Explanation framework for industrial recommendation systems based on the generative adversarial network with embedding constraints

Abstract

The explainability of recommendation systems refers to the ability to explain the logic that guides the system’s decision to endorse or exclude an item. In industrial-grade recommendation systems, the high complexity of features, the presence of embedding layers, the existence of adversarial samples and the requirements for explanation accuracy and efficiency pose significant challenges to current explanation methods. This paper proposes a novel framework AdvLIME (Adversarial Local Interpretable Model-agnostic Explanation) that leverages Generative Adversarial Networks (GANs) with Embedding Constraints to enhance explainability. This method utilizes adversarial samples as references to explain recommendation decisions, generating these samples in accordance with realistic distributions and ensuring they meet the structural constraints of the embedding module. AdvLIME requires no modifications to the existing model architecture and needs only a single training session for global explanation, making it ideal for industrial applications. This work contributes two significant advancements. First, it develops a model-independent global explanation method via adversarial generation. Second, it introduces a model discrimination method to guarantee that the generated samples adhere to the embedding constraints. We evaluate the AdvLIME framework on the Behavior Sequence Transformer (BST) model using the MovieLens 20 M dataset. The experimental results show that AdvLIME outperforms traditional methods such as LIME and DLIME, reducing the approximation error of real samples by 50% and demonstrating improved stability and accuracy.

The original online version of this article was revised: the statement of Data availability and Competing interests have been added.

A correction to this article is available online at https://doi.org/10.1007/s43684-025-00110-3.

Keywords

Explainable recommendation system / Feature embedding / Generative adversarial networks / Deep learning

References

[1] SemwalP., HandaA.. Cyber-attack detection in cyber-physical systems using supervised machine learning. Handbook of Big Data Analytics and Forensics, 2022131140.

[2] EngstromL., TranB., TsiprasD., SchmidtL., MadryA.. Exploring the landscape of spatial robustness. PMLR, 201918021811

[3] ChenL., CruzA., RamseyS., DicksonC.J., DucaJ.S., HornakV., KoesD.R., KurtzmanT.. Hidden bias in the dud-e dataset leads to misleading performance of deep learning in structure-based virtual screening. PLoS ONE, 2019, 14. e0220113https://api.semanticscholar.org/CorpusID:201117942

[4] ChallenR., DennyJ.C., PittM.A., GompelsL., EdwardsT., Tsaneva-AtanasovaK.. Artificial intelligence, bias and clinical safety. BMJ Qual. Saf., 2019, 28: 231-237. https://api.semanticscholar.org/CorpusID:58554131

[5] SinzF.H., PitkowX., ReimerJ., BethgeM., ToliasA.S.. Engineering a less artificial intelligence. Neuron, 2019, 103: 967-979. https://api.semanticscholar.org/CorpusID:202749691

[6] StahlB.C., WrightD.. Ethics and privacy in ai and big data: implementing responsible research and innovation. IEEE Secur. Priv., 2018, 16: 26-33. https://api.semanticscholar.org/CorpusID:49416376

[7] KeskinboraK.H.. Medical ethics considerations on artificial intelligence. J. Clin. Neurosci., 2019, 64: 277-282. https://api.semanticscholar.org/CorpusID:76667734

[8] HengstlerM., EnkelE., DuelliS.. Applied artificial intelligence and trust-the case of autonomous vehicles and medical assistance devices. Technol. Forecast. Soc. Change, 2016, 105: 105-120. https://api.semanticscholar.org/CorpusID:61695662

[9] LuiA., LambG.W.. Artificial intelligence and augmented intelligence collaboration: regaining trust and confidence in the financial sector. Inf. Commun. Technol. Law, 2018, 27: 267-283. https://api.semanticscholar.org/CorpusID:65123095

[10] WeldD.S., BansalG.. The challenge of crafting intelligible intelligence. Commun. ACM, 2018, 62: 70-79. https://api.semanticscholar.org/CorpusID:49563054

[11] TintarevN., MasthoffJ.. Designing and evaluating explanations for recommender systems. Recommender Systems Handbook, 2011https://api.semanticscholar.org/CorpusID:28780910

[12] SharmaA., CosleyD.. Do social explanations work?: studying and modeling the effects of social explanations in recommender systems. Proceedings of the 22nd International Conference on World Wide Web, 2013https://api.semanticscholar.org/CorpusID:6172004

