hPSD: A Hybrid PU-Learning-Based Spammer Detection Model for Product Reviews
The DOI matches its Crossref record; 39 of the 39 references it cites were found in a registry or an index.
Abstract
Spammers, who manipulate online reviews to promote or suppress products, are flooding in online commerce. To combat this trend, there has been a great deal of research focused on detecting review spammers, most of which design diversified features and thus develop various classifiers. The widespread growth of crowdsourcing platforms has created large-scale deceptive review writers who behave more like normal users, that the way they can more easily evade detection by the classifiers that are purely based on fixed characteristics. In this paper, we propose a hybrid semisupervised learning model titled hybrid PU-learning-based spammer detection (hPSD) for spammer detection to leverage both the users' characteristics and the user-product relations. Specifically, the hPSD model can iteratively detect multitype spammers by injecting different positive samples, and allows the construction of classifiers in a semisupervised hybrid learning framework. Comprehensive experiments on movie dataset with shilling injection confirm the superior performance of hPSD over existing baseline methods. The hPSD is then utilized to detect the hidden spammers from real-life Amazon data. A set of spammers and their underlying employers (e.g., book publishers) are successfully discovered and validated. These demonstrate that hPSD meets the real-world application scenarios and can thus effectively detect the potentially deceptive review writers.
Overview
Integrity signals
Registry
Cites
Statistics
Verification
Claims the paper makes
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
In the citing paper
“FL allows ML model training while preserving data privacy [35,36,37]”
From A Novel Blockchain-Based Deepfake Detection Method Using Federated and Deep Learning Models (2024) · cites this paper as [37]
Citations
Cites
Multi-Interval Discretization of Continuous-Valued Attributes for Classification Learning
Fayyad, U. M., & Irani, K. B. (1993) · International Joint Conference on Artificial Intelligence · Cited by 2,870
Zhu, F., & Zhang, X. (. (2010) · Journal of Marketing · doi:10.1509/jm.74.2.133 · Cited by 2,363
Forman, C., Ghose, A., & Wiesenfeld, B. (2008) · Information Systems Research · doi:10.1287/isre.1080.0193 · Cited by 1,695 · cited 1×
“Online product ratings and reviews strongly influence the purchase decisions of a vast number of customers, and thus are strongly correlated with sales [1]–[3].”
BackgroundJindal, N., & Liu, B. (2008) · Proceedings of the international conference on Web search and web data mining - WSDM '08 · doi:10.1145/1341531.1341560 · Cited by 1,542 · cited 1×
“On one hand, the spamming behaviors can be categorized into multiple types, for example, Jindal and Liu [13] summarized three types of them, but it is difficult to unify the detection of multitype spammers with different characteristics into one model.”
Web data mining: exploring hyperlinks, contents, and usage data
(2012) · Choice Reviews Online · doi:10.5860/choice.49-2718 · Cited by 1,179
T-drive: driving directions based on taxi trajectories
Yuan, J., Zheng, Y., Zhang, C., Xie, W., Xie, X., Sun, G., & Huang, Y. (2010) · Proceedings of the 18th SIGSPATIAL International Conference on Advances in Geographic Information Systems · doi:10.1145/1869790.1869807 · Cited by 1,105 · cited 2×
“In [30] and [31], finding the “best” cut points is not a trivial task.”
BackgroundMethodFinding Deceptive Opinion Spam by Any Stretch of the Imagination
Ott, M., Choi, Y., Cardie, C., & Hancock, J. T. (2011) · arXiv · doi:10.48550/arxiv.1107.4557 · Cited by 1,008 · cited 1×
“To detect such spam review, an abundant of linguistic features (e.g., ratio of objective/subjective words computed by WordNet [22]) have been designed based on review content [22], [23].”
