Defending Grey Attacks by Exploiting Wavelet Analysis in Collaborative Filtering Recommender Systems

Nowadays, collaborative filtering recommender systems have been widely deployed in many commercial companies to make profit. Neighborhood‐based collaborative filtering (CF) is common and effective. To date, despite its effectiveness, there has been little effort to explore their robustness and the impact of data poisoning attacks on their performance. Can the neighborhood‐based recommender systems be easily fooled? To this end, we shed light on the robustness of neighborhood‐based recommender systems and propose a novel data poisoning attack framework, encoding the purpose of attack and constraint against them. We first illustrate how to calculate the optimal data poisoning attack, namely, UNAttack. We inject a few well‐designed fake users into the recommender systems such that target items will be recommended to as many normal users as possible. Extensive experiments are conducted on three real‐world datasets to validate the effectiveness and the transferability of our proposed method. In addition, some interesting phenomena can be found. For example, (i) neighborhood‐based recommender systems with Euclidean distance‐based similarity have strong robustness and (ii) the fake users can be transferred to attack the state‐of‐the‐art CF recommender systems such as neural CF and Bayesian personalized ranking matrix factorization.

Collaborative filtering recommender systems are essentially information systems which are capable of combining the judgment of a large group of people to make personalized recommendations and thereby alleviate the so-called information overload problem. However,collaborative filtering recommender systems are generally vulnerable to shilling attacks. Attackers can inject carefully chosen profiles into recommender systems in order to bias the recommendation results to their benefits. This may lead to a significant negative impact on the robustness of the systems. The main contribution of this paper is to build a probabilistic model for attack detection in the framework of probabilistic generative model. Experimental results show that this model can effectively detect shilling attacks of typical types.

With the rapid development of e-commerce, collaborative filtering recommendation system has been widely used in various network platforms. Using recommendation system to accurately predict customers’ preferences for goods can solve the problem of information overload faced by users and improve users’ dependence on the network platform. Because the recommendation system based on collaborative filtering technology has the ability to recommend more abstract or difficult to describe goods in words, the research related to collaborative filtering technology has attracted more and more attention. According to the past research, in collaborative filtering algorithm, if Pearson correlation coefficient is used, errors will occur under special circumstances. In this study, the normal recovery similarity measure is used to modify the similarity value to correct the error value of a collaborative filtering recommendation algorithm. Based on this, a big data analysis method based on a modified collaborative filtering recommendation algorithm is proposed. This research implemented it in the cloud Hadoop environment, and measure the execution time with 2, 5 and 8 nodes. Then the research compared it with the execution time of a single machine, and analyze its speedup ratio and efficiency. The experimental results show that the execution time increases with the number of neighbors. When the number of nodes is 5 and 8, the execution time is greatly improved, which improves the efficiency of collaborative filtering algorithm and can cope with massive data in the future.

Collaborative filtering (CF) technique in recommender systems (RS) is a well-known and popular technique that exploits relationships between users or items to make product recommendations to an active user. The effectiveness of existing memory based algorithms depend on the similarity measure that is used to identify nearest neighbours. However, similarity measures utilize only the ratings of co-rated items while computing the similarity between a pair of users or items. In most of the e-commerce applications, the rating matrix is too sparse since even active users of an online system tend to rate only a few items of the entire set of items. Therefore, co-rated items among users are even sparser. Moreover, the ratings a user gives an individual item tells us nothing about his comprehensive interest without which the generated recommendations may not be satisfactory to a user. In order to be able to address these issues, a comprehensive study is made of the various existing measures of similarity in a collaborative filtering recommender system (CFRS) and a hierarchical categorization of products has been proposed to understand the interest of a user in a wider scope so as to provide better recommendations as well as to alleviate data sparsity.

Collaborative filtering (CF) recommender systems have been extensively developed and widely deployed in various social websites, promoting products or services to the users of interest. Meanwhile, work has been attempted at poisoning attacks to CF recommender systems for distorting the recommend results to reap commercial or personal gains stealthily. While existing poisoning attacks have demonstrated their effectiveness with the offline social datasets, they are impractical when applied to the real setting on online social websites. This paper develops a novel and practical poisoning attack solution toward the CF recommender systems without knowing involved specific algorithms nor historical social data information a priori. Instead of directly attacking the unknown recommender systems, our solution performs certain operations on the social websites to collect a set of sampling data for use in constructing a surrogate model for deeply learning the inherent recommendation patterns. This surrogate model can estimate the item proximities, learned by the recommender systems. By attacking the surrogate model, the corresponding solutions (for availability and target attacks) can be directly migrated to attack the original recommender systems. Extensive experiments validate the generated surrogate model's reproductive capability and demonstrate the effectiveness of our attack upon various CF recommender algorithms.