[13] VigJ., SenS., RiedlJ.. Tagsplanations: explaining recommendations using tags. Proceedings of the 14th International Conference on Intelligent User Interfaces, 2009https://api.semanticscholar.org/CorpusID:5814453

[14] GoodmanB., FlaxmanS.. European Union regulations on algorithmic decision-making and a “right to explanation”. AI Mag., 2016, 38: 50-57https://api.semanticscholar.org/CorpusID:7373959

[15] C. Szegedy, W. Zaremba, I. Sutskever, J. Bruna, D. Erhan, I.J. Goodfellow, R. Fergus, Intriguing properties of neural networks. CoRR (2013). https://api.semanticscholar.org/CorpusID:604334. arXiv:1312.6199

[16] I.J. Goodfellow, J. Shlens, C. Szegedy, Explaining and harnessing adversarial examples. CoRR (2014). https://api.semanticscholar.org/CorpusID:6706414. arXiv:1412.6572

[17] A. Kurakin, I.J. Goodfellow, S. Bengio, Adversarial examples in the physical world. arXiv (2016). https://api.semanticscholar.org/CorpusID:1257772. arXiv:1607.02533

[18] Moosavi-DezfooliS.M., FawziA., FrossardP.. Deepfool: a simple and accurate method to fool deep neural networks. 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 201525742582https://api.semanticscholar.org/CorpusID:12387176

[19] P.Y. Chen, Y. Sharma, H. Zhang, J. Yi, C.J. Hsieh, Ead: elastic-net attacks to deep neural networks via adversarial examples. arXiv (2017). https://api.semanticscholar.org/CorpusID:19157252. arXiv:1709.04114

[20] ChenP.Y., ZhangH., SharmaY., YiJ., HsiehC.J.. Zoo: zeroth order optimization based black-box attacks to deep neural networks without training substitute models. Proceedings of the 10th ACM Workshop on Artificial Intelligence and Security, 2017https://api.semanticscholar.org/CorpusID:2179389

[21] C. Guo, J.R. Gardner, Y. You, A.G. Wilson, K.Q. Weinberger, Simple black-box adversarial attacks. arXiv (2019). https://api.semanticscholar.org/CorpusID:86541092. arXiv:1905.07121

[22] W. Brendel, J. Rauber, M. Bethge, Decision-based adversarial attacks: reliable attacks against black-box machine learning models. arXiv (2017). https://api.semanticscholar.org/CorpusID:2410333. arXiv:1712.04248

[23] C. Xiao, B. Li, J.Y. Zhu, W. He, M. Liu, D.X. Song, Generating adversarial examples with adversarial networks. arXiv (2018). https://api.semanticscholar.org/CorpusID:21949142. arXiv:1801.02610

[24] ManglaP., JandialS., VarshneyS., BalasubramanianV.N.. Advgan++: harnessing latent layers for adversary generation. 2019 IEEE/CVF International Conference on Computer Vision Workshop (ICCVW), 201920452048https://api.semanticscholar.org/CorpusID:199405373

[25] H.N. Mhaskar, T.A. Poggio, Deep vs. shallow networks: an approximation theory perspective. arXiv (2016). https://api.semanticscholar.org/CorpusID:263797010. arXiv:1608.03287

[26] T. Poggio, Q. Liao, Theory II: landscape of the empirical risk in deep learning. Ph.D. thesis, Center for Brains, Minds and Machines (CBMM), arXiv (2017) arXiv:1703.09833

[27] C. Zhang, Q. Liao, A. Rakhlin, K. Sridharan, B. Miranda, N. Golowich, T. Poggio, Musings on deep learning: properties of SGD. Tech. rep., Center for Brains, Minds and Machines (CBMM) (2017) http://hdl.handle.net/1721.1/107841

[28] ChenQ., ZhaoH., LiW., HuangP., OuW.. Behavior sequence transformer for e-commerce recommendation in alibaba. Proceedings of the 1st International Workshop on Deep Learning Practice for High-Dimensional Sparse Data, 2019https://api.semanticscholar.org/CorpusID:155099871

[29] HarperF.M., KonstanJ.A.. The movielens datasets: history and context. ACM Trans. Interact. Intell. Syst., 2016, 5: 19:1-19:19. https://api.semanticscholar.org/CorpusID:16619709

