommendation quality.

Shilling Attack Detection with One Class Support Vector Machines

Öneri sistemleri çeşitli çevrimiçi platformlarda hayati bir rol oynar ve kullanıcıların tercihlerini göz önünde bulundurarak yeni ürünler, hizmetler ve içerikler keşfetmelerine yardımcı olur. Bununla birlikte, bu sistemler, kötü niyetli kullanıcıların derecelendirmeleri yapay olarak şişirdiği veya söndürdüğü ve önyargılı önerilere yol açtığı şilin saldırıları yoluyla manipülasyona karşı savunmasızdır. Bu saldırıları araştırmanın, anlamanın ve hafifletmenin önemini vurgulamak çok önemlidir. Bu tür saldırıları tespit etmek, tavsiye sistemlerinin bütünlüğünü ve etkinliğini korumak için çok önemlidir. Literatürde, şilin saldırılarını tespit etmek için birçok çalışma sunulmuştur. En iyi bilinen kümeleme yöntemleri farklı saldırı modelleri için uyarlanmıştır. Bu makalede, şilin saldırılarını tespit etmek için gürbüz bir teknik olarak Tek Sınıflı Destek Vektör Makineleri kullanımını araştırıyoruz. Tek Sınıflı Destek Vektör Makinaları, öncelikle anomali tespiti ve aykırılık tespiti görevleri için tasarlanmış geleneksel Destek Vektör Makinelerinin özel bir çeşididir. Önerilen yöntemi doğrulamak için MovieLens100K veri kümesi kullanılmıştır. Sonuç olarak, farklı boyut ve doluluk oranlı saldırılar için hassasiyet ve geri çağırma değerleri verilmiştir.

A novel classification‐based shilling attack detection approach for multi‐criteria recommender systems

AbstractRecommender systems are emerging techniques guiding individuals with provided referrals by considering their past rating behaviors. By collecting multi‐criteria preferences concentrating on distinguishing perspectives of the items, a new extension of traditional recommenders, multi‐criteria recommender systems reveal how much a user likes an item and why user likes it; thus, they can improve predictive accuracy. However, these systems might be more vulnerable to malicious attacks than traditional ones, as they expose multiple dimensions of user opinions on items. Attackers might try to inject fake profiles into these systems to skew the recommendation results in favor of some particular items or to bring the system into discredit. Although several methods exist to defend systems against such attacks for traditional recommenders, achieving robust systems by capturing shill profiles remains elusive for multi‐criteria rating‐based ones. Therefore, in this study, we first consider a prominent and novel attack type, that is, the power‐item attack model, and introduce its four distinct variants adapted for multi‐criteria data collections. Then, we propose a classification method detecting shill profiles based on various generic and model‐based user attributes, most of which are new features usually related to item popularity and distribution of rating values. The experiments conducted on three benchmark datasets conclude that the proposed method successfully detects attack profiles from genuine users even with a small selected size and attack size. The empirical outcomes also demonstrate that item popularity and user characteristics based on their rating profiles are highly beneficial features in capturing shilling attack profiles.

Hybrid gated recurrent unit and convolutional neural network-based deep learning mechanism for efficient shilling attack detection in social networks

The degree of openness of the socially aware recommendation systems and the possibility of the attackers injecting vast numbers of fake profiles biases the prediction of the system. Most classical shilling attack discovery mechanisms rely on artificially derived features, generally determined from user-generated data. Moreover, these baseline approaches cannot comprehensively capture the fine-grained relation between users and objects, thereby lagging in detection accuracy. This work introduces a Hybrid deep learning mechanism to address the Shilling Attack discovery with improved accuracy. The proposed mechanism highly enhances the detection accuracy by determining significant feature vectors managed by recurrent gated units and other high dimensional attribute vectors attained by using CNN. The GRU component is incorporated for the dimensionality transition process for existing weight classification data to learn the inherent significant features. The CNN mechanism is mainly used for analyzing the conditions of spatial-temporal data and transforming the data into a dormant feature vector. The experimental analysis of our proposed method is conducted using the datasets of Amazon, Netflix, and Movielens to demonstrate its predominance by enhancing accuracy, precision, and F-score.