MethodSpotting fake reviewer groups in consumer reviews
Mukherjee, A., Liu, B., & Glance, N. (2012) · Proceedings of the 21st international conference on World Wide Web · doi:10.1145/2187836.2187863 · Cited by 773 · cited 6×
“Among them, five features are reformed based on opinion spam behavior indicators [9], [10]: 1) the difference of first and last rating dates; 2) the number of days when posting ratings; 3) the maximal number of ratings posted in one day; 4) the average number of ratings per day; and 5) the entropy…”
BackgroundInfluentialDetecting product review spammers using rating behaviors
Lim, E.-P., Nguyen, V.-A., Jindal, N., Liu, B., & Lauw, H. W. (2010) · Proceedings of the 19th ACM international conference on Information and knowledge management · doi:10.1145/1871437.1871557 · Cited by 715 · cited 4×
“These studies mainly focus on two subareas: 1) the spamming behavior features [5], [8]–[10], [16]–[19] and 2) the spam classification approaches [7], [9], [10], [12], [20], [21].”
BackgroundInfluentialBuilding text classifiers using positive and unlabeled examples
Liu, B., Dai, Y., Li, X., Lee, W. S., & Yu, P. S. (2004) · Third IEEE International Conference on Data Mining · doi:10.1109/icdm.2003.1250918 · Cited by 667 · cited 1×
“In [34] and [35], a similar strategy is used in the context of multiclass text classification, that is, the positive examples correctly classified are maintained after finishing each round of classification with respect to a P set.”
MethodShilling recommender systems for fun and profit
Lam, S. K., & Riedl, J. (2004) · Proceedings of the 13th international conference on World Wide Web · doi:10.1145/988672.988726 · Cited by 635 · cited 1×
“1) Anomaly Injection: We take shilling attackers [11] as malicious users to be detected.”
BackgroundUncovering social spammers: social honeypots + machine learning
Lee, K., Caverlee, J., & Webb, S. (2010) · Proceedings of the 33rd international ACM SIGIR conference on Research and development in information retrieval · doi:10.1145/1835449.1835522 · Cited by 619 · cited 1×
“Consequently, an experimental dataset is acquired with the basic information listed in Table V. 1) Attacker Types: In our experiments, we consider two types of attackers, that is, duplicate spammers and promoters, that are ubiquitous on real e-commerce platforms [38].”
MethodPartially Supervised Classification of Text Documents
Liu, B., Lee, W. S., Yu, P. S., & Li, X. (2002) · Cited by 534
Jindal, N., & Liu, B. (2007) · Proceedings of the 16th international conference on World Wide Web · doi:10.1145/1242572.1242759 · Cited by 449 · cited 3×
“focus on two subareas: 1) the spamming behavior features [5], [8]–[10], [16]–[19] and 2) the spam classification approaches [7], [9], [10], [12], [20], [21].”
BackgroundSpotting opinion spammers using behavioral footprints
Mukherjee, A., Kumar, A., Liu, B., Wang, J., Hsu, M., Castellanos, M., & Ghosh, R. (2013) · Proceedings of the 19th ACM SIGKDD international conference on Knowledge discovery and data mining · doi:10.1145/2487575.2487580 · Cited by 426 · cited 6×
“Most behavioral features used for spammer detection are numerical [5], [9], [10], [19].”
BackgroundInfluentialReview Graph Based Online Store Review Spammer Detection
Wang, G., Xie, S., Liu, B., & Yu, P. S. (2011) · 2011 IEEE 11th International Conference on Data Mining · doi:10.1109/icdm.2011.124 · Cited by 376 · cited 2×
“A few studies propose to exploit the relationship network to enhance the produced classifications [6], [7], [12], [20].”
