Improving top-K recommendation with truster and trustee relationship in user trust network

doi:10.1016/j.ins.2016.09.024

Information Sciences

Volume 374, 20 December 2016, Pages 100-114

https://doi.org/10.1016/j.ins.2016.09.024 Get rights and content

Abstract

Due to the data sparsity problem, social network information is often additionally used to improve the performance of recommender systems. While most existing works exploit social information to reduce the rating prediction error, e.g., RMSE, a few had aimed to improve the top-k ranking prediction accuracy. This paper proposes a novel top-k ranking oriented recommendation method, TRecSo, which incorporates social information into recommendation by modeling two different roles of users as trusters and trustees while considering the structural information of the network. Empirical studies on real-world datasets demonstrate that TRecSo leads to a remarkable improvement compared with previous methods in top-k recommendation.

Introduction

Due to the ever increasing amount of information available to users on the Web, recommender systems have recently gained their popularity in industrial companies such as Amazon¹, Netflix², Facebook³, to name a few. Indeed, it has been reported that Amazon are generating about 35% of their revenue through personalized recommendations provided by their own recommendation algorithms [21]. Moreover, recommender systems are widely used to help academic researchers in their decision making process when searching for relevant scientific articles [37], [38], [41]. Hence, improving the quality of recommender systems has been an attractive research problem for academic researchers.

In the beginning, a significant amount of research was devoted to accurately predicting the overall ratings to reduce the rating prediction error, e.g., RMSE, MAE [2], [13]. Given a user-item rating matrix, their main goal was to provide the correct rating that a user will give to an item. However, rather than better predicted ratings, users are more interested in seeing a list of top-k items that aligns with their preferences. Moreover, it is well known that minimizing the rating prediction error does not always result in better top-k list of items [6], [46]. Consequently, the Learning-To-Rank (LTR) [16] method, a supervised machine learning method that directly builds a ranking list from training data without an intermediate step of the rating prediction, has gained popularity to provide accurate results at top-k.

Despite the success of recommendation approaches, recommender systems suffer from an inherent limitation called the data sparsity problem, that is, the recommendation is hardly accurate due to lack of observations (i.e., ratings) because users typically rate a small number of items. To tackle the data sparsity problem, researchers have tried to incorporate auxiliary information such as social network relationship among users [7], [8], [9], [17], [19], text reviews on items [15], [23], [41], time related information [1], [14], [43] and items’ quality information [28], [37], [38]. Specifically, this paper focuses on incorporating the social network information of users in the top-k recommendation.

Recently, two top-k recommendation methods have been developed to incorporate the social network information based on the LTR approach. Specifically, Yao et al. [48] linearly combine a user's taste and her direct friends' tastes in optimizing the top-k recommendation. However, they do not utilize other important information hidden in the social network such as the structural information or truster-trustee relationship [45]. Zhao et al. [49] optimize the top-k recommendation from relative ordering that can be extracted from purchase history or browsing history, but they cannot handle numerical ratings directly. Note that numerical ratings usually contain much richer information on users preference than relative ordering.

This paper proposes a novel LTR-based top-k recommendation method, TRecSo, which leverages the social network information to optimize top-k recommendation. TRecSo is distinguished from previous methods in that it models two different roles of users as trusters and trustees while considering the structural information of the network. Precisely, we map users into two types of low dimensional spaces according to their roles, that is, truster space and trustee space under the following assumptions:

•
When “user A” is given several choices of items, he turns to people he trusts to ask them their opinions about the items.
•
Consequently, the behavior of “user A” will influence the people that trust “user A”.

We summarize our contributions as follows:

•
We develop a novel Learning-To-Rank (LTR) based top-k recommendation method that takes into account the social network information among users to alleviate the data sparsity problem.
•
We map users into two types of low dimensional spaces according to their roles while considering the structural information of the trust network.
•
Our experimental results on three real-world datasets indicate that our method considerably outperforms previous methods in top-k recommendation.

