Biblio

In this work, we conduct a systematic study on data poisoning attacks to Matrix Factorisation (MF) based Recommender Systems (RS) where a determined attacker injects fake users with false user-item feedback, with an objective to promote a target item by increasing its rating. We explore the capability of a MF based approach to reduce the impact of attack on targeted item in the system. We develop and evaluate multiple techniques to update the user and item feature matrices when incorporating new ratings. We also study the effectiveness of attack under increasing filler items and choice of target item.Our experimental results based on two real-world datasets show that the observations from the study could be used to design a more robust MF based RS.

With the rapid development of artificial intelligence (AI), many companies are moving towards automating their services using automated conversational agents. Dialogue-based conversational recommender agents, in particular, have gained much attention recently. The successful development of such systems in the case of natural language input is conditioned by the ability to understand the users’ utterances. Predicting the users’ intents allows the system to adjust its dialogue strategy and gradually upgrade its preference profile. Nevertheless, little work has investigated this problem so far. This paper proposes an LSTM-based Neural Network model and compares its performance to seven baseline Machine Learning (ML) classifiers. Experiments on a new publicly available dataset revealed The superiority of the LSTM model with 95% Accuracy and 94% F1-score on the full dataset despite the relatively small dataset size (9300 messages and 17 intents) and label imbalance.

Recommenders are central in many applications today. The most effective recommendation schemes, such as those based on collaborative filtering (CF), exploit similarities between user profiles to make recommendations, but potentially expose private data. Federated learning and decentralized learning systems address this by letting the data stay on user's machines to preserve privacy: each user performs the training on local data and only the model parameters are shared. However, sharing the model parameters across the network may still yield privacy breaches. In this paper, we present Rex, the first enclave-based decentralized CF recommender. Rex exploits Trusted execution environments (TEE), such as Intel software guard extensions (SGX), that provide shielded environments within the processor to improve convergence while preserving privacy. Firstly, Rex enables raw data sharing, which ultimately speeds up convergence and reduces the network load. Secondly, Rex fully preserves privacy. We analyze the impact of raw data sharing in both deep neural network (DNN) and matrix factorization (MF) recommenders and showcase the benefits of trusted environments in a full-fledged implementation of Rex. Our experimental results demonstrate that through raw data sharing, Rex significantly decreases the training time by 18.3 x and the network load by 2 orders of magnitude over standard decentralized approaches that share only parameters, while fully protecting privacy by leveraging trustworthy hardware enclaves with very little overhead.

A recommender system aims to suggest the most relevant items to users based on their personal data. However, data privacy is a growing concern for anyone. Secure recommender system is a research direction to preserve user privacy while maintaining as high performance as possible. The most recent strategy is to use Federated Learning, a machine learning technique for privacy-preserving distributed training. In Federated Learning, a subset of users will be selected for training model using data at local systems, the server will securely aggregate the computing result from local models to generate a global model, finally that model will give recommendations to users. In this paper, we present a novel algorithm to train Collaborative Filtering recommender system specialized for the ranking task in Federated Learning setting, where the goal is to protect user interaction information (i.e., implicit feedback). Specifically, with the help of the algorithm, the recommender system will be trained by Neural Collaborative Filtering, one of the state-of-the-art matrix factorization methods and Bayesian Personalized Ranking, the most common pairwise approach. In contrast to existing approaches which protect user privacy by requiring users to download/upload the information associated with all interactions that they can possibly interact with in order to perform training, the algorithm can protect user privacy at low communication cost, where users only need to obtain/transfer the information related to a small number of interactions per training iteration. Above all, through extensive experiments, the algorithm has demonstrated to utilize user data more efficient than the most recent research called FedeRank, while ensuring that user privacy is still preserved.