Shilling Attack Detection in User Based Recommendation System

The majority of the existing unsupervised methods for detecting shilling attacks are based on user rating patterns, ignoring the differences in rating behavior between legitimate users and attack users. These methods have low accuracy in detecting different shilling attacks without having any prior knowledge of the attack types. We provide a novel unsupervised shilling assault detection technique based on an examination of user rating behavior in order to overcome these constraints. By first examining the deviation of rating tendencies on each item, we are able to determine the target item(s) and the accompanying goals of the attack users. Based on the results of this study, a group of suspicious users is then created. Second, we examine the users' rating behaviors in terms of their rating and interest preferences. Finally, using measurements of user rating behavior, we determine the suspicious degree and identify attack users within the collection of suspicious users. The Movie Lens 1M dataset, the sampled Amazon review dataset, and the Netflix dataset all show how good the suggested detection model.

A security detection approach based on autonomy-oriented user sensor in social recommendation network

User social network-based recommender system has achieved significant performance in current recommendation fields. However, the characteristic of openness brings great hidden dangers to the security of recommender systems. Shilling attackers can change the recommendations by foraging user relationships. Most shilling attack detection approaches depend on the explicit user historical data to locate shilling attackers. Some important features such as information propagation and social feedback of users in social networks have not been noticed. We propose a security detection method based on autonomy-oriented user sensor (AOUSD) to identify shilling attackers. Specifically, (1) the user is simulated as a social sensor with autonomous capabilities, (2) the user interaction model is built based on information propagation, information feedback and information disappearance mechanisms of social sensors, and a user dynamic knowledge graph is formed by considering the variable time function, (3) hierarchical clustering method is used to generate preliminary suspicious candidate groups and graph community detection clustering method is applied on the dynamic knowledge graph to detect the attackers. Then, AOUSD is first simulated on NetLogo and it is compared with other algorithms based on the Amazon data. The results prove the advantages of AOUSD in the efficiency and accuracy on shilling attack detection.

Generating A New Shilling Attack for Recommendation Systems

A collaborative filtering-based recommendation system has been an integral part of e-commerce and e-servicing. To keep the recommendation systems reliable, authentic, and superior, the security of these systems is very crucial. Though the existing shilling attack detection methods in collaborative filtering are able to detect the standard attacks, in this paper, we prove that they fail to detect a new or unknown attack. We develop a new attack model, named Obscure attack, with unknown features and observed that it has been successful in biasing the overall top-N list of the target users as intended. The Obscure attack is able to push target items to the top-N list as well as remove the actual rated items from the list. Our proposed attack is more effective at a smaller number of k in top-k similar user as compared to other existing attacks. The effectivity of the proposed attack model is tested on the MovieLens dataset, where various classifiers like SVM, J48, random forest, and naïve Bayes are utilized.

Shilling attack detection for collaborative recommender systems: a gradient boosting method.

Organized malicious shilling attackers influence the output of the collaborative filtering recommendation systems by inserting fake users into the rating matrix within the database. The existence of shilling attack poses a serious risk to the stability of the system. To counter this specific security threat, many attack detection methods are proposed. Some of the past methods suffer from two disadvantages, the first being that they only analyze the rating matrix from a single perspective of user rating values and ignore other perspectives. Another is that some methods only use a single classifier to handle the classification of malicious attackers. Considering the above disadvantages, this paper proposes a gradient boosting method (named XGB-SAD) to achieve attack detection by combining double-view and gradient boosting. We first analyze the rating matrix with a double-view of time and item, which in turn defines the TPUS collection. Then our method uses eXtreme Gradient Boosting to perform heuristic iterative optimization of the model's objective function and uses the idea of ensemble learning to integrate multiple sets of base classifiers into strong classifier. The integrated strong classifiers are used to complete the detection of malicious attackers. Finally, we perform several experiments and the results demonstrate that XGB-SAD outperforms the comparison methods in terms of small-scale attack detection and overall detection, which proves the performance of our method.