The remainder of this paper is organized as follows. Section 2 describes the related work; The problem is formally discussed in Section 3; In Section 4, we describe our proposed method, TRecSo, and demonstrate the learning algorithm; In Section 5, we describe the complexity of our proposed method followed by experimental results in Section 6. Finally in Section 7, we conclude the paper.

Section snippets

Related work

Triggered by the Netflix prize [2] that ended in 2009, the field of recommender system has seen rapid progress since then. In this section, we review the existing methods including traditional collaborative filtering methods, social network based recommender systems, top-k ranking oriented recommender systems and top-k ranking oriented social recommender systems, the most related work to ours.

Problem description

We first introduce the notations that we use throughout the paper. Let $U = {u_{1}, u_{2}, \dots, u_{N}}$ be the set of users and $V = {v_{1}, v_{2}, \dots, v_{M}}$ be the set of items, where N and M are number of users and items, respectively. The ratings given by users in $U$ on items in $V$ are represented by a rating matrix $R = {[r_{i j}]}_{N \times M},$ where r_ij denotes the rating of u_i on v_j. Depending on applications, r_ij can be either a real number or a binary value. When users explicitly express their opinions on items, the rating value is a real

Method

In Section 4.1, we briefly explain the concept of Plackett-Luce model that is required for understanding our proposed model, TRecSo. Then, we explain how the rating and the trust relationship are modeled in Sections 4.2 and 4.3, respectively, and propose a unified model that combines the rating and the trust in Section 4.4, where we show how the unified model is learned.

Complexity analysis

The computational time of learning the proposed model TRecSo is dominated by the computation of the loss function $L$ in Eq. (10) and its gradients with respect to feature vectors given in Eqs. (11)–(19). Let $| R |$ and $| S |$ be the number of observed ratings and trust relations, respectively. Then, the complexity of evaluating $L$ is $O (K | R | c + d | S |),$ where K is the size of latent dimensionality, $c = m a x (a, b)$ and $d = m a x (b, K)$ . Note that a and b are the average ratings an item has received and the average

Experiments

In this section, we carry out experiments to compare the quality of the top-k recommendation of our method, TRecSo, with several state-of-the-art methods on three real-world datasets. Our experiments are designed to verify the following questions:

1.
How does TRecSo perform compared with other related competitors?
2.
Does considering the social network structure as in Eq. (8) enhance the performance of TRecSo?
3.
How does the trade-off parameters of TRecSo affect the quality of the top-k recommendation?
4.

Conclusion and future work

This paper proposes TRecSo, a novel LTR based recommendation method that optimizes the top-k ranking prediction accuracy by additionally considering the social network information. Specifically, TRecSo integrates the social network information into the Learning-To-Rank (LTR) based objective function for recommendation. Thanks to the flexibility (can be generalized to symmetric social relationship) and the low complexity (compared with pair-wise LTR approaches) of our model, our proposed method

Acknowledgment

This work was supported by the Industrial Core Technology Development Program (10049079 , Development of Mining core technology exploiting personal big data) funded by the Ministry of Trade, Industry and Energy (MOTIE, Korea), ICT R&D program of MSIP/IITP [14-824-09-014, Basic Software Research in Human-level Lifelong Machine Learning (Machine Learning Center)]

References (51)

C. Porcel et al.
Dealing with incomplete information in a fuzzy linguistic recommender system to disseminate information in university digital libraries
Knowl. Based Syst.
(2010)
A. Tejeda-Lorente et al.
Integrating quality criteria in a fuzzy linguistic recommender system for digital libraries
Proc. Comput. Sci.
(2014)
A. Tejeda-Lorente et al.
A quality based recommender system to disseminate information in a university digital library
Inf. Sci.
(2014)
G. Adomavicius et al.
Context-aware recommender systems
Recommender Systems Handbook
(2011)
J. Bennett et al.
The netflix prize
Proceedings of KDD Cup and Workshop
(2007)
J.S. Breese et al.
Empirical analysis of predictive algorithms for collaborative filtering
Proceedings of the Fourteenth Conference on Uncertainty in Artificial Intelligence
(1998)
CaoB. et al.
Transfer learning for collective link prediction in multiple heterogenous domains
Proceedings of the 27th International Conference on Machine Learning (ICML-10)
(2010)
CaoZ. et al.
Learning to rank: from pairwise approach to listwise approach
Proceedings of the 24th International Conference on Machine Learning
(2007)
P. Cremonesi et al.
Performance of recommender algorithms on top-n recommendation tasks
Proceedings of the Fourth ACM Conference on Recommender Systems
(2010)
GuoG. et al.
A novel bayesian similarity measure for recommender systems
IJCAI
(2013)