A recommender system is a filtering application based on personalized information from acquired big data to predict a user's preference. Traditional recommender systems primarily rely on keywords or scene patterns. Users' subjective emotion data are rarely utilized for preference prediction. Novel Brain Computer Interfaces hold incredible promise and potential for intelligent applications that rely on collected user data like a recommender system. This paper describes a deep learning method that uses Brain Computer Interfaces (BCI) based neural measures to predict a user's preference on short music videos. Our models are employed on both population-wide and individualized preference predictions. The recognition method is based on dynamic histogram measurement and deep neural network for distinctive feature extraction and improved classification. Our models achieve 97.21%, 94.72%, 94.86%, and 96.34% classification accuracy on two-class, three-class, four-class, and nine-class individualized predictions. The findings provide evidence that a personalized recommender system on an implicit BCI has the potential to succeed.

Ubiquitous environment embedded with artificial intelligent consist of heterogenous smart devices communicating each other in several context for the computation of requirements. In such environment the trust among the smart users have taken as the challenge to provide the secure environment during the communication in the ubiquitous region. To provide the secure trusted environment for the users of ubiquitous system proposed approach aims to extract behavior of smart invisible entities by retrieving their behavior of communication in the network and applying the recommendation-based filters using Deep learning (RBF-DL). The proposed model adopts deep learning-based classifier to classify the unfair recommendation with fair ones to have a trustworthy ubiquitous system. The capability of proposed model is analyzed and validated by considering different attacks and additional feature of instances in comparison with generic recommendation systems.

With the proliferation of false redundant information on various e-commerce platforms, ineffective recommendations and other untrustworthy behaviors have seriously hindered the healthy development of e-commerce platforms. Modern recommendation systems often use side information to alleviate these problems and also increase prediction accuracy. One such piece of side information, which has been widely investigated, is trust. However, it is difficult to obtain explicit trust relationship data, so researchers infer trust values from other methods, such as the user-to-item relationship. In this paper, addressing the problems, we proposed a novel trust-based recommender model called UITrust, which uses user-item relationship value to improve prediction accuracy. With the improvement the traditional similarity measures by employing the entropies of user and item history ratings to reflect the global rating behavior on both. We evaluate the proposed model using two real-world datasets. The proposed model performs significantly better than the baseline methods. Also, we can use the UITrust to alleviate the sparsity problem associated with correlation-based similarity. In addition to that, the proposed model has a better computational complexity for making predictions than the k-nearest neighbor (kNN) method.

Existing group shilling attack detection methods mainly depend on human feature engineering to extract group attack behavior features, which requires a high knowledge cost. To address this problem, we propose a group shilling attack detection method based on maximum density subtensor mining. First, the rating time series of each item is divided into time windows and the item tensor groups are generated by establishing the user-rating-time window data models of three-dimensional tensor. Second, the M-Zoom model is applied to mine the maximum dense subtensor of each item, and the subtensor groups with high consistency of behaviors are selected as candidate groups. Finally, a dual-input convolutional neural network model is designed to automatically extract features for the classification of real users and group attack users. The experimental results on the Amazon and Netflix datasets show the effectiveness of the proposed method.

Recommender systems, which aim to suggest personalized lists of items for users, have drawn a lot of attention. In fact, many of these state-of-the-art recommender systems have been built on deep neural networks (DNNs). Recent studies have shown that these deep neural networks are vulnerable to attacks, such as data poisoning, which generate fake users to promote a selected set of items. Correspondingly, effective defense strategies have been developed to detect these generated users with fake profiles. Thus, new strategies of creating more ‘realistic’ user profiles to promote a set of items should be investigated to further understand the vulnerability of DNNs based recommender systems. In this work, we present a novel framework CopyAttack. It is a reinforcement learning based black-box attacking method that harnesses real users from a source domain by copying their profiles into the target domain with the goal of promoting a subset of items. CopyAttack is constructed to both efficiently and effectively learn policy gradient networks that first select, then further refine/craft user profiles from the source domain, and ultimately copy them into the target domain. CopyAttack’s goal is to maximize the hit ratio of the targeted items in the Top-k recommendation list of the users in the target domain. We conducted experiments on two real-world datasets and empirically verified the effectiveness of the proposed framework. The implementation of CopyAttack is available at https://github.com/wenqifan03/CopyAttack.