Fusing hypergraph spectral features for shilling attack detection

Recommender systems can effectively improve user experience, but they are vulnerable to shilling attacks due to their open nature. Attackers inject fake user profiles to destroy the security and reliability of the recommender systems. Therefore, it is crucial to detect shilling attacks effectively. The primitive detection models are feasible but costly because of the dependence on plenty of hand-engineered explicit features based on statistical measures. Even though the upgraded models based on learning embeddings of the implicit features are more general, they fail to take some distinct features in distinguishing fake users into consideration. Moreover, these primitive and upgraded models are difficult to capture the high order relationships between users and items as the models usually learn the embedding from the first-order interactions. The representation and similarity information learned from the first-order interactions are not comprehensive enough, limiting the detection task. To this end, we propose a novel shilling attack detection model by fusing hypergraph spectral features (SpDetector). The proposed model combines the explicit and implicit features to balance the effectiveness and generality and deal with the high order relationships by hypergraphs-based embedding. From the implicit perspective, SpDetector constructs user hypergraphs and item hypergraphs for the high-order relationships hidden in the interaction and extracts spectral features from hypergraphs to capture high-order similarity for users and items, respectively. From the explicit perspective, it extracts two kinds of explicit features: item similarity offsets (ISO) based on item spectral features and rating prediction errors (RPE), for all users as their distinct capability of distinguishing fake users. Finally, the SpDetector learns to distinguish fake users by training a deep neural network with those features. Experiments conducted on MovieLens and Amazon datasets show that SpDetector outperforms state-of-the-art detection models.

Detecting Group Shilling Attacks in Online Recommender Systems Based on Bisecting K-Means Clustering

Existing shilling attack detection approaches focus mainly on identifying individual attackers in online recommender systems and rarely address the detection of group shilling attacks in which a group of attackers colludes to bias the output of an online recommender system by injecting fake profiles. In this article, we propose a group shilling attack detection method based on the bisecting K-means clustering algorithm. First, we extract the rating track of each item and divide the rating tracks to generate candidate groups according to a fixed time interval. Second, we propose item attention degree and user activity to calculate the suspicious degrees of candidate groups. Finally, we employ the bisecting K-means algorithm to cluster the candidate groups according to their suspicious degrees and obtain the attack groups. The results of experiments on the Netflix and Amazon data sets indicate that the proposed method outperforms the baseline methods.

Hybrid convolutional neural network (CNN) and long-short term memory (LSTM) based deep learning model for detecting shilling attack in the social-aware network

In social aware network (SAN) paradigm, the fundamental activities concentrate on exploring the behavior and attributes of the users. This investigation of user characteristic aids in the design of highly efficient and suitable protocols. In particular, the shilling attack introduces a high degree of vulnerability into the recommender systems. The shilling attackers use the reviews, user ratings and forged user generated content data for the computation of recommendation rankings. The detection of shilling attack in recommender systems is considered to be essential for sustaining their fairness and reliability. In specific, the collaborative filtering strategies utilized for detecting shilling attackers through efficient user behavior mining are considered as the predominant methodologies in the literature. In this paper, a hybrid convolutional neural network (CNN) and long-short term memory (LSTM)-based deep learning model (CNN–LSTM) is proposed for detecting shilling attack in recommender systems. This deep learning model utilizes the transformed network architecture for exploiting the deep-level attributes derived from user rated profiles. It overcomes the limitations of the existing shilling attack detection methods which mostly focuses on identifying spam users by designing features artificially in order to enhance their efficiency and robustness. It is also potent in elucidating deep-level features for efficiently detecting shilling attacks by accurately elaborating the user ratings. The experimental results confirmed the significance of the proposed CNN–LSTM approach by accurately detecting most of the obfuscated attacks compared to the state-of-art algorithms used for investigation.

A novel Kalman Filter based shilling attack detection algorithm.

Collaborative filtering has been widely used in recommendation systems to recommend items that users might like. However, collaborative filtering based recommendation systems are vulnerable to shilling attacks. Malicious users tend to increase or decrease the recommended frequency of target items by injecting fake profiles. In this paper, we propose a Kalman filter-based attack detection model, which statistically analyzes the difference between the actual rating and the predicted rating calculated by this model to find the potential abnormal time period. The Kalman Filter filters out suspicious ratings based on the abnormal time period and identifies suspicious users based on the source of these ratings. The experimental results show that our method performs much better detection performance for the shilling attack than the traditional methods.

