Deep Context-Aware Recommender System Utilizing Sequential Latent Context

Cited by: 1|Views6
Weibo:
We presented deep context-aware recommendation models that utilize explicit and latent context representations and learn nonlinear interaction function between users, items, and contexts

Abstract:

Context-aware recommender systems (CARSs) apply sensing and analysis of user context in order to provide personalized services. Adding context to a recommendation model is challenging, since the addition of context may increases both the dimensionality and sparsity of the model. Recent research has shown that modeling contextual informa...More

Code:

Data:

0
Full Text
Bibtex
Weibo
Introduction
  • Context-aware computing was first defined by [3] as "software that adapts according to its location of use, the collection of nearby people and objects, as well as changes to those objects over time." As such, context-aware services should react to the users’ given contexts and adapt their services .
  • In [23] the authors represent environmental features as low-dimensional unsupervised latent contexts that were extracted by an auto-encoder (AE).
  • While this representation takes the current context of the user into account and can greatly improve recommendation accuracy, it does not model evolution of context over time.
  • Recent studies [4, 6, 16, 22] have shown that incorporating sequential information that models the users’ behavior over time improves the quality of recommendations
Highlights
  • Context-aware computing was first defined by [3] as "software that adapts according to its location of use, the collection of nearby people and objects, as well as changes to those objects over time." As such, context-aware services should react to the users’ given contexts and adapt their services
  • It is important to mention that the number of available datasets with rich context data is scarce, and we focused on these two datasets that contain rich contextual information and up to 4% missing context values
  • It can be seen that in each of the datasets, our neural context-aware models outperformed the various baseline models on all of the measures. These results indicate that adding context information to the neural model improves recommendation accuracy
  • We can see that sequential latent context-aware model (SLCM) obtained the best results among the other suggested neural context models (i.e., explicit neural context-aware model (ENCM) and latent neural context-aware model (LNCM)). This may be due to the fact that SLCM handled long-term context patterns which were modeled by context sequences, while ENCM and LNCM utilized only current context of the user
  • We presented deep context-aware recommendation models that utilize explicit and latent context representations and learn nonlinear interaction function between users, items, and contexts
  • We showed that our models significantly outperform popular baselines used for rating prediction task
Methods
  • The authors aim to design a novel context-aware deep recommendation model that automatically learns the relations between users, items, and contextual information.
  • The authors present novel NCF extensions models for the CARS domain that improve traditional MF.
  • These models utilize contextual information in an explicit and GMF Layer.
  • Element- wise product Score ෞ.
  • MLP Layer 1.
  • MF User Vector MF Item Vector MLP User Vector MLP Item Vector
Results
  • It can be seen that in each of the datasets, the neural context-aware models outperformed the various baseline models on all of the measures.
  • These results indicate that adding context information to the neural model improves recommendation accuracy.
  • The authors can see that SLCM obtained the best results among the other suggested neural context models (i.e., ENCM and LNCM).
  • This may be due to the fact that SLCM handled long-term context patterns which were modeled by context sequences, while ENCM and LNCM utilized only current context of the user
Conclusion
  • The authors presented deep context-aware recommendation models that utilize explicit and latent context representations and learn nonlinear interaction function between users, items, and contexts.
  • The authors conducted several experiments on two context-aware datasets to compare the approaches to state of the art recommendation models with respect to the hit@k, RMSE, and MAE measures.
  • In terms of ranking quality, SLCM improved baseline models up to 16.3% in the hit@k measure.
  • With regard to prediction accuracy, SLCM improved baseline models up to 9.47% for the RMSE measure and 8.98% for the
Summary
  • Introduction:

    Context-aware computing was first defined by [3] as "software that adapts according to its location of use, the collection of nearby people and objects, as well as changes to those objects over time." As such, context-aware services should react to the users’ given contexts and adapt their services .
  • In [23] the authors represent environmental features as low-dimensional unsupervised latent contexts that were extracted by an auto-encoder (AE).
  • While this representation takes the current context of the user into account and can greatly improve recommendation accuracy, it does not model evolution of context over time.
  • Recent studies [4, 6, 16, 22] have shown that incorporating sequential information that models the users’ behavior over time improves the quality of recommendations
  • Objectives:

    The authors aim to design a novel context-aware deep recommendation model that automatically learns the relations between users, items, and contextual information
  • Methods:

    The authors aim to design a novel context-aware deep recommendation model that automatically learns the relations between users, items, and contextual information.
  • The authors present novel NCF extensions models for the CARS domain that improve traditional MF.
  • These models utilize contextual information in an explicit and GMF Layer.
  • Element- wise product Score ෞ.
  • MLP Layer 1.
  • MF User Vector MF Item Vector MLP User Vector MLP Item Vector
  • Results:

    It can be seen that in each of the datasets, the neural context-aware models outperformed the various baseline models on all of the measures.
  • These results indicate that adding context information to the neural model improves recommendation accuracy.
  • The authors can see that SLCM obtained the best results among the other suggested neural context models (i.e., ENCM and LNCM).
  • This may be due to the fact that SLCM handled long-term context patterns which were modeled by context sequences, while ENCM and LNCM utilized only current context of the user
  • Conclusion:

    The authors presented deep context-aware recommendation models that utilize explicit and latent context representations and learn nonlinear interaction function between users, items, and contexts.
  • The authors conducted several experiments on two context-aware datasets to compare the approaches to state of the art recommendation models with respect to the hit@k, RMSE, and MAE measures.
  • In terms of ranking quality, SLCM improved baseline models up to 16.3% in the hit@k measure.
  • With regard to prediction accuracy, SLCM improved baseline models up to 9.47% for the RMSE measure and 8.98% for the
Tables
  • Table1: An Example of Sorted Interactions for Sequence Generation with the CARS Dataset
  • Table2: Description of Context-Aware Datasets
  • Table3: Prediction Results
Download tables as Excel
Related work
  • The CARS domain deals with modeling and predicting users’ tastes and preferences by incorporating available contextual information into the recommendation process. Prior research has shown that adding explicit or latent context to the recommendation process can improve recommendation accuracy [2, 23]. While the latent representation can greatly improve recommendation accuracy, it only considers a single context instance of the user.

    Despite the effectiveness of MF [13] as the primary latent factor model of introducing users’ preferences into recommendation systems, it has some limitations: 1) the multiplication of latent features linearly may not be sufficient to capture the complex structure of user interaction data, and 2) the dimensionality of the latent factor space must be explicitly defined. In order to address these limitations, [10] suggested a deep neural network architecture for learning the interaction function between users and items. We design novel context-aware models which include the context dimension in order to the learn nonlinear interactions between latent representations of user, item and context. We suggest several explicit and latent context representations for this task. The latent representations are obtained by utilizing encoder-decoder neural networks (AE and LSTM encoder-decoder networks) which reduce the high-dimensional context space and reveal complex correlations within the data.
Funding
  • It is important to mention that the number of available datasets with rich context data is scarce, and therefore we focused on these two datasets that contain rich contextual information and up to 4% missing context values
  • In the CARS dataset SLCM achieved the best results and performed 5.6% better than the best baseline model (NeuMF) in terms of the RMSE and 15.2% better than the best baseline model (NeuMF) in terms of the MAE
  • In the Yelp datasets, a similar phenomenon was observed, as SLCM performed the best in terms of both the RMSE (1.239) and MAE (1.011) with up to 12.3% improvement over the best baseline (NeuMF)
  • We showed that our models significantly outperform popular baselines used for rating prediction task
  • In terms of ranking quality, SLCM improved baseline models up to 16.3% in the hit@k measure
  • With regard to prediction accuracy, SLCM improved baseline models up to 9.47% for the RMSE measure and 8.98% for the
Study subjects and analysis
users: 150770
Since the original data was highly sparse, we retained users and items with at least 10 interactions. This results in a subset of data that contains 150,770 users, 60,852 items, and 588,253 interactions. We use the following contextual factors: year, month, day of the week, and city