Recent literature predicts that including context in a recommender system may improve its performance. The context-based recommendation approaches are classified as pre-filtering, post-filtering and contextual modeling. Little research has been done on studying whether including context in a recommender system improves the recommendation performance and no research has compared yet the different approaches to contextual RS. The research contribution of this work lies in studying the effect of the context on the recommendation performance and comparing a pre-filtering approach to a post-filtering using a collaborative filtering recommender system.

Collaborative filtering is one of the most widely used methods for user rating prediction in recommender systems. To evaluate a collaborative filtering system, rating datasets are typically used, which comprise thousands to millions of records consisting of user–item–rating tuples. Initially, a similarity metric is used to quantify the closeness between each user and every other user in the dataset, typically based on the ratings that each pair of users has given to the same items. Subsequently, the K users having the largest similarity to the target user are used to produce rating predictions, which lead to recommendations. A particularly challenging case arises when the rating dataset is very sparse. In this scenario, it is difficult not only to find users with commonly rated items but also to determine the optimal similarity metric and suitable values for variable K. Setting a small value for K results in extremely low prediction coverage, leading to unsuccessful recommendations, while setting a very large K value increases memory requirements and prediction/recommendation generation time. Through a multiparameter experiment, this work aims to determine the optimal settings for rating predictions when very sparse datasets are used in collaborative filtering recommender systems.

Learning to rank with pairwise loss functions has been found useful in collaborative filtering recommender systems. At web scale, the optimization is often based on matrix factorization with stochastic gradient descent (SGD) which has a sequential nature. We investigate two different shared memory lock-free parallel SGD schemes based on block partitioning and no partitioning for use with pairwise loss functions. To speed up convergence to a solution, we extrapolate simple practical algorithms from their application to pointwise learning to rank. Experimental results show that the proposed algorithms are quite useful regarding their ranking ability and speedup patterns in comparison to their sequential counterpart.

Collaborative filtering (CF) is considered a powerful technique for generating personalized recommendation. However, significant vulnerabilities have recently been identified in collaborative filtering recommender systems. Malicious users can inject a large number of biased profiles into such a system in order to make recommendations that favor or disfavor given items. The average attack model is a somewhat more sophisticated attack than the random attack model. In this paper, we examine the robustness of our topic-level trust-based recommendation algorithm that incorporate topic-level trust model into classic collaborative filtering algorithm under the average attack. The results of our experiments show that topic-level trust based collaborative filtering algorithm offers significant improvements in stability over the standard k-nearest neighbor approach under average attack.

Recommender systems provide users with product information and suggestions, which has gradually become an important research tool in e-commerce IT technology, which has attracted a lot of attention of researchers. Collaborative filtering recommendation technology has been the most successful recommendation technology so far, but there are two major problems—recommendation quality and scalability. At present, research at home and abroad mainly focuses on recommendation quality, and there is less discussion on scalability. The scalability problem is that as the size of the system increases, the response time of the system increases to a point where users cannot afford it. Existing solutions often result in a significant drop in recommendation quality while reducing recommendation response time. In this paper, the clustering analysis subsystem based on the genetic algorithm is innovatively introduced into the traditional collaborative filtering recommendation system, and its design and implementation are given. In addition, when obtaining the nearest neighbors, only the clustered users of the target user are searched, making it a collaborative filtering recommender system based on genetic clustering. The experimental results show that the response time of the traditional collaborative filtering recommender system increases linearly with the increase in the number of users while the response time of the collaborative filtering recommender system based on genetic clustering remains unchanged with the increase in the number of users. On the other hand, the recommendation quality of the collaborative filtering recommender system based on genetic clustering is basically not degraded compared with that of the traditional collaborative filtering recommender system. Therefore, the collaborative filtering recommender system based on genetic clustering can effectively solve the scalability problem of the collaborative filtering recommender system.

As e-commerce offers more and more choices for users, its structure becomes more and more complicated. Inevitably, it brings about the problem of information overload. The solution to this problem is an e-commerce personalized recommendation system using machine learning technology. People often seem confused when facing extensive information and cannot grasp the key points. This paper studies the personalized recommendation technology of e-commerce: deeply analyzes the related technologies and algorithms of the e-commerce recommendation system and proposes the latest architecture of the e-commerce recommendation system according to the current development status of the e-commerce recommendation system. The system recommends accuracy and real-time requirements and divides the system into two parts: offline mining and online recommendation and analyzes and implements the functions and technologies of each part. User-based recommender systems, collaborative filtering recommender systems, and content-based recommender systems are analyzed, respectively. The personalized recommendation cannot only quickly help customers find the required commodity information in a wide range of complex information but also can compare more commodity information to help customers to judge. However, the existing recommendation system has some problems such as the lack of recommendation personality, the reduced relevance of recommendation, and the poor timeliness of recommendation. Finally, a recommendation system that combines three recommendation algorithms is designed, and experiments are carried out. The newly designed recommendation system is compared with three different recommendation systems, and a summary and outlook are made. Based on the introduction of the relevant theories, characteristics, and mainstream technologies of personalized recommendation based on machine learning, this document presents a constructive example of a model based on the factors that influence personalized e-commerce information recommendations in the retail sector. Through questionnaire surveys, we analyze and design the influencing factors for consumers to purchase personalized products after the survey and build a project using state-of-the-art field learning techniques. Through the model to test the eight hypotheses proposed in this paper, the results show that customer income level, customer online shopping experience, commodity prices, product quality, recommendation relevance, credit evaluation, and service quality will have a significant positive impact on shopping willingness and ultimately affect the customer’s shopping behavior. e-commerce platform can use this influencing factor to establish personalized information recommendation service mode.