BackgroundMethodLearning to identify review spam
Li, F., Huang, M., Yang, Y., & Zhu, X. (2011) · doi:10.5591/978-1-57735-516-8/ijcai11-414 · Cited by 328
Ketkar, N. (2017) · Deep Learning with Python · doi:10.1007/978-1-4842-2766-4_8 · Cited by 319
Preventing shilling attacks in online recommender systems
Chirita, P.-A., Nejdl, W., & Zamfir, C. (2005) · Proceedings of the 7th annual ACM international workshop on Web information and data management · doi:10.1145/1097047.1097061 · Cited by 309
Estimating the prevalence of deception in online review communities
Ott, M., Cardie, C., & Hancock, J. (2012) · Proceedings of the 21st international conference on World Wide Web · doi:10.1145/2187836.2187864 · Cited by 282 · cited 1×
“Surprisingly, it has been shown that up to 6% of the reviews on sites like Yelp and TripAdvisor may be falsified [4].”
BackgroundRecent Development in Big Data Analytics for Business Operations and Risk Management
Choi, T.-M., Chan, H. K., & Yue, X. (2017) · IEEE Transactions on Cybernetics · doi:10.1109/tcyb.2015.2507599 · Cited by 278 · cited 1×
“Online product ratings and reviews strongly influence the purchase decisions of a vast number of customers, and thus are strongly correlated with sales [1]–[3].”
BackgroundClassification features for attack detection in collaborative recommender systems
Burke, R., Mobasher, B., Williams, C., & Bhaumik, R. (2006) · Proceedings of the 12th ACM SIGKDD international conference on Knowledge discovery and data mining · doi:10.1145/1150402.1150465 · Cited by 247 · cited 3×
“For example, in shilling attack detection [17]–[19], a number of features are defined based on the length variance, deviation, and similarity among rating vectors.”
BackgroundMethodAnalyzing and Detecting Review Spam
Jindal, N., & Liu, B. (2007) · Seventh IEEE International Conference on Data Mining (ICDM 2007) · doi:10.1109/icdm.2007.68 · Cited by 224 · cited 2×
“These studies mainly focus on two subareas: 1) the spamming behavior features [5], [8]–[10], [16]–[19] and 2) the spam classification approaches [7], [9], [10], [12], [20], [21].”
BackgroundText classification without negative examples revisit
Fung, G. P. C., Yu, J. X., Lu, H., & Yu, P. S. (2006) · IEEE Transactions on Knowledge and Data Engineering · doi:10.1109/tkde.2006.16 · Cited by 221 · cited 2×
“First, it adopts the positive and unlabeled dataset learning (PU-learning) [14], [15] scheme to detect multitype spammers, where the positive (P) and unlabeled (U) set denote injected spammers and unknown users, respectively.”
BackgroundMethodUnsupervised strategies for shilling detection and robust collaborative filtering
Mehta, B., & Nejdl, W. (2009) · User Modeling and User-Adapted Interaction · doi:10.1007/s11257-008-9050-4 · Cited by 170 · cited 2×
“For instance, Mehta and Nejdl [28] utilized the principal component analysis (PCA) on user–product rating matrix to detect shilling attackers, and Mukherjee et al.”
BackgroundMethodLies and propaganda: detecting spam users in collaborative filtering
Mehta, B., Hofmann, T., & Fankhauser, P. (2007) · ACM International Conference on Intelligent User Interfaces (IUI) · doi:10.1145/1216295.1216307 · Cited by 159 · cited 3×
“, probability) to rank suspected spammers [9], [10], [26], [27].”
BackgroundMethodModel-based collaborative filtering as a defense against profile injection attacks
Mobasher, B., Burke, R., & Sandvig, J. J. (2006) · National Conference on Artificial Intelligence · Cited by 153
HySAD: a semi-supervised hybrid shilling attack detector for trustworthy product recommendation
Wu, Z., Wu, J., Cao, J., & Tao, D. (2012) · Proceedings of the 18th ACM SIGKDD international conference on Knowledge discovery and data mining · doi:10.1145/2339530.2339684 · Cited by 119 · cited 2×
“For instance, Wu et al. [21] have shown the effectiveness of semisupervised learning on shilling attack detection, that is, it is more suitable for the real-world e-commerce sites that are lacking labeled spammers.”