Recommender systems have been proved to be effective techniques to provide users with better experiences. However, when a recommender knows the user's preference characteristics or gets their sensitive information, then a series of privacy concerns are raised. A amount of solutions in the literature have been proposed to enhance privacy protection degree of recommender systems. Although the existing solutions have enhanced the protection, they led to a decrease in recommendation accuracy simultaneously. In this paper, we propose a security-aware hybrid recommendation method by combining the factorization and regression techniques. Specifically, the differential privacy mechanism is integrated into data pre-processing for data encryption. Firstly data are perturbed to satisfy differential privacy and transported to the recommender. Then the recommender calculates the aggregated data. However, applying differential privacy raises utility issues of low recommendation accuracy, meanwhile the use of a single model may cause overfitting. In order to tackle this challenge, we adopt a fusion prediction model by combining linear regression (LR) and matrix factorization (MF) for collaborative recommendation. With the MovieLens dataset, we evaluate the recommendation accuracy and regression of our recommender system and demonstrate that our system performs better than the existing recommender system under privacy requirement.

In a group shilling attack, attackers work collaboratively to inject fake profiles aiming to obtain desired recommendation result. This type of attacks is more harmful to recommender systems than individual shilling attacks. Previous studies pay much attention to detect individual attackers, and little work has been done on the detection of shilling groups. In this work, we introduce a topic modeling method of natural language processing into shilling attack detection and propose a shilling group detection method on the basis of hierarchical topic model. First, we model the given dataset to a series of user rating documents and use the hierarchical topic model to learn the specific topic distributions of each user from these rating documents to describe user rating behaviors. Second, we divide candidate groups based on rating value and rating time which are not involved in the hierarchical topic model. Lastly, we calculate group suspicious degrees in accordance with several indicators calculated through the analysis of user rating distributions, and use the k-means clustering algorithm to distinguish shilling groups. The experimental results on the Netflix and Amazon datasets show that the proposed approach performs better than baseline methods.

Session-based recommendation is the task of predicting user actions during short online sessions. The user is considered to be anonymous in this setting, with no past behavior history available. Predicting anonymous users' next actions and their preferences in the absence of historical user behavior information is valuable from a cybersecurity and aerospace perspective, as cybersecurity measures rely on the prompt classification of novel threats. Our offered solution builds upon the previous representation learning work originating from natural language processing, namely BERT, which stands for Bidirectional Encoder Representations from Transformers (Devlin et al., 2018). In this paper we propose CyberBERT, the first deep session-based recommender system to employ bidirectional transformers to model the intent of anonymous users within a session. The session-based setting lends itself to applications in threat recognition, through monitoring of real-time user behavior using the CyberBERT architecture. We evaluate the efficiency of this dynamic state method using the Windows PE Malware API sequence dataset (Catak and Yazi, 2019), which contains behavior for 7107 API call sequences executed by 8 classes of malware. We compare the proposed CyberBERT solution to two high-performing benchmark algorithms on the malware dataset: LSTM (Long Short-term Memory) and transformer encoder (Vaswani et al., 2017). We also evaluate the method using the YOOCHOOSE 1/64 dataset, which is a session-based recommendation dataset that contains 37,483 items, 719,470 sessions, and 31,637,239 clicks. Our experiments demonstrate the advantage of a bidirectional architecture over the unidirectional approach, as well as the flexibility of the CyberBERT solution in modelling the intent of anonymous users in a session. Our system achieves state-of-the-art measured by F1 score on the Windows PE Malware API sequence dataset, and state-of-the-art for P@20 and MRR@20 on YOOCHOOSE 1/64. As CyberBERT allows for user behavior monitoring in the absence of behavior history, it acts as a robust malware classification system that can recognize threats in aerospace systems, where malicious actors may be interacting with a system for the first time. This work provides the backbone for systems that aim to protect aviation and aerospace applications from prospective third-party applications and malware.