[30] RibeiroM.T., SinghS., GuestrinC.. “Why should I trust you?”: explaining the predictions of any classifier. Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 2016https://api.semanticscholar.org/CorpusID:13029170

[31] M.R. Zafar, N.M. Khan, Dlime: a deterministic local interpretable model-agnostic explanations approach for computer-aided diagnosis systems. arXiv (2019). https://api.semanticscholar.org/CorpusID:195584114. arXiv:1906.10263

[32] GuidottiR., MonrealeA., TuriniF., PedreschiD., GiannottiF.. A survey of methods for explaining black box models. ACM Comput. Surv., 2018, 51: 1-42. https://api.semanticscholar.org/CorpusID:3342225

[33] M.D. Zeiler, R. Fergus, Visualizing and understanding convolutional networks. arXiv (2013). https://api.semanticscholar.org/CorpusID:3960646. arXiv:1311.2901

[34] KohP.W., LiangP.. Understanding black-box predictions via influence functions. International Conference on Machine Learning, 2017https://api.semanticscholar.org/CorpusID:13193974

[35] L.M. Zintgraf, T. Cohen, T. Adel, M. Welling, Visualizing deep neural network decisions: prediction difference analysis. arXiv (2017). https://api.semanticscholar.org/CorpusID:2103669. arXiv:1702.04595

[36] V. Petsiuk, A. Das, K. Saenko, Rise: randomized input sampling for explanation of black-box models. arXiv (2018). https://api.semanticscholar.org/CorpusID:49324724. arXiv:1806.07421

[37] K. Simonyan, A. Vedaldi, A. Zisserman, Deep inside convolutional networks: visualising image classification models and saliency maps. CoRR (2013). https://api.semanticscholar.org/CorpusID:1450294. arXiv:1312.6034

[38] J.T. Springenberg, A. Dosovitskiy, T. Brox, M.A. Riedmiller, Striving for simplicity: the all convolutional net. CoRR (2014). https://api.semanticscholar.org/CorpusID:12998557. arXiv:1412.6806

[39] M. Sundararajan, A. Taly, Q. Yan, Gradients of counterfactuals. arXiv (2016). https://api.semanticscholar.org/CorpusID:11880723. arXiv:1611.02639

[40] D. Smilkov, N. Thorat, B. Kim, F.B. Vi´egas, M. Wattenberg, Smoothgrad: removing noise by adding noise. arXiv (2017). https://api.semanticscholar.org/CorpusID:11695878. arXiv:1706.03825

[41] ZhouB., KhoslaA., LapedrizaA., OlivaA., TorralbaA.. Learning deep features for discriminative localization. 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 201529212929https://api.semanticscholar.org/CorpusID:6789015

[42] SelvarajuR.R., DasA., VedantamR., CogswellM., ParikhD., BatraD.. Gradcam: visual explanations from deep networks via gradient-based localization. Int. J. Comput. Vis., 2016, 128: 336-359. https://api.semanticscholar.org/CorpusID:15019293

[43] ChattopadhyayA., SarkarA., HowladerP., BalasubramanianV.N.. Gradcam++: generalized gradient-based visual explanations for deep convolutional networks. 2018 IEEE Winter Conference on Applications of Computer Vision (WACV), 2017839847https://api.semanticscholar.org/CorpusID:13678776

[44] LundbergS.M., LeeS.I.. A unified approach to interpreting model predictions. Neural Information Processing Systems, 2017https://api.semanticscholar.org/CorpusID:21889700

[45] BucilaC., CaruanaR., Niculescu-MizilA.. Model compression. Knowledge Discovery and Data Mining, 2006https://api.semanticscholar.org/CorpusID:11253972

[46] CravenM.W., ShavlikJ.W.. Extracting tree-structured representations of trained networks. Neural Information Processing Systems, 1995https://api.semanticscholar.org/CorpusID:265038754

[47] WuM., HughesM.C., ParbhooS., ZazziM., RothV., Doshi-VelezF.. Beyond sparsity: tree regularization of deep models for explainability. AAAI Conference on Artificial Intelligence, 2017https://api.semanticscholar.org/CorpusID:7614825