Recommender system is the system which, by using automated information filtering technology, recommends products or services to the customers who are likely to be interested in. Those systems are widely used in many different Web retailers such as Amazon.com, Netfix.com, and CDNow.com. Various recommender systems have been developed. Among them, Collaborative Filtering (CF) has been known as the most successful and commonly used approach. CF identifies customers whose tastes are similar to those of a given customer, and recommends items those customers have liked in the past. Numerous CF algorithms have been developed to increase the performance of recommender systems. However, the relative performances of CF algorithms are known to be domain and data dependent. It is very time-consuming and expensive to implement and launce a CF recommender system, and also the system unsuited for the given domain provides customers with poor quality recommendations that make them easily annoyed. Therefore, predicting in advance whether the performance of CF recommender system is acceptable or not is practically important and needed. In this study, we propose a decision making guideline which helps decide whether CF is adoptable for a given application with certain transaction data characteristics. Several previous studies reported that sparsity, gray sheep, cold-start, coverage, and serendipity could affect the performance of CF, but the theoretical and empirical justification of such factors is lacking. Recently there are many studies paying attention to Social Network Analysis (SNA) as a method to analyze social relationships among people. SNA is a method to measure and visualize the linkage structure and status focusing on interaction among objects within communication group. CF analyzes the similarity among previous ratings or purchases of each customer, finds the relationships among the customers who have similarities, and then uses the relationships for recommendations. Thus CF can be modeled as a social network in which customers are nodes and purchase relationships between customers are links. Under the assumption that SNA could facilitate an exploration of the topological properties of the network structure that are implicit in transaction data for CF recommendations, we focus on density, clustering coefficient, and centralization which are ones of the most commonly used measures to capture topological properties of the social network structure. While network density, expressed as a proportion of the maximum possible number of links, captures the density of the whole network, the clustering coefficient captures the degree to which the overall network contains localized pockets of dense connectivity. Centralization reflects the extent to which connections are concentrated in a small number of nodes rather than distributed equally among all nodes. We explore how these SNA measures affect the performance of CF performance and how they interact to each other. Our experiments used sales transaction data from H department store, one of the well?known department stores in Korea. Total 396 data set were sampled to construct various types of social networks. The dependant variable measuring process consists of three steps; analysis of customer similarities, construction of a social network, and analysis of social network patterns. We used UCINET 6.0 for SNA. The experiments conducted the 3-way ANOVA which employs three SNA measures as dependant variables, and the recommendation accuracy measured by F1-measure as an independent variable. The experiments report that 1) each of three SNA measures affects the recommendation accuracy, 2) the density's effect to the performance overrides those of clustering coefficient and centralization (i.e., CF adoption is not a good decision if the density is low), and 3) however though the density is low, the performance of CF is comparatively good when the clustering coefficient is low. We expect that these experiment results help firms decide whether CF recommender system is adoptable for their business domain with certain transaction data characteristics.

Predicting products a customer would like on the basis of other customers ratings for these products has become a well-known approach adopted by many personalized recommendation systems on the Internet. With the development of electronic commerce, the number of customers and products grows rapidly, resulted in the sparsely of the rating dataset. Poor quality is one major challenge in collaborative filtering recommender systems. To solve this problem, slope one algorithm has a good performance. But there is also some places that not make sense. Slope one algorithm assumes that all users are at the same level, and it has neglected the individual differences. So, we propose an improved slope one algorithm, which will consider the weights of the user. Finally, we experimentally evaluate our approach and compare it to the original Slope One. The experiment shows that our method provides better recommendation results than it.

