A Fast Parallel Stochastic Gradient Method for Matrix Factorization in Shared Memory Systems

Chin, Wei-Sheng; Zhuang, Yun; Juan, Yu-Chin; Lin, Chih-Jen

doi:10.1145/2668133

Cited by 84 publications

(62 citation statements)

References 13 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We consider observed entries with ratings 4 as positive. 4 For some problems, training and test sets are available, but the test sets are too small. Therefore, other than delicious, we merge training and test sets of every problem first, and then do a 9-to-1 split to obtain training/test sets for our experiments.…”

Section: Resultsmentioning

confidence: 99%

See 1 more Smart Citation

Selection of Negative Samples for One-class Matrix Factorization

Yu¹,

Bilenko²

2017

Proceedings of the 2017 SIAM International Conference on Data Mining

Self Cite

View full text Add to dashboard Cite

Many recommender systems have only implicit user feedback. The two possible ratings are positive and negative, but only part of positive entries are observed. One-class matrix factorization (MF) is a popular approach for such scenarios by treating some missing entries as negative. Two major ways to select negative entries are by sub-sampling a set with similar size to that of observed positive entries or by including all missing entries as negative. They are referred to as "subsampled" and "full" approaches in this work, respectively. Currently detailed comparisons between these two selection schemes on large-scale data are still lacking. One important reason is that the "full" approach leads to a hard optimization problem after treating all missing entries as negative. In this paper, we successfully develop e cient optimization techniques to solve this challenging problem so that the "full" approach becomes practically viable. We then compare in detail the two approaches "subsampled" and "full" for selecting negative entries. Results show that the "full" approach of including much more missing entries as negative yields better results.

show abstract

Section: Resultsmentioning

confidence: 99%

“…However, such settings do not guarantee the complexity reduction like what we achieved for ALS and CD. Further, existing methods to parallelize SG for MF such as [4,8] may become not applicable because A is split into blocks. In contrast, ALS and CD under our new settings for one-class MF can be easily parallelized.…”

Section: Stochastic Gradient (Sg)mentioning

confidence: 99%

Selection of Negative Samples for One-class Matrix Factorization

Yu¹,

Bilenko²

2017

Proceedings of the 2017 SIAM International Conference on Data Mining

Self Cite

View full text Add to dashboard Cite

show abstract

“…Various ranges of the entries in R and bi are specified in our experiments (see Table 1). The number of steps T is set to 10 and the rating logs are then generated according to equations (7) and (11).…”

Section: Experiments On the Synthetic Datasetmentioning

confidence: 99%

“…In Table 1, we report the RMSE for both the original MF (implemented by the LIBMF library [7,8]) and our method for various parameter settings. The parameters that we choose for MF are the learning rate α= 0.01, the regulator parameter λ = 0.02, and the number of factors D = 30.…”

Section: Experiments On the Synthetic Datasetmentioning

confidence: 99%

Temporal Matrix Factorization for Tracking Concept Drift in Individual User Preferences

Liao

Chang

et al. 2018

IEEE Trans. Comput. Soc. Syst.

View full text Add to dashboard Cite

The matrix factorization (MF) technique has been widely adopted for solving the rating prediction problem in recommender systems. The MF technique utilizes the latent factor model to obtain static user preferences (user latent vectors) and item characteristics (item latent vectors) based on historical rating data. However, in the real world user preferences are not static but full of dynamics. Though there are several previous works that addressed this time varying issue of user preferences, it seems (to the best of our knowledge) that none of them is specifically designed for tracking concept drift in individual user preferences. Motivated by this, we develop a Temporal Matrix Factorization approach (TMF) for tracking concept drift in each individual user latent vector. There are two key innovative steps in our approach: (i) we develop a modified stochastic gradient descent method to learn an individual user latent vector at each time step, and (ii) by the Lasso regression we learn a linear model for the transition of the individual user latent vectors. We test our method on a synthetic dataset and several real datasets. In comparison with the original MF, our experimental results show that our temporal method is able to achieve lower root mean square errors (RMSE) for both the synthetic and real datasets. One interesting finding is that the performance gain in RMSE is mostly from those users who indeed have concept drift in their user latent vectors at the time of prediction. In particular, for the synthetic dataset and the Ciao dataset, there are quite a few users with that property and the performance gains for these two datasets are roughly 20% and 5%, respectively.

show abstract

“…Matrix Factorization and its application to personalized recommendation demonstrated the effectiveness of directly modelling all the dimensions simultaneously in a unified framework. These among other works presupposes that, tensor decomposition models performed well in terms of prediction efficiency and effectiveness compared to the various matrix factorization algorithms, in particular application to massive data processing [24]- [26]. However, the numerous literature concerning the subject.…”

Section: Introductionmentioning

confidence: 99%