Recommendation in evolving online networks

Recommender system is an effective tool to find the most relevant information for onlineusers. By analyzing the historical selection records of users, recommender system predictsthe most likely future links in the user-item network and accordingly constructs apersonalized recommendation list for each user. So far, the recommendation process ismostly investigated in static user-item networks. In this paper, we propose a model whichallows us to examine the performance of the state-of-the-art recommendation algorithms inevolving networks. We find that the recommendation accuracy in general decreases with timeif the evolution of the online network fully depends on the recommendation. Interestingly,some randomness in users’ choice can significantly improve the long-term accuracy of therecommendation algorithm. When a hybrid recommendation algorithm is applied, we find thatthe optimal parameter gradually shifts towards the diversity-favoring recommendationalgorithm, indicating that recommendation diversity is essential to keep a high long-termrecommendation accuracy. Finally, we confirm our conclusions by studying therecommendation on networks with the real evolution data.     

MIGAN: Mutual-Interaction Graph Attention Network for Collaborative Filtering

Many web platforms now include recommender systems. Network representation learning has been a successful approach for building these efficient recommender systems. However, learning the mutual influence of nodes in the network is challenging. Indeed, it carries collaborative signals accounting for complex user-item interactions on user decisions. For this purpose, in this paper, we develop a Mutual Interaction Graph Attention Network “MIGAN”, a new algorithm based on self-supervised representation learning on a large-scale bipartite graph (BGNN). Experimental investigation with real-world data demonstrates that MIGAN compares favorably with the baselines in terms of prediction accuracy and recommendation efficiency.     

Heterogeneity,quality, and reputation in an adaptive recommendation model

Recommender systems help people cope with the problem of information overload. A recently proposed adaptive news recommender model [M. Medo, Y.-C. Zhang, T. Zhou, Europhys. Lett. 88, 38005 (2009)] is based on epidemic-like spreading of news in a social network. By means of agent-based simulations we study a “good get richer” feature of the model and determine which attributes are necessary for a user to play a leading role in the network. We further investigate the filtering efficiency of the model as well as its robustness against malicious and spamming behaviour. We show that incorporating user reputation in the recommendation process can substantially improve the outcome.     

Personalized recommendation via integrated diffusion on user-item-tag tripartite graphs

Personalized recommender systems are confronting great challenges of accuracy, diversification and novelty, especially when the data set is sparse and lacks accessorial information, such as user profiles, item attributes and explicit ratings. Collaborative tags contain rich information about personalized preferences and item contents, and are therefore potential to help in providing better recommendations. In this article, we propose a recommendation algorithm based on an integrated diffusion on user-item-tag tripartite graphs. We use three benchmark data sets, Del.icio.us, MovieLens and BibSonomy, to evaluate our algorithm. Experimental results demonstrate that the usage of tag information can significantly improve accuracy, diversification and novelty of recommendations.     

Time-Aware Explainable Recommendation via Updating Enabled Online Prediction

There has been growing attention on explainable recommendation that is able to provide high-quality results as well as intuitive explanations. However, most existing studies use offline prediction strategies where recommender systems are trained once while used forever, which ignores the dynamic and evolving nature of user–item interactions. There are two main issues with these methods. First, their random dataset split setting will result in data leakage that knowledge should not be known at the time of training is utilized. Second, the dynamic characteristics of user preferences are overlooked, resulting in a model aging issue where the model’s performance degrades along with time. In this paper, we propose an updating enabled online prediction framework for the time-aware explainable recommendation. Specifically, we propose an online prediction scheme to eliminate the data leakage issue and two novel updating strategies to relieve the model aging issue. Moreover, we conduct extensive experiments on four real-world datasets to evaluate the effectiveness of our proposed methods. Compared with the state-of-the-art, our time-aware approach achieves higher accuracy results and more convincing explanations for the entire lifetime of recommendation systems, i.e., both the initial period and the long-term usage.     

Behavior patterns of online users and the effect on information filtering

Understanding the structure and evolution of web-based user-item bipartite networks is an important task since they play a fundamental role in online information filtering. In this paper, we focus on investigating the patterns of online users’ behavior and the effect on recommendation process. Empirical analysis on the e-commercial systems show that users’ taste preferences are heterogeneous in general but their interests for niche items are highly clustered. Additionally, recommendation processes are investigated on both the real networks and the reshuffled networks in which real users’ behavior patterns can be gradually destroyed. We find that the performance of personalized recommendation methods is strongly related to the real network structure. Detailed study on each item shows that most hot items are accurately recommended and their recommendation accuracy is robust to the reshuffling process. However, the accuracy for niche items is relatively low and drops significantly after removing users’ behavior patterns. Our work is also meaningful in practical sense since it reveals an effective direction to improve the accuracy and the robustness of the existing recommender systems.     