Los sistemas de recomendación han tenido un rápido crecimiento tanto en investigación académica como en aplicaciones industriales en los últimos años. Las recomendaciones se nos proporcionan cuando elegimos elementos como música, películas, restaurantes, noticias, etc. Como parte de las técnicas de recomendación más prometedoras, los sistemas basados en recomendación de filtros colaborativos desempeñan un papel importante en las aplicaciones actuales. La aparición de nuevas fuentes de información requiere la combinación de sistemas de recomendación con técnicas avanzadas de otros campos, como el análisis de redes sociales, el procesamiento del lenguaje natural, el aprendizaje en profundidad, etc. Esta combinación mejora la calidad de las recomendaciones. Los sistemas de recomendación anteriores se centran principalmente en las mejoras de la precisión de la recomendación. Recientemente, surge una tendencia para la evaluación del rendimiento de los sistemas de recomendación que toma muchas propiedades nuevas en consideración. La fiabilidad es uno de los aspectos más notables y será profundamente estudiado en esta tesis. Se han propuesto algunas medidas para medir la fiabilidad de los sistemas de recomendación de filtrado colaborativo basados en la memoria. Sin embargo, observamos un área sin cubrir en el diseño de las medidas de fiabilidad adecuadas para los sistemas de recomendación de filtros colaborativos basados en modelos, especialmente aquellos basados en la factorización matricial. También encontramos que hay una falta de medidas de calidad de fiabilidad adecuadas. Esta tesis se concentra en proporcionar soluciones a ambas ausencias. Después de un estudio exhaustivo de trabajos de investigación del estado del arte, proponemos una nueva medida de fiabilidad que supera los desafíos de generalización y abstracción causados por los factores latentes de los métodos de factorización matricial. La medida propuesta se compara con con métodos y algoritmos representativos, usando bases de datos publicas, tanto académicas como comerciales. Los resultados experimentales muestran que nuestra nueva medida logra mejoras en la calidad de la recomendación. Además, la medida de fiabilidad propuesta también tiene un impacto muy positivo contra los "Shilling Attacks". Para obtener una mejora respecto a las medidas existentes de calidad de la fiabilidad, también proponemos, hasta donde sabemos, la primera medida de calidad de fiabilidad universal que es capaz de proporcionar valores numéricos para cualquier tipo de sistema de recomendación. Esta medida también se ha probado usando conjuntos de datos públicos reales y ha sido comparada con los métodos de calidad de la fiabilidad existentes. Basándonos en los resultados de los experimentos, concluimos que nuestra medida supera los últimos trabajos de investigación, proporcionando valores de calidad de fiabilidad más precisos e informativos. ----------ABSTRACT---------- Recommender systems have experienced a rapid growth in both academical research and industrial applications in latest years. Recommendations are provided to us when we choose items such as music, movies, restaurants, news, etc. As the most promising recommendation techniques, collaborative filtering recommender systems play an important role in current applications. The emergence of novel information sources requires to combine recommender systems with several advanced techniques coming from other fields like social network analysis, natural language processing, deep learning etc. This combination improves the quality of recommendations. Previous recommender systems research mainly focuses on the improvements of recommendation accuracy. Recently a new trend arises for the performance evaluation of recommender systems that takes many novel properties into consideration. Reliability is one of the most remarkable aspects, which will be deeply studied in this thesis. Some measures have been proposed for measuring reliability of memory based collaborative filtering recommender systems. However, we observe a gap in the design of proper reliability measures for model based collaborative filtering recommender systems, especially those based on matrix factorization. We also found that there is a lack in proper reliability quality measures. This thesis concentrates on providing solutions to both lacks. After a comprehensive study of state-of-the-art research works, we propose a novel reliability measure that overcomes the challenges of generality and abstraction caused by latent factors of matrix factorization methods. The proposed measure is tested against current baselines on public academic and commercial datasets; experiment results show that our novel measure achieves better improvements on recommendation quality. What’s more, the proposed reliability measure also has a very positive impact against shilling attacks. In order to fill in the blank of reliability quality measures, we also propose, as far as we know, the first universal reliability quality measure that is able to provide numerical values for any type of recommender systems. This measure has also been tested on public academic and commercial datasets using current methods. Based on the experiment results we conclude that our quality measure outperforms state-of-the-art research works by providing more accurate and informative reliability quality values.

Collaborative filtering recommendation systems that utilize graph convolutional neural networks (GCNs) often emphasize on training speed and accuracy, while overlooking the critical issue of fairness. This paper introduces FairGCF, a novel graph-based collaborative filtering model designed to address fairness concerns in GCN-based recommendations. FairGCF improves the representational capacity of graph collaborative filtering models by identifying dependencies between nodes in the graph, resulting in fairer and more accurate recommendations. The model constructs a dependency graph between users and items to capture higher order interaction patterns, aggregating features from users and items with similar dependencies to enrich their representations. Additionally, a fairness loss factor is incorporated to minimize disparities between positive and negative samples, thereby promoting fairness. Experiments on three public datasets demonstrate that FairGCF improves the Recall metric by an average of 6%, NDCG by 8.79%, Precision by 3.68%, and F1 score by 3.62%. Notably, the model is carefully designed to avoid bias toward specific users or items.