Embedding-Augmented Generalized Matrix Factorization for Recommendation With Implicit Feedback

Feng, Lei; Wei, Hongxin; Guo, Qingyu; Lin, Zhuoyi; An, Bo

doi:10.1109/mis.2020.3036136

“…A consequence of this line of thinking is that one should try to select neighbors that are not only strong predictors but also independent of each other, as also observed in [14]. Equation (11) and the independence assumption will be further discussed in Sect. 4.4.…”

Section: Ensemble Predictionmentioning

confidence: 93%

“…• In Sect. 4.2, using mild independence assumptions, we derive formula (11) for computing the OR of all past interactions that indicate a given new one and show how this formula can efficiently be computed. • In Sect.…”

Section: Our New Algorithm In the Probabilistic Interpretation Of Nea...mentioning

confidence: 99%

“…• In Sect. 4.4, we enhance the main formula (11) to decouple popularity from user preference. We introduce and handle separate taste and popularity components of a given user interaction.…”

Section: Our New Algorithm In the Probabilistic Interpretation Of Nea...mentioning

confidence: 99%

“…However, it makes sense to view Eq. (11) as the probability that the interaction happened because one of the modeled events caused it to happen. In essence, we need to view the value defined in Eq.…”

Section: Ensemble Predictionmentioning

confidence: 99%

“…Our contribution to non-sequential implicit feedback recommendation is the mathematical understanding and improvement of nearest neighbor, one of the oldest and most popular approaches for collaborative filtering [8]. While a multitude of more sophisticated methods has been developed over the years [9][10][11], very few can match the simplicity of the classic k-nearest neighbor. This simplicity also makes the recommendations and behavior of the method easier to interpret than many more recent counterparts, which makes it attractive in practice.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

A probabilistic perspective on nearest neighbor for implicit recommendation

Kelen

¹

,

Benczúr

²

2021

2021 International Conference on Data Mining Workshops (ICDMW)

1

0

View full text Add to dashboard Cite

Over the past years, the recommender systems community invented several novel approaches that reached better and better prediction accuracy. Sequential recommendation, such as music recommendation, has seen large improvements from neural network-based models such as recurrent neural networks or transformers. When no sequential information is available or not relevant, such as for book, movie, or product recommendation, however, the classic k-nearest neighbor algorithm appears to remain competitive, even when compared to much more sophisticated methods. In this paper, we attempt to explain the inner workings of the nearest neighbor using probabilistic tools, treating similarity as conditional probability and presenting a novel model for explaining and removing popularity bias. First, we provide a probabilistic formulation of similarity and the classic prediction formula. Second, by modeling user behavior as a combination of personal preference and global influence, we are able to explain the presence of popularity bias in the predictions. Finally, we utilize Bayesian inference to construct a theoretically grounded variant of the widely used inverse frequency scaling, which we use to mitigate the effect of popularity bias in the predictions. By replacing the formerly ad hoc choices of nearest neighbor with probabilistically founded counterparts, we are able to improve prediction accuracy over a variety of data sets and gain an increased understanding of the theory behind the method.

show abstract

A novel recommendation system comprising WNMF with graph-based static and temporal similarity estimators

Gupta

¹

,

Shrinath²

2022

Int J Data Sci Anal

View full text Add to dashboard Cite

A probabilistic perspective on nearest neighbor for implicit recommendation

Kelen

¹

,

Benczúr

²

2022

Int J Data Sci Anal

2

0

View full text Add to dashboard Cite

Over the past years, the recommender systems community invented several novel approaches that reached better and better prediction accuracy. Sequential recommendation, such as music recommendation, has seen large improvements from neural network-based models such as recurrent neural networks or transformers. When no sequential information is available or not relevant, such as for book, movie, or product recommendation, however, the classic k-nearest neighbor algorithm appears to remain competitive, even when compared to much more sophisticated methods. In this paper, we attempt to explain the inner workings of the nearest neighbor using probabilistic tools, treating similarity as conditional probability and presenting a novel model for explaining and removing popularity bias. First, we provide a probabilistic formulation of similarity and the classic prediction formula. Second, by modeling user behavior as a combination of personal preference and global influence, we are able to explain the presence of popularity bias in the predictions. Finally, we utilize Bayesian inference to construct a theoretically grounded variant of the widely used inverse frequency scaling, which we use to mitigate the effect of popularity bias in the predictions. By replacing the formerly ad hoc choices of nearest neighbor with probabilistically founded counterparts, we are able to improve prediction accuracy over a variety of data sets and gain an increased understanding of the theory behind the method.

show abstract

Embedding-Augmented Generalized Matrix Factorization for Recommendation With Implicit Feedback

Cited by 5 publications

References 16 publications

A probabilistic perspective on nearest neighbor for implicit recommendation

A probabilistic perspective on nearest neighbor for implicit recommendation

A novel recommendation system comprising WNMF with graph-based static and temporal similarity estimators

A probabilistic perspective on nearest neighbor for implicit recommendation

Contact Info

Product

Resources

About