Preference of online users and personalized recommendations

In a recent work [T. Zhou, Z. Kuscsik, J.-G. Liu, M. Medo, J.R. Wakeling, Y.-C. Zhang, Proc. Natl. Acad. Sci. 107 (2010) 4511], a personalized recommendation algorithm with high performance in both accuracy and diversity is proposed. This method is based on the hybridization of two single algorithms called probability spreading and heat conduction, which respectively are inclined to recommend popular and unpopular products. With a tunable parameter, an optimal balance between these two algorithms in system level is obtained. In this paper, we apply this hybrid method in individual level, namely each user has his/her own personalized hybrid parameter to adjust. Interestingly, we find that users are quite different in personalized hybrid parameters and the recommendation performance can be significantly improved if each user is assigned with his/her optimal personalized hybrid parameter. Furthermore, we find that users’ personalized parameters are negatively correlated with users’ degree but positively correlated with the average degree of the items collected by each user. With these understandings, we propose a strategy to assign users with suitable personalized parameters, which leads to a further improvement of the original hybrid method. Finally, our work highlights the importance of considering the heterogeneity of users in recommendation.     

Recommender systems

The ongoing rapid expansion of the Internet greatly increases the necessity of effective recommender systems for filtering the abundant information. Extensive research for recommender systems is conducted by a broad range of communities including social and computer scientists, physicists, and interdisciplinary researchers. Despite substantial theoretical and practical achievements, unification and comparison of different approaches are lacking, which impedes further advances. In this article, we review recent developments in recommender systems and discuss the major challenges. We compare and evaluate available algorithms and examine their roles in the future developments. In addition to algorithms, physical aspects are described to illustrate macroscopic behavior of recommender systems. Potential impacts and future directions are discussed. We emphasize that recommendation has great scientific depth and combines diverse research fields which makes it interesting for physicists as well as interdisciplinary researchers.     

Towards novelty-driven recommender systems

We get recommendations about everything and in a pervasive way. Recommender systems act like compasses for our journey in complex conceptual spaces and we more and more rely on recommendations to ground most of our decisions. Despite their extraordinary efficiency and reliability, recommender systems are far from being flawless. They display instead serious drawbacks that might seriously reduce our open-mindedness and our capacity of experiencing diversity and possibly conflicting views. In this paper, we carefully investigate the very foundations of recommendation algorithms in order to identify the determinants of what could be the next generation of recommender systems. We postulate that it is possible to overcome the limitations of current recommender systems, by getting inspiration from the way in which people seek for novelties and give value to new experiences. From this perspective, the notion of adjacent possible seems a relevant one to redesign recommender systems in a way that better aligns with the natural inclination of human beings towards new and pleasant experiences. We claim that this new generation of recommenders could help in overcoming the pitfalls of current technologies, namely the tendency towards a lack of diversity, polarization, the emergence of echo-chambers and misinformation.     

QCF: Quantum Collaborative Filtering Recommendation Algorithm

With the rapid development of the Internet, e-commerce plays an important role in people’s lives, and the recommendation system is one of the most critical technologies. However, as the number of users and the scale of goods increase sharply, the traditional collaborative filtering recommendation algorithm has a large computational complexity in the part of calculating the user similarity, which leads to a low recommendation efficiency. In response to the above problems, this paper introduces the concept of quantum computing theory. The user score vector is first prepared into a quantum state, the similarity score is calculated in parallel, then the similarity information is saved into the quantum bit, and finally the similar user is searched by the Grover search algorithm. Compared with the traditional collaborative filtering recommendation algorithm, the time complexity of the collaborative filtering recommendation algorithm based on Grover algorithm can be effectively reduced under certain conditions.