GuoG. et al.

Etaf: an extended trust antecedents framework for trust prediction

ASONAM

(2014)

GuoG. et al.

Trustsvd: collaborative filtering with both the explicit and implicit influence of user trust and of item ratings.

AAAI

(2015)

T. Hofmann

Latent semantic models for collaborative filtering

ACM Trans. Inf. Syst.

(2004)

M. Jamali et al.

Trustwalker: a random walk model for combining trust-based and item-based recommendation

Proceedings of the 15th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining

(2009)

M. Jamali et al.

A matrix factorization technique with trust propagation for recommendation in social networks

Proceedings of the Fourth ACM Conference on Recommender Systems

(2010)

Y. Koren

Factorization meets the neighborhood: a multifaceted collaborative filtering model

Proceedings of the 14th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining

(2008)

Y. Koren

Collaborative filtering with temporal dynamics

Commun. ACM

(2010)

LingG. et al.

Ratings meet reviews, a combined approach to recommend

Proceedings of the 8th ACM Conference on Recommender Systems

(2014)

LiuT.-Y.

Learning to rank for information retrieval

Found. Trends Inf. Retr.

(2009)

MaH. et al.

Sorec: social recommendation using probabilistic matrix factorization

Proceedings of the 17th ACM Conference on Information and Knowledge Management

(2008)

MaH. et al.

Learning to recommend with social trust ensemble

Proceedings of the 32nd International ACM SIGIR Conference on Research and Development in Information Retrieval

(2009)

MaH. et al.

Recommender systems with social regularization

Proceedings of the Fourth ACM International Conference on Web Search and Data Mining

(2011)

J.I. Marden

Analyzing and Modeling Rank Data

(1996)

M. Marshall, Aggregate Knowledge raises 5M from Kleiner, on a roll, 2006. URL...

P. Massa et al.

Trust-aware recommender systems

Proceedings of the 2007 ACM Conference on Recommender Systems

(2007)

Cited by (49)