Reference
  • Gediminas Adomavicius and Alexander Tuzhilin. 2015. Context-aware recommender systems. In Recommender systems handbook. Springer, 191–226.
    Google ScholarFindings
  • Linas Baltrunas, Bernd Ludwig, and Francesco Ricci. 2011. Matrix factorization techniques for context aware recommendation. In Proceedings of the fifth ACM conference on Recommender systems. ACM, 301–304.
    Google ScholarLocate open access versionFindings
  • Daniel Billsus, Clifford A Brunk, Craig Evans, Brian Gladish, and Michael Pazzani. 2002. Adaptive interfaces for ubiquitous web access. Commun. ACM 45, 5 (2002), 34–38.
    Google ScholarLocate open access versionFindings
  • Xu Chen, Hongteng Xu, Yongfeng Zhang, Jiaxi Tang, Yixin Cao, Zheng Qin, and Hongyuan Zha. 2018. Sequential recommendation with user memory networks. In Proceedings of the Eleventh ACM International Conference on Web Search and Data Mining. ACM, 108–116.
    Google ScholarLocate open access versionFindings
  • Kyunghyun Cho, Bart Van Merriënboer, Caglar Gulcehre, Dzmitry Bahdanau, Fethi Bougares, Holger Schwenk, and Yoshua Bengio. 2014. Learning phrase representations using RNN encoder-decoder for statistical machine translation. arXiv preprint arXiv:1406.1078 (2014).
    Findings
  • Robin Devooght and Hugues Bersini. 2017. Long and Short-Term Recommendations with Recurrent Neural Networks. In Proceedings of the 25th Conference on User Modeling, Adaptation and Personalization. ACM, 13–21.
    Google ScholarLocate open access versionFindings
  • PN Druzhkov and VD Kustikova. 2016. A survey of deep learning methods and software tools for image classification and object detection. Pattern Recognition and Image Analysis 26, 1 (2016), 9–15.
    Google ScholarLocate open access versionFindings
  • Alex Graves. 2013. Generating sequences with recurrent neural networks. arXiv preprint arXiv:1308.0850 (2013).
    Findings
  • Elad Hazan, Tomer Koren, and Nati Srebro. 2011. Beating sgd: Learning svms in sublinear time. In Advances in Neural Information Processing Systems. 1233–1241.
    Google ScholarLocate open access versionFindings
  • Xiangnan He, Lizi Liao, Hanwang Zhang, Liqiang Nie, Xia Hu, and Tat-Seng Chua. 2017. Neural collaborative filtering. In Proceedings of the 26th International Conference on World Wide Web. International World Wide Web Conferences Steering Committee, 173–182.
    Google ScholarLocate open access versionFindings
  • Sepp Hochreiter and Jürgen Schmidhuber. 1997. Long short-term memory. Neural computation 9, 8 (1997), 1735–1780.
    Google ScholarLocate open access versionFindings
  • Nal Kalchbrenner and Phil Blunsom. 2013. Recurrent continuous translation models. In Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing. 1700–1709.
    Google ScholarLocate open access versionFindings
  • Yehuda Koren. 2008. Factorization meets the neighborhood: a multifaceted collaborative filtering model. In Proceedings of the 14th ACM SIGKDD international conference on Knowledge discovery and data mining. ACM, 426–434.
    Google ScholarLocate open access versionFindings
  • Yucheng Low, Joseph E Gonzalez, Aapo Kyrola, Danny Bickson, Carlos E Guestrin, and Joseph Hellerstein. 20Graphlab: A new framework for parallel machine learning. arXiv preprint arXiv:1408.2041 (2014).
    Findings
  • Charith Perera, Arkady Zaslavsky, Peter Christen, and Dimitrios Georgakopoulos. 2014. Context aware computing for the internet of things: A survey. IEEE communications surveys & tutorials 16, 1 (2014), 414–454.
    Google ScholarLocate open access versionFindings
  • Massimo Quadrana, Paolo Cremonesi, and Dietmar Jannach. 2018. Sequenceaware recommender systems. arXiv preprint arXiv:1802.08452 (2018).
    Findings
  • Steffen Rendle, Christoph Freudenthaler, Zeno Gantner, and Lars Schmidt-Thieme. 2009. BPR: Bayesian personalized ranking from implicit feedback. In Proceedings of the twenty-fifth conference on uncertainty in artificial intelligence. AUAI Press, 452–461.
    Google ScholarLocate open access versionFindings
  • Alan Said and Alejandro Bellogín. 2014. Comparative Recommender System Evaluation: Benchmarking Recommendation Frameworks. In Proceedings of the 8th ACM Conference on Recommender Systems (RecSys ’14). ACM, New York, NY, USA, 129–136. https://doi.org/10.1145/2645710.2645746
    Locate open access versionFindings
  • Elena Smirnova and Flavian Vasile. 2017. Contextual Sequence Modeling for Recommendation with Recurrent Neural Networks. arXiv preprint arXiv:1706.07684 (2017).
    Findings
  • Harold Soh, Scott Sanner, Madeleine White, and Greg Jamieson. 2017. Deep sequential recommendation for personalized adaptive user interfaces. In Proceedings of the 22nd International Conference on Intelligent User Interfaces. ACM, 589–593.
    Google ScholarLocate open access versionFindings
  • Ilya Sutskever, Oriol Vinyals, and Quoc V Le. 2014. Sequence to sequence learning with neural networks. In Advances in neural information processing systems. 3104– 3112.
    Google ScholarFindings
  • Yong Kiam Tan, Xinxing Xu, and Yong Liu. 2016. Improved recurrent neural networks for session-based recommendations. In Proceedings of the 1st Workshop on Deep Learning for Recommender Systems. ACM, 17–22.
    Google ScholarLocate open access versionFindings
  • Moshe Unger, Ariel Bar, Bracha Shapira, and Lior Rokach. 2016. Towards latent context-aware recommendation systems. Knowledge-Based Systems 104 (2016), 165–178.
    Google ScholarLocate open access versionFindings
  • Moshe Unger, Bracha Shapira, Lior Rokach, and Amit Livne. 2018. Inferring contextual preferences using deep encoder-decoder learners. New Review of Hypermedia and Multimedia 24, 3 (2018), 262–290.
    Google ScholarLocate open access versionFindings
  • Yao Wu, Christopher DuBois, Alice X Zheng, and Martin Ester. 2016. Collaborative denoising auto-encoders for top-n recommender systems. In Proceedings of the Ninth ACM International Conference on Web Search and Data Mining. ACM, 153–162.
    Google ScholarLocate open access versionFindings
  • Yong Zheng. 2015. A revisit to the identification of contexts in recommender systems. In Proceedings of the 20th International Conference on Intelligent User Interfaces Companion. ACM, 133–136.
    Google ScholarLocate open access versionFindings
Your rating :
0

 

Tags
Comments