A novel shilling attack detection model based on particle filter and gravitation

With the rapid development of e-commerce, the security issues of collaborative filtering recommender systems have been widely investigated. Malicious users can benefit from injecting a great quantities of fake profiles into recommender systems to manipulate recommendation results. As one of the most important attack methods in recommender systems, the shilling attack has been paid considerable attention, especially to its model and the way to detect it. Among them, the loose version of Group Shilling Attack Generation Algorithm (GSAGen l ) has outstanding performance. It can be immune to some PCC (Pearson Correlation Coefficient)-based detectors due to the nature of anti-Pearson correlation. In order to overcome the vulnerabilities caused by GSAGen l a gravitation-based detection model (GBDM) is presented, integrated with a sophisticated gravitational detector and a decider. And meanwhile two new basic attributes and a particle filter algorithm are used for tracking prediction. And then, whether an attack occurs can be judged according to the law of universal gravitation in decision-making. The detection performances of GBDM, HHT-SVM, UnRAP, AP-UnRAP Semi-SAD, SVM-TIA and PCA-P are compared and evaluated. And simulation results show the effectiveness and availability of GBDM.

An Outlier Degree Shilling Attack Detection Algorithm Based on Dynamic Feature Selection

Recommender system is widely used in various fields for dealing with information overload effectively, and collaborative filtering plays a vital role in the system. However, recommender system suffers from its vulnerabilities by malicious attacks significantly, especially, shilling attacks because of the open nature of recommender system and the dependence on data. Therefore, detecting shilling attack has become an important issue to ensure the security of recommender system. Most of the existing methods of detecting shilling attack are based on user ratings, and one limitation is that they are likely to be interfered by obfuscation techniques. Moreover, traditional detection algorithms cannot handle different types of shilling attacks flexibly. In order to solve the problems, we proposed an outlier degree shilling attack detection algorithm by using dynamic feature selection. Considering the differences when users choose items, we combined rating-based indicators with user popularity, and utilized the information entropy to select detection indicators dynamically. Therefore, a variety of shilling attack models can be dealt with flexibility in this way. The experiments show that the proposed algorithm can achieve better detection performance and interference immunity.

Trustworthy and profit: A new value-based neighbor selection method in recommender systems under shilling attacks

User-based collaborative filtering recommender systems are widely deployed by e-retailers to facilitate customer’ decision-making and enhance e-retailers' profitability. Despite the advantages these systems provide, their recommendation effectiveness is vulnerable to attacks from malicious users who inject biased ratings. Such attacks against recommender systems are called shilling attacks. Although several shilling attack detection mechanisms have been proposed in previous studies, their detection performance is limited in various attack conditions. Furthermore, few of these mechanisms consider the value-dimension associated with recommendations, which is crucial for e-retailers. This research proposes a novel approach called Value-based Neighbor Selection (VNS) to address the above limitations. The objective of this approach is to protect recommender systems from shilling attacks while improving e-retailers' profitability. It alleviates the aforementioned problems through strategically selecting neighbors whose preferences are then used to make recommendations. We have performed a series of empirical validations in various attack conditions to compare the performance of the proposed method and three benchmark methods, in terms of both recommendation accuracy and e-retailer profitability. The results show the advantages of the proposed method in balancing customer satisfaction and e-retailer profitability.

Shilling attack detection in binary data: a classification approach

Reliability of a recommender system is extremely substantial for the continuity of the system. Malicious users may harm the reliability of predictions by injecting fake profiles called shilling attacks into the system. Therefore, the detection of such attacks is vital for a recommender system. Thus, many shilling attack detection methods have been studied. However, the proposed solutions work only on numerical rating based recommender systems. On the other hand, it has been shown that collaborative filtering systems utilizing binary ratings are also vulnerable to shilling attacks. In this work, we propose a detection method, which finds out six well-known shilling attack models against binary ratings-based collaborative filtering systems. Besides deriving generic attributes from user profiles, we generate additional model-specific attributes in order to deal with fake profiles. Our empirical results show that the proposed method successfully detects attack profiles even with low attack size and filler size values.

