User-object Bipartite Network Research Articles

Characterizing the reputation of evaluators using vectors in the object feature space

Evaluation is a frequent occurrence in our daily lives, especially in online systems. Establishing ways to characterize the reputation of an evaluator is therefore becoming an important problem in online systems, and this has attracted much attention. Most existing evaluation methods use the rating information in user–object bipartite networks to evaluate the reputations of evaluators. These methods use a scalar quantity to represent the reputation of each user, which is applied to try to identify malicious ratings and spamming attacks. In this paper, we suggest that the reputation should be characterized by a vector, and propose a reputation–evaluation algorithm based on the classification of objects. More specifically, the objects are classified into several categories according to the community structure information obtained from a one-mode projection onto the objects of a bipartite network. The reputation vector of an evaluator then represents his/her reputation for the objects in different categories. The evaluator is also assigned an appropriate attribute according to his/her reputation vector using K-means clustering. The results from both artificial and real rating data have shown that the presented method is better than a correlation-based ranking method in terms of both accuracy and robustness.

Empirical analysis of the user reputation and clustering property for user-object bipartite networks

User reputation is of great significance for online rating systems which can be described by user-object bipartite networks, measuring the user ability of rating accurate assessments of various objects. The clustering coefficients have been widely investigated to analyze the local structural properties of complex networks, analyzing the diversity of user interest. In this paper, we empirically analyze the relation of user reputation and clustering property for the user-object bipartite networks. Grouping by user reputation, the results for the MovieLens dataset show that both the average clustering coefficient and the standard deviation of clustering coefficient decrease with the user reputation, which are different from the results that the average clustering coefficient and the standard deviation of clustering coefficient remain stable regardless of user reputation in the null model, suggesting that the user interest tends to be multiple and the diversity of the user interests is centralized for users with high reputation. Furthermore, we divide users into seven groups according to the user degree and investigate the heterogeneity of rating behavior patterns. The results show that the relation of user reputation and clustering coefficient is obvious for small degree users and weak for large degree users, reflecting an important connection between user degree and collective rating behavior patterns. This work provides a further understanding on the intrinsic association between user collective behaviors and user reputation.

A general growth model for online emerging user–object bipartite networks

The last couple of years have witnessed a huge surge in usage of the Internet due to the availability of several interactive applications of Web 2.0. Growth in both web users and varieties of online objects has been observed. In this paper, we propose a growth model for online emerging user–object bipartite networks to investigate selection behaviors of web users. According to model, both sets of users and objects grow constantly but edges arrive only from user set. The network evolves by arrival of external edges brought by new users and/or internal edges from old users. These external and internal edges are attached with a combination of preferential and random attachment mechanism. We have considered different attachment procedures for external edges of new users and internal edges of old users. To validate our model, we have taken nine different real online networks with distinct objects. Our results show good agreement between empirical data and theoretical model. We report mean absolute deviation (MAD), mean square error (MSE) and root mean square error (RMSE) for the proposed growth model. We emphasize on significant inferences about selection behaviors of web users after interpreting parameters’ values of the model. We also report detailed analysis of randomness in selection of new and old users that gives interesting inferences for each of the considered online networks.

Modeling online social signed networks

People’s online rating behavior can be modeled by user–object bipartite networks directly. However, few works have been devoted to reveal the hidden relations between users, especially from the perspective of signed networks. We analyze the signed monopartite networks projected by the signed user–object bipartite networks, finding that the networks are highly clustered with obvious community structure. Interestingly, the positive clustering coefficient is remarkably higher than the negative clustering coefficient. Then, a Signed Growing Network model (SGN) based on local preferential attachment is proposed to generate a user’s signed network that has community structure and high positive clustering coefficient. Other structural properties of the modeled networks are also found to be similar to the empirical networks.

Identifying online user reputation of user–object bipartite networks

Abstract Identifying online user reputation based on the rating information of the user–object bipartite networks is important for understanding online user collective behaviors. Based on the Bayesian analysis, we present a parameter-free algorithm for ranking online user reputation, where the user reputation is calculated based on the probability that their ratings are consistent with the main part of all user opinions. The experimental results show that the AUC values of the presented algorithm could reach 0.8929 and 0.8483 for the MovieLens and Netflix data sets, respectively, which is better than the results generated by the CR and IARR methods. Furthermore, the experimental results for different user groups indicate that the presented algorithm outperforms the iterative ranking methods in both ranking accuracy and computation complexity. Moreover, the results for the synthetic networks show that the computation complexity of the presented algorithm is a linear function of the network size, which suggests that the presented algorithm is very effective and efficient for the large scale dynamic online systems.

