Abstract
Context-aware recommendation algorithms focus on refining recommendations by considering additional information, available to the system. This topic has gained a lot of attention recently. Among others, several factorization methods were proposed to solve the problem, although most of them assume explicit feedback which strongly limits their real-world applicability. While these algorithms apply various loss functions and optimization strategies, the preference modeling under context is less explored due to the lack of tools allowing for easy experimentation with various models. As context dimensions are introduced beyond users and items, the space of possible preference models and the importance of proper modeling largely increases. In this paper we propose a general factorization framework (GFF), a single flexible algorithm that takes the preference model as an input and computes latent feature matrices for the input dimensions. GFF allows us to easily experiment with various linear models on any context-aware recommendation task, be it explicit or implicit feedback based. The scaling properties makes it usable under real life circumstances as well. We demonstrate the framework’s potential by exploring various preference models on a 4-dimensional context-aware problem with contexts that are available for almost any real life datasets. We show in our experiments—performed on five real life, implicit feedback datasets—that proper preference modelling significantly increases recommendation accuracy, and previously unused models outperform the traditional ones. Novel models in GFF also outperform state-of-the-art factorization algorithms. We also extend the method to be fully compliant to the Multidimensional Dataspace Model, one of the most extensive data models of context-enriched data. Extended GFF allows the seamless incorporation of information into the factorization framework beyond context, like item metadata, social networks, session information, etc. Preliminary experiments show great potential of this capability.
This is a preview of subscription content, log in via an institution to check access.
Similar content being viewed by others
Notes
User purchased an item or viewed an product page, etc. Interactions also called events or transactions.
We use the classic notion of implicit and explicit feedback here. In some cases explicit feedback can also be positive only, called also unary rating, typical is the voting scenario, such as Facebook likes, or Google’s \(+1\), Ricci et al. (2011a). However our focus is the easily collectable implicit feedback, that is unary data. While one can (and must) infer negative signs of preference from such data, e.g., by considering missing feedback or using additional information such as time spent on page, negative and positive preferences are not explicitly distinguished.
More precisely: the sum of elements in the elementwise product of corresponding vectors.
Meaning that a dimension can not directly interact with itself in the model.
Note that by setting \(w^0=0\) and \(w^1=1\) and using ratings in \(R\) we get the standard explicit setting in \(N_D\) dimensions.
We omit regularization for clearer presentation, but \(\ell _2\) regularization is used in the actual algorithm.
Omitted from the deduction for clearer presentation.
To avoid more complex notation, we assume that the columns of \(M^{(1)}\) are the first members in the products where they are present.
This value is 1.0 at TV1 and TV2. This is possibly due to preprocessing by the original authors that removed duplicate events.
We also measured recall@10 and recall@5 (not shown); the relation between different models are the same.
If we have no highlighted items in the recommendations (i.e. all recommended items are equal), then it makes sense to disregard the order of the recommended items. Whether this is true is determined by both the interface and the recommendation logic. For example, if we want to show more items or more diverse itemset to a user during a session while still giving relevant recommendations, we can randomize the top \(N\) recommendation and recommend the first \(K\) of this random order. This way we can overcome showing users the same \(K\) items multiple times and have a higher chance for clicking. The goal of the system is to recommend items that the user likes. The @20 comes from a very average setting of recommending 5 items (from a randomized pool of top 20 items) per page and the user having 4–6 page views in a session. Of course these numbers are highly varied in different applications, but we still think that this is a realistic proxy for a real recommender as it can get.
We rephrased here the feature matrix based introduction of the original paper.
A certain version of ALS, which optimizes for one parameter at a time.