BackgroundSelf-citationPositive Unlabeled Learning for Data Stream Classification
Li, X.-L., Yu, P. S., Liu, B., & Ng, S.-K. (2009) · Proceedings of the 2009 SIAM International Conference on Data Mining · doi:10.1137/1.9781611972795.23 · Cited by 115 · cited 1×
“The RN set consists of a small set of instances extracted from U that are remarkably different from the instances in the positive set P [32].”
ResultCollective Spammer Detection in Evolving Multi-Relational Social Networks
Fakhraei, S., Foulds, J., Shashanka, M., & Getoor, L. (2015) · Proceedings of the 21th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining · doi:10.1145/2783258.2788606 · Cited by 109 · cited 7×
“Along this line, some recent studies [7], [12] proposed utilizing these relationships to define new topological features.”
BackgroundMethodInfluentialShilling Attack Detection—A New Approach for a Trustworthy Recommender System
Lee, J., & Zhu, D. (2011) · INFORMS journal on computing · doi:10.1287/ijoc.1100.0440 · Cited by 103 · cited 3×
“In [27] and [36], the shilling attackers are summarized into several types according to different methods for selecting filler items and assigning ratings to them.”
BackgroundMethodUncovering Crowdsourced Manipulation of Online Reviews
Fayazi, A., Lee, K., Caverlee, J., & Squicciarini, A. (2015) · Proceedings of the 38th International ACM SIGIR Conference on Research and Development in Information Retrieval · doi:10.1145/2766462.2767742 · Cited by 90 · cited 6×
“Hence, product review spammer detection [5]–[7] has attracted an increasing amount of attention in the recent years.”
BackgroundInfluentialLearning Rate Adaptation in Stochastic Gradient Descent
Plagianakos, V. P., Magoulas, G. D., & Vrahatis, M. N. (2001) · Nonconvex optimization and its applications · doi:10.1007/978-1-4613-0279-7_27 · Cited by 69
Positive and Unlabeled Multi-Graph Learning
Wu, J. X., Pan, S., Xingquan Zhu, A., Chengqi Zhang, B., & Wu, X. (2016) · IEEE Transactions on Cybernetics · doi:10.1109/tcyb.2016.2527239 · Cited by 65 · cited 2×
“First, it adopts the positive and unlabeled dataset learning (PU-learning) [14], [15] scheme to detect multitype spam-mers, where the positive ( P ) and unlabeled ( U ) set denote injected spammers and unknown users, respectively.”
BackgroundSelf-citationGLAD: group anomaly detection in social media analysis
Yu, R., He, X., & Liu, Y. (2014) · ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD) · doi:10.1145/2623330.2623719 · Cited by 49 · cited 1×
“Researchers have also begun analyzing how to detect groups of review spammers [9], [24], [25].”
BackgroundDiscovering shilling groups in a real e-commerce platform
Wang, Y., Wu, Z., Bu, Z., Cao, J., & Yang, D. (2016) · Online Information Review · doi:10.1108/oir-03-2015-0073 · Cited by 34 · cited 1×
“Researchers have also begun analyzing how to detect groups of review spammers [9], [24], [25].”
BackgroundSelf-citationProfile Injection Attack Detection for Securing Collaborative Recommender Systems 1
Williams, C., & Mobasher, B. (2006) · Cited by 31
Liu, H., Zhang, Y., Lin, H., Wu, J., Wu, Z., & Zhang, X. (2013) · Proceedings · doi:10.1109/icdm.2013.166 · Cited by 23 · cited 1×
“Second, a majority of approaches leverage indicative features of spam for building supervised classification models [7], [12], [19], [20], [29].”
BackgroundSelf-citationDetecting Marionette Microblog Users for Improved Information Credibility
Wu, X., Fan, W., Gao, J., Feng, Z., & Yu, Y. H. (2015) · Journal of Computer Science and Technology · doi:10.1007/s11390-015-1584-4 · Cited by 7 · cited 5×
“These studies mainly focus on two subareas: 1) the spamming behavior features [5], [8]–[10], [16]–[19] and 2) the spam classification approaches [7], [9], [10], [12], [20], [21].”