Uncovering the essential links in online commercial networks

Recommender systems are designed to effectively support individuals' decision-making process on various web sites. It can be naturally represented by a user-object bipartite network, where a link indicates that a user has collected an object. Recently, research on the information backbone has attracted researchers' interests, which is a sub-network with fewer nodes and links but carrying most of the relevant information. With the backbone, a system can generate satisfactory recommenda- tions while saving much computing resource. In this paper, we propose an enhanced topology-aware method to extract the information backbone in the bipartite network mainly based on the information of neighboring users and objects. Our backbone extraction method enables the recommender systems achieve more than 90% of the accuracy of the top-L recommendation, however, consuming only 20% links. The experimental results show that our method outperforms the alternative backbone extraction methods. Moreover, the structure of the information backbone is studied in detail. Finally, we highlight that the information backbone is one of the most important properties of the bipartite network, with which one can significantly improve the efficiency of the recommender system.

Popularity and user diversity of online objects

Abstract The popularity has been widely used to describe the object property of online user–object bipartite networks regardless of the user characteristics. In this paper, we introduce a measurement namely user diversity to measure diversity of users who select or rate one type of objects by using the information entropy. We empirically calculate the user diversity of objects with specific degree for both MovieLens and Diggs data sets. The results indicate that more types of users select normal-degree objects than those who select large-degree and small-degree objects. Furthermore, small-degree objects are usually selected by large-degree users while large-degree objects are usually selected by small-degree users. Moreover, we define 15% objects of smallest degrees as unpopular objects and 10% ones of largest degrees as popular objects. The timestamp is introduced to help further analyze the evolution of user diversity of popular objects and unpopular objects. The dynamic analysis shows that as objects become popular gradually, they are more likely accepted by small-degree users but lose attention among the large-degree users.

Information filtering via balanced diffusion on bipartite networks

The recent decade has witnessed the increasing popularity of recommender systems, which help users acquire relevant commodities and services from overwhelming resources on Internet. Some simple physical diffusion processes have been used to design effective recommendation algorithms for user–object bipartite networks, such as mass diffusion (MD) and heat conduction (HC) algorithms, which have different advantages respectively on accuracy and diversity. In this paper, we explore how to combine MD and HC processes to get better recommendation performance and propose a new algorithm mimicking the hybrid of MD and HC processes, named balanced diffusion (BD) algorithm. Numerical experiments on three benchmark data sets, MovieLens, Netflix and RateY ourMusic, show that BD algorithm outperforms three typical diffusion-like algorithms on the three important metrics, accuracy, diversity and novelty. Specifically, it not only provides accurate recommendation results, but also yields higher diversity and novelty in recommendations by accurately recommending unpopular objects.

Avoiding congestion in recommender systems

Recommender systems use the historical activities and personal profiles of users to uncover their preferences and recommend objects. Most of the previous methods are based on objects’ (and/or users’) similarity rather than on their difference. Such approaches are subject to a high risk of increasingly exposing users to a narrowing band of popular objects. As a result, a few objects may be recommended to an enormous number of users, resulting in the problem of recommendation congestion, which is to be avoided, especially when the recommended objects are limited resources. In order to quantitatively measure a recommendation algorithmʼs ability to avoid congestion, we proposed a new metric inspired by the Gini index, which is used to measure the inequality of the individual wealth distribution in an economy. Besides this, a new recommendation method called directed weighted conduction (DWC) was developed by considering the heat conduction process on a user–object bipartite network with different thermal conductivities. Experimental results obtained for three benchmark data sets showed that the DWC algorithm can effectively avoid system congestion, and greatly improve the novelty and diversity, while retaining relatively high accuracy, in comparison with the state-of-the-art methods.

Personal Recommendation Via Heterogeneous Diffusion on Bipartite Network

Recommender systems have proven to be an effective method to deal with the problem of information overload in finding interesting products. It is still a challenge to increase the accuracy and diversity of recommendation algorithms to fulfill users' preferences. To provide a better solution, in this paper, we propose a novel recommendation algorithm based on heterogeneous diffusion process on a user-object bipartite network. This algorithm generates personalized recommendation results on the basis of the physical dynamic feature of resources diffusion which is influenced by objects' degrees and users' interest degrees. Detailed numerical analysis on two benchmark datasets shows that the presented algorithm is of high accuracy, and also generates more diversity.