A generic Bayesian-based framework for enhancing top-N recommender algorithms
2021, Information Sciences
Personalized top-N recommender algorithms have been investigated widely in decades. The core task of different recommender algorithms is to estimate user-item preference scores and then to suggest, for each user, top-N items that have high preference scores. However, little attention was paid to the recommendation of items with low preference scores. In this work, we use the bayesian estimation theory to build the relationship between the estimated preference scores and the actual recommendation accuracy. We then propose a novel metric RNR (Recall-to-Noise Ratio) to recommend items with low estimated preference scores. An interesting counterintuitive phenomenon is found that user-item links with low preference scores may achieve higher accuracy than high preference score ones in the recommendation. Based on RNR, we design a generic framework that could recommend user-item links that have low preference scores without the loss of recommendation accuracy. The effectiveness of the proposed framework is illustrated by both theoretical analysis and empirical experiments.
Content-Aware Listwise Collaborative Filtering
2021, Neurocomputing
Recently, listwise collaborative filtering (CF) algorithms are attracting increasing interest due to their efficiency and prediction quality. Different from rating-oriented (pointwise) CF, they recommend a preference ranking of items to each user without estimating the absolute value of the ratings. In practice, there are extensive side information about users and items in many corpora that can improve recommendation performance and specially help CF methods to address the cold-start problem. However, a model has not been proposed that incorporates side information to listwise CF. Therefore, in this work, a Bayesian graphical model, called Content-Aware Listwise Collaborative Filtering (CALCF), is developed which incorporates text information, represented as a bag of words, to listwise CF using topic models. We propose a Gibbs sampler with closed-form samples using data augmentation techniques to infer the latent variables. CALCF has been validated by comparison with previous listwise and pointwise algorithms on implicit (click, view, purchase) and explicit (numeric ratings) feedback data in both cold-start and warm-start scenarios. The results demonstrate that in most cases CALCF achieves better Normalized Discounted Cumulative Gain (NDCG) and Recall at top M recommendation compared with the previous models.
Data set quality in Machine Learning: Consistency measure based on Group Decision Making
2021, Applied Soft Computing
Citation Excerpt :
This study focuses on the evaluation of training data consistency for ranking problems that are mainly solved using Learning to Rank algorithms (LTR) [1]. Among others, Learning to Rank algorithms are successfully applied to domains such as recommender systems [2–4], and Information Retrieval [5–7]. These algorithms rely on the assumption that similar labelers are agreed, not only about the highest-ranked alternative but on a ranked list of possible options.
Performance of Machine Learning models heavily depends on the quality of the training dataset. Among others, the quality of training data relies on the consistency of the labels assigned to similar items. Indeed, the labels should be coherently assigned (or collected) by avoiding inconsistencies for increasing the performance of the machine learning model. This study focuses on evaluating training data consistency for machine learning algorithms dealing with ranking problems, i.e., the Learning to Rank methods (LTR). This work defines a training data consistency measure based on the consensus value introduced in Group Decision Making. It investigates the statistical relationship between the proposed consistency measure and the performance of a deep neural network implementing an LTR method. This measure could drive data filtering at the training stage and guide model update decisions. Experimentation reveals a strong correlation between the proposed consistency measure and the performance of the model.
BSPR: Basket-sensitive personalized ranking for product recommendation
2020, Information Sciences
Product recommendation has played an important role in improving user experiences and obtaining more profits. To optimize recommendation models, pairwise learning has become a mainstream method for modeling user preferences from implicit feedback. Nevertheless, most existing pairwise methods optimize the relative order among products from only the user perspective. In fact, the purchase decision making of a given user depends on not only individual taste, but also the complementary relationships between the recommended products and the products in his/her historical basket. We argue that it is challenging to uncover meaningful user and product representations by only utilizing the user-side pairwise ranking. Towards this end, we propose a novel probabilistic pairwise method named BSPR, short for basket-sensitive personalized ranking, which solves both user- and product-side pairwise ranking problems in a unified manner. Specifically, BSPR discovers mutual correlations among users and products by exploiting co-pairwise ranking, alleviating the inherent flaw in existing pairwise methods. Considering that the negative sampler is one of the key components for pairwise learning, we devise a position-aware sampling strategy for the proposed method. To solve the optimization problem in BSPR, we further design an alternative optimization algorithm to efficiently learn the model parameters. Extensive experiments on multiple real-world datasets demonstrate significant improvements of our method over a series of state-of-the-art methods. Our implementation of BSPR is publicly available at:https://github.com/wubinzzu/BSPR.
Autoencoder-based personalized ranking framework unifying explicit and implicit feedback for accurate top-N recommendation
2019, Knowledge-Based Systems
Existing top- $N$ recommendation models can be classified according to the following two criteria: way of optimization and type of data. In terms of optimization, the models can either minimize the mean squared error (MSE) of rating predictions, which is so-called pointwise learning, or maximize the likelihood of pairwise preferences over more preferred and less preferred items (e.g., rated and unrated items), which is so-called pairwise learning. According to the data type, the models use either explicit feedback or implicit feedback. Most existing models use one of the optimization methods with either explicit or implicit feedback. However, we believe that pairwise learning and pointwise learning (resp. using explicit and implicit feedback) are complementary, thus employing both optimization methods and both forms of data together would bring a synergy effect in recommendation. Along this line, we propose a novel, unified recommendation framework based on deep neural networks, in which the pointwise and pairwise learning are employed together while using both the users’ explicit and implicit feedback. The experimental results on four real-life datasets confirm the effectiveness of our proposed framework over the state-of-the-art ones.
A personalized clustering-based and reliable trust-aware QoS prediction approach for cloud service recommendation in cloud manufacturing
2019, Knowledge-Based Systems
With the growing popularity of cloud manufacturing (CMfg), an increasing number of functionally equivalent cloud services are available on the Internet, which makes cloud service recommendation an important topic recently. Since the quality-of-service (QoS) varies largely in cloud environment, many Collaborative Filtering (CF) approaches are recently proposed to predict the uncertain QoS by utilizing the historical QoS records of similar users. In addition to the data sparsity problem, however, most existing approaches also ignore the task similarity among different users, while dissimilar tasks requested by similar users may lead to quite different QoS prediction results, and thus cannot achieve higher prediction accuracy and recommendation quality. To address this problem, we develop a novel clustering-based and trust-aware approach. Firstly, a set of similar users can be identified from the view of task similarity using clustering-based algorithm, where task similarity can be computed by incorporating both explicit textual information and rating information as well as implicit context information. Secondly, considering the fact that the QoS values may be contributed by unreliable users, we design a trust-aware CF approach by merging local and global trust values, to reconstruct trust network of the clustered users. Finally, we combine the clustering-based algorithm and trust-aware approach to make a more personalized QoS prediction and reliable cloud service recommendation for the active user in CMfg. Experimental results on two real-world data sets demonstrate the effectiveness of the proposed approach compared with other state-of-the-art approaches.