In the economics environment, Job Matching is always a challenge involving the evolution of knowledge and skills. A good matching of skills and jobs can stimulate the growth of economics. Recommender System (RecSys), as one kind of Job Matching, can help the candidates predict the future job relevant to their preferences. However, RecSys still has the problem of cold start and data sparsity. The content-based filtering in RecSys needs the adaptive data for the specific staffing tasks of Bidirectional Encoder Representations from Transformers (BERT). In this paper, we propose a job RecSys based on skills and locations using a domain-specific Knowledge Graph (KG). This system has three parts: a pipeline of Named Entity Recognition (NER) and Relation Extraction (RE) using BERT; a standardization system for pre-processing, semantic enrichment and semantic similarity measurement; a domain-specific Knowledge Graph (KG). Two different relations in the KG are computed by cosine similarity and Term Frequency-Inverse Document Frequency (TF-IDF) respectively. The raw data used in the staffing RecSys include 3000 descriptions of job offers from Indeed, 126 Curriculum Vitae (CV) in English from Kaggle and 106 CV in French from Linx of Capgemini Engineering. The staffing RecSys is integrated under an architecture of Microservices. The autonomy and effectiveness of the staffing RecSys are verified through the experiment using Discounted Cumulative Gain (DCG). Finally, we propose several potential research directions for this research.

Aiming at group shilling attacks in recommender systems, a shilling group detection approach based on triangle dense subgraph mining is proposed. First, the user relation graph is built by mining the relations among users in the rating dataset. Second, the improved triangle dense subgraph mining method and the personalizing PageRank seed expansion algorithm are used to divide candidate shilling groups. Finally, the suspicious degrees of candidate groups are calculated using several group detection indicators and the attack groups are obtained. Experiments indicate that our method has better detection performance on the Amazon and Yelp datasets than the baselines.

In Social Aware Network (SAN) model, the elementary actions focus on investigating the attributes and behaviors of the customer. This analysis of customer attributes facilitate in the design of highly active and improved protocols. In specific, the recommender systems are highly vulnerable to the shilling attack. The recommender system provides the solution to solve the issues like information overload. Collaborative filtering based recommender systems are susceptible to shilling attack known as profile injection attacks. In the shilling attack, the malicious users bias the output of the system's recommendations by adding the fake profiles. The attacker exploits the customer reviews, customer ratings and fake data for the processing of recommendation level. It is essential to detect the shilling attack in the network for sustaining the reliability and fairness of the recommender systems. This article reviews the most prominent issues and challenges of shilling attack. This paper presents the literature survey which is contributed in focusing of shilling attack and also describes the merits and demerits with its evaluation metrics like attack detection accuracy, precision and recall along with different datasets used for identifying the shilling attack in SAN network.

Recommender systems in software engineering provide developers with a wide range of valuable items to help them complete their tasks. Among others, API recommender systems have gained momentum in recent years as they became more successful at suggesting API calls or code snippets. While these systems have proven to be effective in terms of prediction accuracy, there has been less attention for what concerns such recommenders’ resilience against adversarial attempts. In fact, by crafting the recommenders’ learning material, e.g., data from large open-source software (OSS) repositories, hostile users may succeed in injecting malicious data, putting at risk the software clients adopting API recommender systems. In this paper, we present an empirical investigation of adversarial machine learning techniques and their possible influence on recommender systems. The evaluation performed on three state-of-the-art API recommender systems reveals a worrying outcome: all of them are not immune to malicious data. The obtained result triggers the need for effective countermeasures to protect recommender systems against hostile attacks disguised in training data.

In the last few years, cluster ensembles have emerged as powerful techniques that integrate multiple clustering methods into recommender systems. Such integration leads to improving the performance, quality and the accuracy of the generated recommendations. This paper proposes a novel recommender system based on a cluster ensemble technique for big data. The proposed system incorporates the collaborative filtering recommendation technique and the cluster ensemble to improve the system performance. Besides, it integrates the Expectation-Maximization method and the HyperGraph Partitioning Algorithm to generate new recommendations and enhance the overall accuracy. We use two real-world datasets to evaluate our system: TED Talks and MovieLens. The experimental results show that the proposed system outperforms the traditional methods that utilize single clustering techniques in terms of recommendation quality and predictive accuracy. Most importantly, the results indicate that the proposed system provides the highest precision, recall, accuracy, F1, and the lowest Root Mean Square Error regardless of the used similarity strategy.