     

基于视频推荐技术的无线网络视频服务器设计

无线网络视频服务器中视频推荐技术已成为重要技术之一,视频推荐技术是为了用户在使用无线网络是视频务器时,更快的找到感兴趣的视频。采用当前方法对用户进行视频推荐时,未考虑用户的兴趣偏好是否随着时间有所变化,使视频推荐出现偏差。为此,提出一种基于视频推荐技术的无线网络视频服务器设计方法。该方法首先使用无线网络视频服务器硬件部分的MPEG-4进行视频数据采集,并对MPEG-4采集的视频数据进行整理。在由软件部分把采集到的视频数据储存到缓冲区,进行视频缓冲,并建立视频数据队列进行视频数据输送。以计算无线网络视频的相似度来搜索相近视频,将搜索到的视频与目标用户观看过的视频进行对比,对比相似度越接近1,则说明用户对推荐视频感兴趣的几率大,反之越接近-1用户对推荐视频越不感兴趣。此计算方法能有效的从海量视频数据中快速的搜索出目标用户感兴趣视频。实验结果表明,将视频推荐技术应用到无线网络视频中可以迅速准确的搜索出目标用户感兴趣视频。     

An improved recommendation algorithm via weakening indirect linkage effect

We propose an indirect-link-weakened mass diffusion method(IMD), by considering the indirect linkage and the source object heterogeneity effect in the mass diffusion(MD) recommendation method. Experimental results on the MovieLens, Netflix, and RYM datasets show that, the IMD method greatly improves both the recommendation accuracy and diversity, compared with a heterogeneity-weakened MD method(HMD), which only considers the source object heterogeneity. Moreover, the recommendation accuracy of the cold objects is also better elevated in the IMD than the HMD method. It suggests that eliminating the redundancy induced by the indirect linkages could have a prominent effect on the recommendation efficiency in the MD method.     

Adaptive social recommendation in a multiple category landscape

People in the Internet era have to cope with the information overload, striving to find what they are interested in, and usually face this situation by following a limited number of sources or friends that best match their interests. A recent line of research, namely adaptive social recommendation, has therefore emerged to optimize the information propagation in social networks and provide users with personalized recommendations. Validation of these methods by agent-based simulations often assumes that the tastes of users can be represented by binary vectors, with entries denoting users’ preferences. In this work we introduce a more realistic assumption that users’ tastes are modeled by multiple vectors. We show that within this framework the social recommendation process has a poor outcome. Accordingly, we design novel measures of users’ taste similarity that can substantially improve the precision of the recommender system. Finally, we discuss the issue of enhancing the recommendations’ diversity while preserving their accuracy.     

Weighted Similarity and Core-User-Core-Item Based Recommendations

In traditional recommendation algorithms, the users and/or the items with the same rating scores are equally treated. In real world, however, a user may prefer some items to other items and some users are more loyal to a certain item than other users. In this paper, therefore, we propose a weighted similarity measure by exploiting the difference in user-item relationships. In particular, we refer to the most important item of a user as his core item and the most important user of an item as its core user. We also propose a Core-User-Item Solver (CUIS) to calculate the core users and core items of the system, as well as the weighting coefficients for each user and each item. We prove that the CUIS algorithm converges to the optimal solution efficiently. Based on the weighted similarity measure and the obtained results by CUIS, we also propose three effective recommenders. Through experiments based on real-world data sets, we show that the proposed recommenders outperform corresponding traditional-similarity based recommenders, verify that the proposed weighted similarity can improve the accuracy of the similarity, and then improve the recommendation performance.     

Comprehensive Empirical Evaluation of Deep Learning Approaches for Session-Based Recommendation in E-Commerce

Boosting the sales of e-commerce services is guaranteed once users find more items matching their interests in a short amount of time. Consequently, recommendation systems have become a crucial part of any successful e-commerce service. Although various recommendation techniques could be used in e-commerce, a considerable amount of attention has been drawn to session-based recommendation systems in recent years. This growing interest is due to security concerns over collecting personalized user behavior data, especially due to recent general data protection regulations. In this work, we present a comprehensive evaluation of the state-of-the-art deep learning approaches used in the session-based recommendation. In session-based recommendation, a recommendation system counts on the sequence of events made by a user within the same session to predict and endorse other items that are more likely to correlate with their preferences. Our extensive experiments investigate baseline techniques (e.g., nearest neighbors and pattern mining algorithms) and deep learning approaches (e.g., recurrent neural networks, graph neural networks, and attention-based networks). Our evaluations show that advanced neural-based models and session-based nearest neighbor algorithms outperform the baseline techniques in most scenarios. However, we found that these models suffer more in the case of long sessions when there exists drift in user interests, and when there are not enough data to correctly model different items during training. Our study suggests that using the hybrid models of different approaches combined with baseline algorithms could lead to substantial results in session-based recommendations based on dataset characteristics. We also discuss the drawbacks of current session-based recommendation algorithms and further open research directions in this field.     