Detecting shilling attacks in social recommender systems based on time series analysis and trust features

In social recommender systems or trust-based recommender systems, malicious users can bias the recommendations by injecting a large number of fake profiles and by building bogus trust relationships. The existing shilling attack detection methods suffer from low precision when detecting attacks in social recommender systems because they focus mainly on the rating pattern differences between attack profiles and genuine ones and ignore the trust relationships between users. In this paper, we propose an approach for detecting shilling attacks in social recommender systems based on time series analysis and trust features (TSA–TF). Firstly, we construct rating distribution time series for items and propose a dynamic rating distribution prediction model to detect suspicious items by using a single exponential smoothing method. Then, we filter out a part of genuine user profiles by analyzing suspicious items and obtain the set of suspicious user profiles. Secondly, we propose four features by combining rating patterns and trust relationships and train a support vector machine (SVM) classifier to discriminate attack profiles in the set of suspicious user profiles. Experiments on the CiaoDVD dataset and Epinions dataset show that the proposed approach can improve the detection precision while maintaining a high recall.

Detecting shilling attacks in recommender systems based on analysis of user rating behavior

The existing unsupervised methods for detecting shilling attacks are mostly based on the rating patterns of users, ignoring the rating behavior difference between genuine users and attack users, and these methods suffer from low accuracy in detecting various shilling attacks without a priori knowledge of the attacks. To address these limitations, we propose a novel unsupervised shilling attack detection model based on an analysis of user rating behavior. First, we identify the target item(s) and the corresponding intentions of the attack users by analyzing the deviation of rating tendencies on each item, and based on this analysis, a set of suspicious users is constructed. Second, we analyze the users’ rating behaviors from an interest preference and rating preference perspective. In particular, we measure the diversity and memory of users’ interest preferences by entropy and block entropy, respectively, and we analyze the memory of user rating preferences by a self-correlation analysis. Finally, we calculate the suspicious degree and spot attack users in the set of suspicious users based on measurements of user rating behavior. Experimental results on the Netflix dataset, the MovieLens 1M dataset and the sampled Amazon review dataset demonstrate the effectiveness of the proposed detection model.

Improving the Shilling Attack Detection in Recommender Systems Using an SVM Gaussian Mixture Model

With recent advances in e-commerce platforms, the information overload has grown due to increasing number of users, rapid generation of data and items in the recommender system. This tends to create serious problems in such recommender systems. The increasing features in recommender systems pose some new challenges due to poor resilience to mitigate against vulnerable attacks. In particular, the recommender systems are more prone to be attacked by shilling attacks, which creates more vulnerability. A recommender system with poor detection of attacks leads to a reduced detection rate. The performance of the recommender system is thus affected with poor detection ability. Hence, in this paper, we improve the resilience against shilling attacks using a modified Support Vector Machine (SVM) and a machine learning algorithm. The Gaussian Mixture Model is used as a machine learning algorithm to increase the detection rate and it further reduces the dimensionality of data in recommender systems. The proposed method is evaluated against several result metrics, such as the recall rate, precision rate and false positive rate between different attacks. The results of the proposed system are evaluated against probabilistic recommender approaches to demonstrate the efficacy of machine learning language in recommender systems.

Privacy-Aware Detection of Shilling Profiles on Arbitrarily Distributed Recommender Systems

Due to the mutual advantage of small-scale online service providers, they need to collaborate to deliver recommendations based on arbitrarily distributed preference data without jeopardizing their confidentiality. Besides privacy issues, parties also have concerns regarding the vulnerability against recommendation manipulation attempts, referred to as shilling attacks. Although there are methods for detecting these injected malicious profiles in central server-based configurations, they are not readily suitable for employing arbitrarily distributed data. In this paper, we present a novel classification-based shilling attack detection protocol enabling the recognition of malicious profiles in arbitrarily distributed configurations without compromising the privacy of collaborating parties. The analysis of the proposed protocol regarding confidentiality of parties reveals that the process is bound to collaboration by design, which does not allow parties to achieve detection by themselves. Furthermore, empirical evaluations using real-world preference data demonstrate that the protocol can achieve significantly high detection rates facilitating privacy-aware data collaboration.