Recommender systems help people in finding a particular item based on their preference from a wide range of products in online shopping rapidly. One of the most popular models of recommendation systems is the Collaborative Filtering Recommendation System (CFRS) that recommend the top-K items to active user based on peer grouping user ratings. The implementation of CFRS is easy and it can easily be attacked by fake users and affect the recommendation. Fake users create a fake profile to attack the RS and change the output of it. Different attack types with different features and attacking methods exist in which decrease the accuracy. It is important to detect fake users, remove their rating from rating matrix and recognize the items has been attacked. In the recent years, many algorithms have been proposed to detect the attackers but first, researchers have to inject the attack type into their dataset and then evaluate their proposed approach. The purpose of this article is to develop a tool to inject the different attack types to datasets. Proposed tool constructs a new dataset containing the fake users therefore researchers can use it for evaluating their proposed attack detection methods. Researchers could choose the attack type and the size of attack with a user interface of our proposed tool easily.

Recommender system helps people in decision making by asking their preferences about various items and recommends other items that have not been rated yet and are similar to their taste. A traditional recommendation system aims at generating a set of recommendations based on inter-user similarity that will satisfy the target user. Positive preferences as well as negative preferences of the users are taken into account so as to find strongly related users. Weighted entropy is usedz as a similarity measure to determine the similar taste users. The target user is asked to fill in the ratings so as to identify the closely related users from the knowledge base and top N recommendations are produced accordingly. Results show a considerable amount of improvement in accuracy after using weighted entropy and opposite preferences as a similarity measure.

Volume, Variety, Velocity, Veracity & Value of data has drawn the attention of many analysts in the last few years. Performance optimization and comparison are the main challenges we face when we talk about the humongous volume of data. Data Analysts use data for activities like forecasting or deep learning and to process these data various tools are available which helps to achieve this task with minimum efforts. Recommendation System plays a crucial role while running any business such as a shopping website or travel agency where the system recommends the user according to their search history, likes, comments, or their past order/booking details. Recommendation System works on various strategies such as Content Filtering, Collaborative Filtering, Neighborhood Methods, or Matrix Factorization methods. For achieving maximum efficiency and accuracy based on the data a specific strategy can be the best case or the worst case for that scenario. Matrix Factorization is the key point of interest in this work. Matrix Factorization strategy includes multiplication of user matrix and item matrix in-order to get a rating matrix that can be recommended to the users. Matrix Multiplication can be achieved by using various algorithms such as Naive Algorithm, Strassen Algorithm, Coppersmith - Winograd (CW) Algorithm. In this work, a new algorithm is proposed to achieve less amount of time and space complexity used in-order for performing matrix multiplication which helps to get the results much faster. By using the Matrix Factorization strategy with various Matrix Multiplication Algorithm we are going to perform a comparative analysis of the same to conclude the proposed algorithm is more efficient.

Recommendation systems, especially collaborative filtering recommenders, are vulnerable to shilling attacks as some profit-driven users may inject fake profiles into the system to alter recommendation outputs. Current shilling attack detection methods are mostly based on feature extraction techniques. The hand-designed features can confine the model to specific domains or datasets while deep learning techniques enable us to derive deeper level features, enhance detection performance, and generalize the solution on various datasets and domains. This paper illustrates the application of two deep learning methods to detect shilling attacks. We conducted experiments on the MovieLens 100K and Netflix Dataset with different levels of attacks and types. Experimental results show that deep learning models can achieve an accuracy of up to 99%.