Combining content and relation analysis for recommendation in social tagging systems

Social tagging is one of the most important ways to organize and index online resources. Recommendation in social tagging systems, e.g. tag recommendation, item recommendation and user recommendation, is used to improve the quality of tags and to ease the tagging or searching process. Existing works usually provide recommendations by analyzing relation information in social tagging systems, suffering a lot from the over sparse problem. These approaches ignore information contained in the content of resources, which we believe should be considered to improve recommendation quality and to deal with the over sparse problem. In this paper we propose a recommendation approach for social tagging systems that combines content and relation analysis in a single model. By modeling the generating process of social tagging systems in a latent Dirichlet allocation approach, we build a fully generative model for social tagging, leverage it to estimate the relation between users, tags and resources and achieve tag, item and user recommendation tasks. The model is evaluated using a CiteULike data snapshot, and results show improvements in metrics for various recommendation tasks.     

Personalized recommendation via an improved NBI algorithm and user influence model in a Microblog network

Bipartite network based recommendations have attracted extensive attentions in recent years. Differing from traditional object-oriented recommendations, the recommendation in a Microblog network has two crucial differences. One is high authority users or one’s special friends usually play a very active role in tweet-oriented recommendation. The other is that the object in a Microblog network corresponds to a set of tweets on same topic instead of an actual and single entity, e.g. goods or movies in traditional networks. Thus repeat recommendations of the tweets in one’s collected topics are indispensable. Therefore, this paper improves network based inference (NBI) algorithm by original link matrix and link weight on resource allocation processes. This paper finally proposes the Microblog recommendation model based on the factors of improved network based inference and user influence model. Adjusting the weights of these two factors could generate the best recommendation results in algorithm accuracy and recommendation personalization.     

Preference-Tree-Based Real-Time Recommendation System

In the current era of online information overload, recommendation systems are very useful for helping users locate content that may be of interest to them. A personalized recommendation system presents content based on information such as a user’s browsing history and the videos watched. However, information filtering-based recommendation systems are vulnerable to data sparsity and cold-start problems. Additionally, existing recommendation systems suffer from the large overhead incurred in learning regression models used for preference prediction or in selecting groups of similar users. In this study, we propose a preference-tree-based real-time recommendation system that uses various tree models to predict user preferences with a fast runtime. The proposed system predicts preferences based on two balance constants and one similarity threshold to recommend content with a high accuracy while balancing generalized and personalized preferences. The results of comparative experiments and ablation studies confirm that the proposed system can accurately recommend content to users. Specifically, we confirmed that the accuracy and novelty of the recommended content were, respectively, improved by 12.1% and 27.2% compared to existing systems. Furthermore, we verified that the proposed system satisfies real-time requirements and mitigates both cold-start and overfitting problems.     

Information filtering via weighted heat conduction algorithm

In this paper, by taking into account effects of the user and object correlations on a heat conduction (HC) algorithm, a weighted heat conduction (WHC) algorithm is presented. We argue that the edge weight of the user-object bipartite network should be embedded into the HC algorithm to measure the object similarity. The numerical results indicate that both the accuracy and diversity could be improved greatly compared with the standard HC algorithm and the optimal values reached simultaneously. On the Movielens and Netflix datasets, the algorithmic accuracy, measured by the average ranking score, can be improved by 39.7% and 56.1% in the optimal case, respectively, and the diversity could reach 0.9587 and 0.9317 when the recommendation list equals to 5. Further statistical analysis indicates that, in the optimal case, the distributions of the edge weight are changed to the Poisson form, which may be the reason why HC algorithm performance could be improved. This work highlights the effect of edge weight on a personalized recommendation study, which maybe an important factor affecting personalized recommendation performance.     

Personal artist recommendation via a listening and trust preference network

Trust information provided by a user unfolds his/her reliable friends with similar tastes. It not only has the potential to help provide better recommendations but also emancipates the recommendation process from heavy computation for seeking friends. In this paper, by taking into account the latent value of trust information, our personal artist recommendation algorithm via a listening and trust preference network (LTPN for short) is presented. We argue that the excellent recommendation should be acquired via the listening and trust preference network instead of the original listening and trust relation information. Experimental results demonstrate LTPN can not only provide better recommendation but also help relieve the cold start problem caused by new users.