[Paper Review] How to Retrain Recommender System? A Sequential Meta-Learning Method

Yang Zhang, Fuli Feng+5 more|arXiv (Cornell University)|May 27, 2020

Recommender Systems and Techniques46 references8 citations

TL;DR

This paper proposes Sequential Meta-Learning (SML), a method that retrains recommender systems by learning a transfer network to adapt old models to new data without retraining on historical data. SML optimizes for future recommendation performance, achieving faster training and better accuracy than full retraining on two real-world datasets.

ABSTRACT

Practical recommender systems need be periodically retrained to refresh the model with new interaction data. To pursue high model fidelity, it is usually desirable to retrain the model on both historical and new data, since it can account for both long-term and short-term user preference. However, a full model retraining could be very time-consuming and memory-costly, especially when the scale of historical data is large. In this work, we study the model retraining mechanism for recommender systems, a topic of high practical values but has been relatively little explored in the research community. Our first belief is that retraining the model on historical data is unnecessary, since the model has been trained on it before. Nevertheless, normal training on new data only may easily cause overfitting and forgetting issues, since the new data is of a smaller scale and contains fewer information on long-term user preference. To address this dilemma, we propose a new training method, aiming to abandon the historical data during retraining through learning to transfer the past training experience. Specifically, we design a neural network-based transfer component, which transforms the old model to a new model that is tailored for future recommendations. To learn the transfer component well, we optimize the "future performance" -- i.e., the recommendation accuracy evaluated in the next time period. Our Sequential Meta-Learning(SML) method offers a general training paradigm that is applicable to any differentiable model. We demonstrate SML on matrix factorization and conduct experiments on two real-world datasets. Empirical results show that SML not only achieves significant speed-up, but also outperforms the full model retraining in recommendation accuracy, validating the effectiveness of our proposals. We release our codes at: https://github.com/zyang1580/SML.

Motivation & Objective

• To address the inefficiency and high cost of full model retraining in large-scale recommender systems.
• To overcome overfitting and forgetting issues in fine-tuning by preserving long-term user preference without accessing historical data.
• To optimize the retraining process explicitly for future recommendation performance, rather than relying on heuristic data sampling or manual tuning.
• To develop a general, model-agnostic training paradigm applicable to differentiable recommender models.

Proposed method

• Proposes a sequential meta-learning framework that learns a transfer network to map an old model to a new one based on new interaction data.
• Uses future validation data (next time period interactions) to optimize the transfer network, directly targeting recommendation accuracy on upcoming data.
• Employs a meta-optimization objective that minimizes the expected loss on future data, enabling the model to generalize well to upcoming user preferences.
• Designs a neural network-based transfer component that learns to adapt the old model's parameters to the new data distribution without reprocessing historical data.
• Applies the method to matrix factorization models, demonstrating its applicability to standard collaborative filtering architectures.
• Treats each retraining cycle as a meta-task, where the model learns to generalize across sequential data periods.

Experimental results

Research questions

• RQ1 Can we retrain recommender models efficiently without reprocessing large historical datasets?
• RQ2 How can we preserve long-term user preference signals while adapting to new short-term preferences?
• RQ3 Can future recommendation performance be used as a direct optimization objective for model retraining?
• RQ4 Does learning to transfer past training experience outperform full retraining in both speed and accuracy?

Key findings

• SML achieves significant speed-up compared to full model retraining, as it avoids processing historical data during each retraining cycle.
• On two real-world datasets, SML outperforms full retraining in recommendation accuracy, demonstrating superior generalization to future data.
• SML effectively mitigates overfitting and catastrophic forgetting, even when training only on new data, by leveraging learned transfer from past models.
• The method maintains high model fidelity without accessing historical interactions, validating the hypothesis that past training knowledge can be effectively transferred.
• The transfer network learns meaningful representations, as visualized through filter analysis, showing effective adaptation to new data patterns.