Ranking Reputation and Quality in Online Rating Systems

How to design an accurate and robust ranking algorithm is a fundamental problem with wide applications in many real systems. It is especially significant in online rating systems due to the existence of some spammers. In the literature, many well-performed iterative ranking methods have been proposed. These methods can effectively recognize the unreliable users and reduce their weight in judging the quality of objects, and finally lead to a more accurate evaluation of the online products. In this paper, we design an iterative ranking method with high performance in both accuracy and robustness. More specifically, a reputation redistribution process is introduced to enhance the influence of highly reputed users and two penalty factors enable the algorithm resistance to malicious behaviors. Validation of our method is performed in both artificial and real user-object bipartite networks.

Information Filtering in Sparse Online Systems: Recommendation via Semi-Local Diffusion

With the rapid growth of the Internet and overwhelming amount of information and choices that people are confronted with, recommender systems have been developed to effectively support users’ decision-making process in the online systems. However, many recommendation algorithms suffer from the data sparsity problem, i.e. the user-object bipartite networks are so sparse that algorithms cannot accurately recommend objects for users. This data sparsity problem makes many well-known recommendation algorithms perform poorly. To solve the problem, we propose a recommendation algorithm based on the semi-local diffusion process on the user-object bipartite network. The simulation results on two sparse datasets, Amazon and Bookcross, show that our method significantly outperforms the state-of-the-art methods especially for those small-degree users. Two personalized semi-local diffusion methods are proposed which further improve the recommendation accuracy. Finally, our work indicates that sparse online systems are essentially different from the dense online systems, so it is necessary to reexamine former algorithms and conclusions based on dense data in sparse systems.

Effect of the social influence on topological properties of user-object bipartite networks

Social influence plays an important role in analyzing online users’ collective behaviors [Salganik et al., Science 311, 854 (2006)]. However, the effect of the social influence from the viewpoint of theoretical model is missing. In this paper, by taking into account the social influence and users’ preferences, we develop a theoretical model to analyze the topological properties of user-object bipartite networks, including the degree distribution, average nearest neighbor degree and the bipartite clustering coefficient, as well as topological properties of the original user-object networks and their unipartite projections. According to the users’ preferences and the global ranking effect, we analyze the theoretical results for two benchmark data sets, Amazon and Bookcrossing, which are approximately consistent with the empirical results. This work suggests that this model is feasible to analyze topological properties of bipartite networks in terms of the social influence and the users’ preferences.

Membership in social networks and the application in information filtering

During the past few years, users’ membership in the online system (i.e. the social groups that online users joined) were widely investigated. Most of these works focus on the detection, formulation and growth of online communities. In this paper, we study users’ membership in a coupled system which contains user-group and user-object bipartite networks. By linking users’ membership information and their object selection, we find that the users who have collected only a few objects are more likely to be “influenced” by the membership when choosing objects. Moreover, we observe that some users may join many online communities though they collected few objects. Based on these findings, we design a social diffusion recommendation algorithm which can effectively solve the user cold-start problem. Finally, we propose a personalized combination of our method and the hybrid method in [T. Zhou, Z. Kuscsik, J.G. Liu, M. Medo, J.R. Wakeling, Y.C. Zhang, Proc. Natl. Acad. Sci. 107, 4511 (2010)], which leads to a further improvement in the overall recommendation performance.

EMPIRICAL ANALYSIS OF THE CLUSTERING COEFFICIENT IN THE USER-OBJECT BIPARTITE NETWORKS

The clustering coefficient of the bipartite network, C4, has been widely used to investigate the statistical properties of the user-object systems. In this paper, we empirically analyze the evolution patterns of C4 for a nine year MovieLens data set, where C4 is used to describe the diversity of the user interest. First, we divide the MovieLens data set into fractions according to the time intervals and calculate C4 of each fraction. The empirical results show that, the diversity of the user interest changes periodically with a round of one year, which reaches the smallest value in spring, then increases to the maximum value in autumn and begins to decrease in winter. Furthermore, a null model is proposed to compare with the empirical results, which is constructed in the following way. Each user selects each object with a turnable probability p, and the numbers of users and objects are equal to that of the real MovieLens data set. The comparison result indicates that the user activity has greatly influenced the structure of the user-object bipartite network, and users with the same degree information may have two totally different clustering coefficients. On the other hand, the same clustering coefficient also corresponds to different degrees. Therefore, we need to take the clustering coefficient into consideration together with the degree information when describing the user selection activity.