View all citing articles on Scopus

View full text

Improving top-K recommendation with truster and trustee relationship in user trust network

Abstract

Introduction

Section snippets

Related work

Problem description

Method

Complexity analysis

Experiments

Conclusion and future work

Acknowledgment

Knowl. Based Syst.

Proc. Comput. Sci.

Inf. Sci.

Context-aware recommender systems

Recommender Systems Handbook

The netflix prize

Proceedings of KDD Cup and Workshop

Empirical analysis of predictive algorithms for collaborative filtering

Proceedings of the Fourteenth Conference on Uncertainty in Artificial Intelligence

Transfer learning for collective link prediction in multiple heterogenous domains

Proceedings of the 27th International Conference on Machine Learning (ICML-10)

Learning to rank: from pairwise approach to listwise approach

Proceedings of the 24th International Conference on Machine Learning

Performance of recommender algorithms on top-n recommendation tasks

Proceedings of the Fourth ACM Conference on Recommender Systems

A novel bayesian similarity measure for recommender systems

IJCAI

Etaf: an extended trust antecedents framework for trust prediction

ASONAM

Trustsvd: collaborative filtering with both the explicit and implicit influence of user trust and of item ratings.

AAAI

Latent semantic models for collaborative filtering

ACM Trans. Inf. Syst.

Trustwalker: a random walk model for combining trust-based and item-based recommendation

Proceedings of the 15th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining

A matrix factorization technique with trust propagation for recommendation in social networks

Proceedings of the Fourth ACM Conference on Recommender Systems

Factorization meets the neighborhood: a multifaceted collaborative filtering model

Proceedings of the 14th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining

Collaborative filtering with temporal dynamics

Commun. ACM

Ratings meet reviews, a combined approach to recommend

Proceedings of the 8th ACM Conference on Recommender Systems

Learning to rank for information retrieval

Found. Trends Inf. Retr.

Sorec: social recommendation using probabilistic matrix factorization

Proceedings of the 17th ACM Conference on Information and Knowledge Management

Learning to recommend with social trust ensemble

Proceedings of the 32nd International ACM SIGIR Conference on Research and Development in Information Retrieval

Recommender systems with social regularization

Proceedings of the Fourth ACM International Conference on Web Search and Data Mining

Analyzing and Modeling Rank Data

Trust-aware recommender systems

Proceedings of the 2007 ACM Conference on Recommender Systems