Improving pairwise learning for item recommendation from implicit feedback

By Steffen Rendle, Christoph Freudenthaler. 2014

416 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Information Systems, Management Science and Operations Research

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

1. Xiangnan He
2. Xiang Wang
3. Tat-seng Chua
4. Julian McAuley
5. Liqiang Nie
6. Meng Wang
7. Fuli Feng
8. Yongfeng Zhang

In Information Systems

1. Xiangnan He
2. Xiang Wang
3. Tat-seng Chua
4. Julian McAuley
5. Meng Wang
6. Fuli Feng
7. Yongfeng Zhang
8. Liqiang Nie

In Management Science and Operations Research

1. Xiangnan He
2. Julian McAuley
3. Tat-seng Chua
4. Xiang Wang
5. Wang-Cheng Kang
6. Ruining He
7. Jiaxi Tang
8. Ke Wang

Citations in the corpus

• Joint Topic-Semantic-aware Social Recommendation for Online Voting (2017, methodology): “Collaborative filtering methods [22, 25, 28, 31] use either explicit feedback (e.”
• Leveraging Social Connections to Improve Personalized Ranking for Collaborative Filtering (2014, methodology): “In [17], the authors propose two strategies to sample negative feedback and also design an efficient implementation to improve the convergence of the learning algorithm.”
• LightGCN: Simplifying and Powering Graph Convolution Network for Recommendation (2020, background): “We are aware of other advanced negative sampling strategies which might improve the LightGCN training, such as the hard negative sampling [31] and adversarial sampling [9].”
• Personalizing Session-based Recommendations with Hierarchical Recurrent Neural Networks (2017, background): “Both properties are known to be beneficial for pairwise learning with implicit user feedback [13].”
• Modeling User Exposure in Recommendation (2015, background): “Bayesian personalized ranking (BPR) [25, 24] instead treats the problem as a one of ranking consumed user-item pairs above unobserved pairs.”
• Online Machine Learning in Big Data Streams (2018, background): “nts with positive feedback. To fit a model, one requires negative feedback for training as well. Usually, such elements are selected by sampling from those that the user has not interacted with before [195]. We can also introduce confidence values for ratings and consider lower confidence for the artifi”
• Ranking Distillation: Learning Compact Ranking Models With High Performance for Recommender System (2018, methodology): “s weight wa should be inversely proportional to the rank: wa r ∝r −1 and r ∈[1,K], (6) where r is the rank range from 1 to K. As pointed out above, this scheme pre-determines the weight. Rendle et al [29] proposed an empirical weight for sampling a single position from a top-K ranking, following a g”
• Fast Matrix Factorization for Online Recommendation with Implicit Feedback (2016, methodology): “Our proposed popularity-aware weighting strategy has the same intuition with Rendle’s popularity-based oversampling [24] for learning BPR, which basically samples popular items as negative feedback with a higher probability.”
• Bias and Debias in Recommender System: A Survey and Future Directions (2020, background): “An example is the adaptive negative sampler [40], [41], [42], [43], which aims to over-sample the “difficult” instances in order to increase learning speed, even though the resultant loss function will differ significantly from the original.”
• A Generic Coordinate Descent Framework for Learning from Implicit Feedback (2016, background): “It requires more complex, non-uniform, sampling strategies for dealing with this problem [12, 6].”
• Translation-based Recommendation (2017, background): “, [18, 19, 21]) make a weaker assumption that users simply prefer observed feedback over unobserved feedback and optimize the pairwise rankings of (positive, non-positive) pairs.”
• Self-supervised Graph Learning for Recommendation (2020, background): “This indicates that hard negative nodes offer much larger gradients to guide the optimization, thus making node representations more discriminative and accelerating the training process [33].”
• Social Recommendation with Strong and Weak Ties (2016, background): “Although collaborative filtering approaches can be adapted [24], pairwise ranking methods have gained more traction lately [30, 39, 41, 42, 48].”
• Contrastive Learning for Cold-Start Recommendation (2021, background): “LU I is benefited by its multiple negative samples, which obviously increases the probability of hard negative detection, thus offers meaningful and larger gradients to the optimization [28].”