When users in various web and mobile applications enjoy the convenience of recommendation systems, they are vulnerable to attribute inference attacks. The accumulating online behaviors of users (e.g., clicks, searches, ratings) naturally brings out user preferences, and poses an inevitable threat of privacy that adversaries can infer one's private profiles (e.g., gender, sexual orientation, political view) with AI-based algorithms. Existing defense methods assume the existence of a trusted third party, rely on computationally intractable algorithms, or have impact on recommendation utility. These imperfections make them impractical for privacy preservation in real-life scenarios. In this work, we introduce BiasBooster, a practical proactive defense method based on behavior segmentation, to protect user privacy against attribute inference attacks from user behaviors, while retaining recommendation utility with a heuristic recommendation aggregation module. BiasBooster is a user-centric approach from client side, which proactively divides a user's behaviors into weakly related segments and perform them with several dummy identities, then aggregates real-time recommendations for user from different dummy identities. We estimate its effectiveness of preservation on both privacy and recommendation utility through extensive evaluations on two real-world datasets. A Chrome extension is conducted to demonstrate the feasibility of applying BiasBooster in real world. Experimental results show that compared to existing defenses, BiasBooster substantially reduces the averaged accuracy of attribute inference attacks, with minor utility loss of recommendations.

Cloud computing is applied in various domains, among which social recommender systems are well-received because of their effectivity to provide suggestions for users. Social recommender systems perform well in alleviating cold start problem, but it suffers from shilling attack due to its natural openness. Shilling attack is an injection attack mainly acting on the training process of machine learning, which aims to advance or suppress the recommendation ranking of target items. Some researchers have studied the influence of shilling attacks in two perspectives simultaneously, which are user-item's rating and user-user's relation. However, they take more consideration into user-item's rating, and up to now, the construction of user-user's relation has not been explored in depth. To explore shilling attacks with complex relations, in this paper, we propose two novel attack models based on triangle relations in social networks. Furthermore, we explore the influence of these models on five social recommendation algorithms. The experimental results on three datasets show that the recommendation can be affected by the triangle relation attacks. The attack model combined with triangle relation has a better attack effect than the model only based on rating injection and the model combined with random relation. Besides, we compare the functions of triangle relations in friend recommendation and product recommendation.

With the development of the Internet, the amount of information on the Internet is increasing rapidly, which makes it difficult for people to select the information they really want. A recommendation system is an effective way to solve this problem. Fake users can be injected by criminals to attack the recommendation system; therefore, accurate identification of fake users is a necessary feature of the recommendation system. Existing fake user detection algorithms focus on designing recognition methods for different types of attacks and have limited detection capabilities against unknown or hybrid attacks. The use of deep learning models can automate the extraction of false user scoring features, but neural network models are not applicable to discrete user scoring data. In this paper, random walking is used to rearrange the otherwise discrete user rating data into a rating feature matrix with spatial continuity. The rating data and the text data have some similarity in the distribution mode. By effective analogy, the TextCNN model originally used in NLP domain can be improved and applied to the classification task of rating feature matrix. Combining the ideas of random walking and word vector processing, this paper proposes a TextCNN detection model for user rating data. To verify the validity of the proposed model, the model is tested on MoiveLens dataset against 7 different attack detection algorithms, and exhibits better performance when compared with 4 attack detection algorithms. Especially for the Aop attack, the proposed model has nearly 100% detection performance with F1 - value as the evaluation index.

Recent advances in information filtering have resulted in effective recommender systems that are able to provide online personalized recommendations to millions of users from all over the world. However, most of these systems ignore the explanation purpose while producing recommendations with high-quality results. Moreover, the classification of reviews given to users as explanations is not fully exploited in previous studies. In this paper, we develop a convolutional neural network-based reviews classification method for explainable recommendation systems. The convolutional neural network is used to extract the reviews features for predicting whether the reviews provided as explanations are positive or negative. Based on such additional information, users can understand not only why certain items are recommended for them but also get support to know the nature of such explanations. We conduct experiments on a dataset from Amazon. The experimental results show that our method outperforms state-of-the-art methods.