Session 4: Collaborative Filtering

Date: Tuesday October 15, 15:15 PM – 16:25 PM (GMT+2)
Room: Petruzzelli Theater
Session Chair: Rodrygo Santos

• RES 🕓5The Role of Unknown Interactions in Implicit Matrix Factorization — A Probabilistic View
  by Joey De Pauw (University of Antwerp) and Bart Goethals (University of Antwerp)

  Matrix factorization is a well-known and effective methodology for top-k list recommendation. It became widely known during the Netflix challenge in 2006, and since then, many adapted and improved versions have been published. A particularly interesting matrix factorization algorithm called iALS (for implicit Alternating Least Squares) adapts the method for implicit feedback, i.e. a setting where only a very small amount of positive labels are available along with a majority of unknown labels. Compared to the classical task of rating prediction, learning from implicit feedback is applicable to many more domains, as the data is more abundant and requires less effort to elicit from users. However, the sparsity, imbalance, and implicit nature of the signal also pose unique challenges to retrieving the most relevant items to recommend.

  We revisit the role of unknown interactions in implicit matrix factorization. Traditionally, all unknowns are interpreted as negative samples and their importance in the training objective is then down-weighted to balance them out with the known, positive interactions. Interestingly, by adapting a probabilistic view of matrix factorization, we can retain the unknown nature of these interactions by modelling them as either positive or negative. With this new formulation that better fits the underlying data, we gain improved performance on the downstream recommendation task without any computational overhead compared to the popular iALS method.

  This paper outlines the key insights needed to adapt iALS to use logistic regression. Furthermore, a logistic version of the popular full-rank EASE model is introduced in a similar fasion. An extensive experimental evaluation on several real-world datasets demonstrates the effectiveness of our approach. Additionally, a discrepancy between the need for weighting between factorization and autoencoder models is discovered, leading towards a better understanding of these methods.

  Full text in ACM Digital Library

• RES 🕓15Adaptive Fusion of Multi-View for Graph Contrastive Recommendation
  by Mengduo Yang (Zhejiang University), Yi Yuan (Zhejiang University), Jie Zhou (Zhejiang University), Meng Xi (Zhejiang University), Xiaohua Pan (Zhejiang University), Ying Li (Zhejiang University), Yangyang Wu (Zhejiang University), Jinshan Zhang (Zhejiang University) and Jianwei Yin (Zhejiang University)

  Recommendation is a key mechanism for modern users to access items of their interests from massive entities and information. Recently, graph contrastive learning (GCL) has demonstrated satisfactory results on recommendation, due to its ability to enhance representation by integrating graph neural networks (GNNs) with contrastive learning. However, those methods often generate contrastive views by performing random perturbation on edges or embeddings, which is likely to bring noise in representation learning. Besides, in all these methods, the degree of user preference on items is omitted during the representation learning process, which may cause incomplete user/item modeling. To address these limitations, we propose the Adaptive Fusion of Multi-View Graph Contrastive Recommendation (AMGCR) model. Specifically, to generate the informative and less noisy views for better contrastive learning, we design four view generators to learn the edge weights focusing on weight adjustment, feature transformation, neighbor aggregation, and attention mechanism, respectively. Then, we employ an adaptive multi-view fusion module to combine different views from both the view-shared and the view-specific levels. Moreover, to make the model capable of capturing preference information during the learning process, we further adopt a preference refinement strategy on the fused contrastive view. Experimental results on three real-world datasets demonstrate that AMGCR consistently outperforms the state-of-the-art methods, with average improvements of over 10% in terms of Recall and NDCG. Our code is available on https://github.com/Du-danger/AMGCR.

  Full text in ACM Digital Library

• RES 🕓15Low Rank Field-Weighted Factorization Machines for Low Latency Item Recommendation
  by Alex Shtoff (Yahoo Research), Michael Viderman (Yahoo Research), Naama Haramaty-Krasne, Oren Somekh (Yahoo Research), Ariel Raviv (Meta) and Tularam Ban (Yahoo Research)

  Factorization machine (FM) variants are widely used in recommendation systems that operate under strict throughput and latency requirements, such as online advertising systems. FMs have two prominent strengths. First, is their ability to model pairwise feature interactions while being resilient to data sparsity by learning factorized representations. Second, their computational graphs facilitate fast inference and training. Moreover, when items are ranked as a part of a query for each incoming user, these graphs facilitate computing the portion stemming from the user and context fields only once per query. Thus, the computational cost for each ranked item is proportional only to the number of fields that vary among the ranked items. Consequently, in terms of inference cost, the number of user or context fields is practically unlimited.

  More advanced variants of FMs, such as field-aware and field-weighted FMs, provide better accuracy by learning a representation of field-wise interactions, but require computing all pairwise interaction terms explicitly. In particular, the computational cost during inference is proportional to the square of the number of fields, including user, context, and item. When the number of fields is large, this is prohibitive in systems with strict latency constraints, and imposes a limit on the number of user and context fields for a given computational budget. To mitigate this caveat, heuristic pruning of low intensity field interactions is commonly used to accelerate inference.