References
Adomavicius G, Tuzhilin A (2008) Context-aware recommender systems. In: Recsys’08: ACM conference on recommender systems. pp 335–336
Adomavicius G, Sankaranarayanan R, Sen S, Tuzhilin A (2005) Incorporating contextual information in recommender systems using a multidimensional approach. ACM Trans Inf Syst 23(1):103–145
Bell R, Koren Y (2007) Scalable collaborative filtering with jointly derived neighborhood interpolation weights. In: ICDM’07: IEEE international conference on data mining. pp 43–52
Burke R (2007) Hybrid web recommender systems. In: The adaptive web. Springer, Berlin, pp 377–408
Celma O (2010) Music recommendation and discovery in the long tail. Springer, Berlin
Chen T, Zheng Z, Lu Q, Zhang W, Yu Y (2011) Feature-based matrix factorization. CoRR abs/1109.2271
Chen T, Zhang W, Lu Q, Chen K, Zheng Z, Yu Y (2012) SVDFeature: a toolkit for feature-based collaborative filtering. J Mach Learn Res 13:3619–3622
Cremonesi P, Turrin R (2009) Analysis of cold-start recommendations in IPTV systems. In: Recsys’09: ACM conference on recommender systems
Hidasi B (2014) Factorization models for context-aware recommendations. Infocommun J VI(4):27–34
Hidasi B, Tikk D (2012) Fast ALS-based tensor factorization for context-aware recommendation from implicit feedback. In: ECML-PKDD’12, Part II, no. 7524 in LNCS. Springer, Berlin, pp 67-82
Hidasi B, Tikk D (2013) Context-aware recommendations from implicit data via scalable tensor factorization. ArXiv e-prints
Hu Y, Koren Y, Volinsky C (2008) Collaborative filtering for implicit feedback datasets. In: ICDM’08: IEEE international conference on data mining. pp 263–272
Karatzoglou A, Amatriain X, Baltrunas L, Oliver N (2010) Multiverse recommendation: N-dimensional tensor factorization for context-aware collaborative filtering. In: Recsys’10: ACM conference on recommender systems. pp 79–86
Koren Y (2008) Factorization meets the neighborhood: a multifaceted collaborative filtering model. In: SIGKDD’08: ACM international conference on knowledge discovery and data mining. pp 426–434
Koren Y, Bell R (2011) Advances in collaborative filtering. In: Ricci F (ed) Recommender systems handbook. Springer, Berlin, pp 145–186
Little RJA, Rubin DB (1987) Statistical analysis with missing data. Willey, New York
Liu NN, Zhao BCM, Yang Q (2010) Adapting neighborhood and matrix factorization models for context aware recommendation. In: CAMRa’10: workshop on context-aware movie recommendation. pp 7–13
Liu Q, Chen T, Cai J, Yu D (2012) Enlister: Baidu’s recommender system for the biggest Chinese Q&A website. In: RecSys-12: proceedings of the 6th ACM conference on recommender systems. pp 285–288
Nguyen TV, Karatzoglou A, Baltrunas L (2014) Gaussian process factorization machines for context-aware recommendations. In: SIGIR-14: ACM SIGIR conference on research & development in information retrieval. pp 63–72
Paterek A (2007) Improving regularized singular value decomposition for collaborative filtering. In: Proceedings of the KDD cup and workshop. pp 5–8
Pilászy I, Tikk D (2009) Recommending new movies: even a few ratings are more valuable than metadata. In: Recsys’09: ACM conference on recommender systems. pp 93–100
Rendle S (2012) Factorization machines with libFM. ACM Trans Intell Syst Technol (TIST) 3(3):57
Rendle S (2013) Scaling factorization machines to relational data. In: PVLDB’13: 39th international conference on very large data bases. pp 337–348
Rendle S, Schmidt-Thieme L (2010) Pairwise interaction tensor factorization for personalized tag recommendation. In: WSDM’10: ACM international conference on web search and data mining. pp 81–90
Rendle S, Freudenthaler C, Gantner Z, Schmidt-Thieme L (2009) BPR: Bayesian personalized ranking from implicit feedback. In: UAI’09: 25th conference on uncertainty in artificial intelligence. pp 452–461
Ricci F, Rokach L, Shapira B (2011a) Introduction to recommender systems handbook. Recommender systems handbook. Springer, US, pp 1–35
Ricci F (ed) (2011b) Recommender systems handbook. Springer, Berlin
Salakhutdinov R, Mnih A (2008) Probabilistic matrix factorization. In: Advances in neural information processing systems, vol 20. MIT Press, Cambridge, MA
Shi Y, Karatzoglou A, Baltrunas L, Larson M, Hanjalic A, Oliver N (2012) TFMAP: Optimizing MAP for top-N context-aware recommendation. In: SIGIR-12: ACM SIGIR conference on research and development in information retrieval, ACM. pp 155–164
Takács G, Pilászy I, Németh B, Tikk D (2007) Major components of the Gravity recommendation system. SIGKDD Explor Newsl 9:80–83
Acknowledgments
The work leading to these results has received funding from the European Union’s Seventh Framework Programme (FP7/2007-2013) under CrowdRec Grant Agreement n\(^{\circ }\) 610594.
Author information
Authors and Affiliations
Corresponding author
Additional information
Responsible editors: Toon Calders, Floriana Esposito, Eyke Hüllermeier, and Rosa Meo.
Rights and permissions
About this article
Cite this article
Hidasi, B., Tikk, D. General factorization framework for context-aware recommendations. Data Min Knowl Disc 30, 342–371 (2016). https://doi.org/10.1007/s10618-015-0417-y
Received:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10618-015-0417-y