[48] H. Lakkaraju, E. Kamar, R. Caruana, J. Leskovec, Interpretable & explorable approximations of black box models. arXiv (2017). https://api.semanticscholar.org/CorpusID:19598815. arXiv:1707.01154

[49] TanS., CaruanaR., HookerG., KochP., GordoA.. Learning global additive explanations for neural nets using model distillation. Stat, 2018, 10503https://doi.org/10.48550/arXiv.1801.08640

[50] Z. Hu, X. Ma, Z. Liu, E.H. Hovy, E.P. Xing, Harnessing deep neural networks with logic rules. arXiv (2016). https://api.semanticscholar.org/CorpusID:7663461. arXiv:1603.06318

[51] GuoW., MuD., XuJ., SuP., WangG., XingX.. Lemna: explaining deep learning based security applications. Proceedings of the 2018 ACM SIGSAC Conference on Computer and Communications Security, 2018https://api.semanticscholar.org/CorpusID:52027892

[52] S. Shi, X. Zhang, W. Fan, A modified perturbed sampling method for local interpretable model-agnostic explanation. arXiv (2020). https://api.semanticscholar.org/CorpusID:211146282. arXiv:2002.07434

[53] YuX., RenX., GuQ., SunY., HanJ.. Collaborative filtering with entity similarity regularization in heterogeneous information networks. IJCAI HINA, 2013, 27: 1-6

[54] GaoL., YangH., WuJ., ZhouC., LuW., HuY.. Recommendation with multisource heterogeneous information. IJCAI International Joint Conference on Artificial Intelligence, 201833783384

[55] WangX., WangD., XuC., HeX., CaoY., ChuaT.S.. Explainable reasoning over knowledge graphs for recommendation. AAAI Conference on Artificial Intelligence, 2018https://api.semanticscholar.org/CorpusID:267895341

[56] XianY., FuZ., MuthukrishnanS., de MeloG., ZhangY.. Reinforcement knowledge graph reasoning for explainable recommendation. Proceedings of the 42nd International ACM SIGIR Conference on Research and Development in Information Retrieval, 2019https://api.semanticscholar.org/CorpusID:186206810

[57] BordesA., UsunierN., GarciadA., WestonJ., YakhnenkoO.. Translating embeddings for modeling multi-relational data. Neural Information Processing Systems, 2013https://api.semanticscholar.org/CorpusID:14941970

[58] LinY., LiuZ., SunM., LiuY., ZhuX.. Learning entity and relation embeddings for knowledge graph completion. AAAI Conference on Artificial Intelligence, 2015https://api.semanticscholar.org/CorpusID:2949428

[59] Q. Ai, V. Azizi, X. Chen, Y. Zhang, Learning over knowledge-base embeddings for recommendation. arXiv (2018). https://api.semanticscholar.org/CorpusID:4371098. arXiv:1805.03352

[60] GaoJ., WangX., WangY., XieX.. Explainable recommendation through attentive multi-view learning. AAAI Conference on Artificial Intelligence, 2019https://api.semanticscholar.org/CorpusID:69392767

[61] M. Hardt, B. Recht, Y. Singer, Train faster, generalize better: stability of stochastic gradient descent. arXiv (2015). https://api.semanticscholar.org/CorpusID:49015. arXiv:1509.01240

[62] GonenA., Shalev-ShwartzS.. Fast rates for empirical risk minimization of strict saddle problems. Annual Conference Computational Learning Theory, 2017https://api.semanticscholar.org/CorpusID:8426367

[63] CarliniN., WagnerD.A.. Towards evaluating the robustness of neural networks. 2017 IEEE Symposium on Security and Privacy (SP), 20163957https://api.semanticscholar.org/CorpusID:2893830

[64] Y. Liu, X. Chen, C. Liu, D.X. Song, Delving into transferable adversarial examples and black-box attacks arXiv (2016). https://api.semanticscholar.org/CorpusID:17707860. arXiv:1611.02770

Funding

the National Natural Science Foundation of China(62403360)

the Aeronautical Science Foundation of China(2023Z066038001)

the National Natural Science Foundation of China Basic Science Research Center Program(62088101)

Shanghai Municipal Science and Technology Major Project(2021SHZDZX0100)

iF open-funds from Xinghuo Eco and China Institute of Communications

RIGHTS & PERMISSIONS

The Author(s)