BackgroundMethodInfluential
Research works printed but not linked: 2
- bausch, 2003, Amazon Hacks
- 10.1509/jmkg.74.2.133 · doi:10.1509/jmkg.74.2.133 · Paper page
Cited by
Fake online reviews: Literature review, synthesis, and directions for future research
Wu, Y., Ngai, E. W. T., Wu, P., & Wu, C. (2020) · Decision Support Systems · doi:10.1016/j.dss.2020.113280 · Cited by 359
A Deep Learning-based Fast Fake News Detection Model for Cyber-Physical Social Services
Zhang, Q., Guo, Z., Zhu, Y., Vijayakumar, P., Castiglione, A., & Gupta, B. B. (2023) · Pattern Recognition Letters · doi:10.1016/j.patrec.2023.02.026 · Cited by 183
Robust Spammer Detection Using Collaborative Neural Network in Internet-of-Things Applications
Guo, Z., Shen, Y., Bashir, A. K., Imran, M., Kumar, N., Zhang, D., & Yu, K. (2021) · IEEE Internet of Things Journal · doi:10.1109/jiot.2020.3003802 · Cited by 177
A Novel Blockchain-Based Deepfake Detection Method Using Federated and Deep Learning Models
Heidari, A., Navimipour, N. J., Dag, H., Talebi, S., & Unal, M. (2024) · Cognitive Computation · doi:10.1007/s12559-024-10255-7 · Cited by 164
“FL allows ML model training while preserving data privacy [35–37], and blockchain ensures data trustworthiness by authenticating source data and enhancing deepfake detection.”
Fuzzy Detection System for Rumors Through Explainable Adaptive Learning
Guo, Z., Yu, K., Jolfaei, A., Bashir, A. K., Almagrabi, A. O., & Kumar, N. (2021) · IEEE Transactions on Fuzzy Systems · doi:10.1109/tfuzz.2021.3052109 · Cited by 144
“Predictably, as the social applications constantly gain their popularity, situations where labels are missing tend to become more universal [30].”
Background
Metadata
- Authors
- Wu, Z., Cao, J., Wang, Y., Wang, Y., Zhang, L., & Wu, J.
- Affiliations
- Nanjing University of Finance and Economics (CN)
- University of Minnesota (US)
- Beihang University (CN)
- Venue
- IEEE Transactions on Cybernetics
- Published
- 2020-04
- Volume
- 50
- Issue
- 4
- Pages
- 1595-1606
- ISSN
- 2168-2267 (ISSN-L)
- 2168-2275
- OpenAlex
- W2899676218
- Cited by
- 166
- References
- 39
- Citations by year
- 2026: 1 · 2025: 7 · 2024: 28 · 2023: 52 · 2022: 38 · 2021: 20 · 2020: 9
- FWCI
- 16.13
- Type
- journal article
- Language
- en
- Open access
- closed
- Licence
- Locations
- IEEE Transactions on Cybernetics (publishedVersion)
- PubMed (publishedVersion)
- Funders
- National Basic Research Program of China (2016YFB1000901)
- National Natural Science Foundation of China (71571093, 91646204, 71701089, 71801123, 71725002, 71531001, U1636210, 71471009)
- National Center for International Joint Research on E-Business Information Processing (2013B01035)
- Fundamental Research Funds for the Central Universities
- Topics
- Physical Sciences › Computer Science › Information Systems › Spam and Phishing Detection
- Physical Sciences › Computer Science › Artificial Intelligence › Hate Speech and Cyberbullying Detection
- Social Sciences › Sociology and Political Science › Misinformation and Its Impacts
- Keywords
- positive-unlabeled learning, semi-supervised learning, hybrid learning, user characteristics, Amazon reviews, crowdsourcing platforms
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
Data sources
Bibliographic data: Semantic Scholar, OpenAlex, Crossref.
Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).
Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.