Extracting the Information Backbone in Online System

Information overload is a serious problem in modern society and many solutions such as recommender system have been proposed to filter out irrelevant information. In the literature, researchers have been mainly dedicated to improving the recommendation performance (accuracy and diversity) of the algorithms while they have overlooked the influence of topology of the online user-object bipartite networks. In this paper, we find that some information provided by the bipartite networks is not only redundant but also misleading. With such “less can be more” feature, we design some algorithms to improve the recommendation performance by eliminating some links from the original networks. Moreover, we propose a hybrid method combining the time-aware and topology-aware link removal algorithms to extract the backbone which contains the essential information for the recommender systems. From the practical point of view, our method can improve the performance and reduce the computational time of the recommendation system, thus improving both of their effectiveness and efficiency.

Heat conduction information filtering via local information of bipartite networks

Information filtering based on structure properties of user-object bipartite networks is of both theoretical interest and practical significance in modern science. In this paper, we empirically investigate the framework of heat-conduction-based (HC) information filtering [Y.-C. Zhang et al., Phys. Rev. Lett. 99, 154301 (2007)] in terms of the local node similarity. We compare nine well-known local similarity measures on four real networks. The results indicate that the local-heat-conduction-based similarity has the best accuracy and diversity simultaneously. Embedding the object degree effect into the heat conduction process, we present a new user similarity measure. Experimental results on four real networks demonstrate that the improved similarity measure could generate remarkably higher diversity and novelty results than the state-of-the-art HC information filtering algorithms based on local information, and the accuracy is also increased greatly or approximately unchanged. Since the improved similarity index only need the local information of user-object bipartite networks, it is therefore a strong candidate for potential application in information filtering of large-scale bipartite networks.

INFORMATION FILTERING VIA CLUSTERING COEFFICIENTS OF USER–OBJECT BIPARTITE NETWORKS

The clustering coefficient of user–object bipartite networks is presented to evaluate the overlap percentage of neighbors rating lists, which could be used to measure interest correlations among neighbor sets. The collaborative filtering (CF) information filtering algorithm evaluates a given user's interests in terms of his/her friends' opinions, which has become one of the most successful technologies for recommender systems. In this paper, different from the object clustering coefficient, users' clustering coefficients of user–object bipartite networks are introduced to improve the user similarity measurement. Numerical results for MovieLens and Netflix data sets show that users' clustering effects could enhance the algorithm performance. For MovieLens data set, the algorithmic accuracy, measured by the average ranking score, can be improved by 12.0% and the diversity could be improved by 18.2% and reach 0.649 when the recommendation list equals to 50. For Netflix data set, the accuracy could be improved by 14.5% at the optimal case and the popularity could be reduced by 13.4% comparing with the standard CF algorithm. Finally, we investigate the sparsity effect on the performance. This work indicates the user clustering coefficients is an effective factor to measure the user similarity, meanwhile statistical properties of user–object bipartite networks should be investigated to estimate users' tastes.

Negative ratings play a positive role in information filtering

The explosive growth of information asks for advanced information filtering techniques to solve the so-called information overload problem. A promising way is the recommender system which analyzes the historical records of users’ activities and accordingly provides personalized recommendations. Most recommender systems can be represented by user-object bipartite networks where users can evaluate and vote for objects, and ratings such as “dislike” and “I hate it” are treated straightforwardly as negative factors or are completely ignored in traditional approaches. Applying a local diffusion algorithm on three benchmark data sets, MovieLens, Netflix and Amazon, our study arrives at a very surprising result, namely the negative ratings may play a positive role especially for very sparse data sets. In-depth analysis at the microscopic level indicates that the negative ratings from less active users to less popular objects could probably have positive impacts on the recommendations, while the ones connecting active users and popular objects mostly should be treated negatively. We finally outline the significant relevance of our results to the two long-term challenges in information filtering: the sparsity problem and the cold-start problem.

Information filtering via preferential diffusion

Recommender systems have shown great potential in addressing the information overload problem, namely helping users in finding interesting and relevant objects within a huge information space. Some physical dynamics, including the heat conduction process and mass or energy diffusion on networks, have recently found applications in personalized recommendation. Most of the previous studies focus overwhelmingly on recommendation accuracy as the only important factor, while overlooking the significance of diversity and novelty that indeed provide the vitality of the system. In this paper, we propose a recommendation algorithm based on the preferential diffusion process on a user-object bipartite network. Numerical analyses on two benchmark data sets, MovieLens and Netflix, indicate that our method outperforms the state-of-the-art methods. Specifically, it can not only provide more accurate recommendations, but also generate more diverse and novel recommendations by accurately recommending unpopular objects.