  In this work we propose an alternative to the pruning heuristic in field-weighted FMs using a diagonal plus symmetric low-rank decomposition. Our technique reduces the computational cost of inference, by allowing it to be proportional to the number of item fields only. Using a set of experiments on real-world datasets, we show that aggressive rank reduction outperforms similarly aggressive pruning in both accuracy and item recommendation speed. Beyond computational complexity analysis, we corroborate our claim of faster inference experimentally, both via a synthetic test, and by having deployed our solution to a major online advertising system, where we observed significant ranking latency improvements. We have made the code to reproduce the results on public datasets and synthetic tests available at https://github.com/michaelviderman/pytorch-fm.

  Full text in ACM Digital Library

• RES 🕓15Unlocking the Hidden Treasures: Enhancing Recommendations with Unlabeled Data
  by Yuhan Zhao (Harbin Engineering University), Rui Chen (Harbin Engineering University), Qilong Han (Harbin Engineering University), Hongtao Song (Harbin Engineering University) and Li Chen (Hong Kong Baptist University)

  Collaborative filtering (CF) stands as a cornerstone in recommender systems, yet effectively leveraging the massive unlabeled data presents a significant challenge. Current research focuses on addressing the challenge of unlabeled data by extracting a subset that closely approximates negative samples. Regrettably, the remaining data are overlooked, failing to fully integrate this valuable information into the construction of user preferences. To address this gap, we introduce a novel positive-neutral-negative (PNN) learning paradigm. PNN introduces a neutral class, encompassing intricate items that are challenging to categorize directly as positive or negative samples. By training a model based on this triple-wise partial ranking, PNN offers a promising solution to learning complex user preferences. Through theoretical analysis, we connect PNN to one-way partial AUC (OPAUC) to validate its efficacy. Implementing the PNN paradigm is, however, technically challenging because: (1) it is difficult to classify unlabeled data into neutral or negative in the absence of supervised signals; (2) there does not exist any loss function that can handle set-level triple-wise ranking relationships. To address these challenges, we propose a semi-supervised learning method coupled with a user-aware attention model for knowledge acquisition and classification refinement. Additionally, a novel loss function with a two-step centroid ranking approach enables handling set-level rankings. Extensive experiments on four real-world datasets demonstrate that, when combined with PNN, a wide range of representative CF models can consistently and significantly boost their performance. Even with a simple matrix factorization, PNN can achieve comparable performance to sophisticated graph neutral networks. Our code is publicly available at https://github.com/Asa9aoTK/PNN-RecBole.

  Full text in ACM Digital Library

• REPR 🕓10One-class Matrix Factorization: Point-Wise Regression-Based or Pair-Wise Ranking-Based?
  by Sheng-Wei Chen (National Taiwan University) and Chih-Jen Lin (National Taiwan University)

  One-class matrix factorization (MF) is an important technique for recommender systems with implicit feedback. In one widely used setting, a regression function is fit in a point-wise manner on observed and some unobserved (user, item) entries. Recently, in AAAI 2019, Chen et al. [2] proposed a pair-wise ranking-based approach for observed (user, item) entries to be compared against unobserved ones. They concluded that the pair-wise setting performs consistently better than the more traditional point-wise setting. However, after some detailed investigation, we explain by mathematical derivations that their method may perform only similar to the point-wise ones. We also identified some problems when reproducing their experimental results. After considering suitable settings, we rigorously compare point-wise and pair-wise one-class MFs, and show that the pair-wise method is actually not better. Therefore, for one-class MF, the more traditional and mature point-wise setting should still be considered. Our findings contradict the conclusions in [2] and serve as a call for caution when researchers are comparing between two machine learning methods.

  Full text in ACM Digital Library

• REPR 🕓10Revisiting BPR: A Replicability Study of a Common Recommender System Baseline
  by Aleksandr Milogradskii (National Research University Higher School of Economics; TBank), Oleg Lashinin (Moscow Institute of Physics and Technology; TBank), Alexander P (Independent Researcher), Marina Ananyeva (National Research University Higher School of Economics; TBank) and Sergey Kolesnikov (TBank)

  Bayesian Personalized Ranking (BPR), a collaborative filtering approach based on matrix factorization, frequently serves as a benchmark for recommender systems research. However, numerous studies often overlook the nuances of BPR implementation, claiming that it performs worse than newly proposed methods across various tasks. In this paper, we thoroughly examine the features of the BPR model, indicating their impact on its performance, and investigate open-source BPR implementations. Our analysis reveals inconsistencies between these implementations and the original BPR paper, leading to a significant decrease in performance of up to 50% for specific implementations. Furthermore, through extensive experiments on real-world datasets under modern evaluation settings, we demonstrate that with proper tuning of its hyperparameters, the BPR model can achieve performance levels close to state-of-the-art methods on the top-n recommendation tasks and even outperform them on specific datasets. Specifically, on the Million Song Dataset, the BPR model with hyperparameters tuning statistically significantly outperforms Mult-VAE by 10% in NDCG@100 with binary relevance function.

  Full text in ACM Digital Library

Back to program