ADER: Adaptively Distilled Exemplar Replay Towards Continual Learning for Session-based Recommendation

Abstract

Session-based recommendation has received growing attention recently due to the increasing privacy concern. Despite the recent success of neural session-based recommenders, they are typically developed in an offline manner using a static dataset. However, recommendation requires continual adaptation to take into account new and obsolete items and users, and requires "continual learning" in real-life applications. In this case, the recommender is updated continually and periodically with new data that arrives in each update cycle, and the updated model needs to provide recommendations for user activities before the next model update. A major challenge for continual learning with neural models is catastrophic forgetting, in which a continually trained model forgets user preference patterns it has learned before. To deal with this challenge, we propose a method called Adaptively Distilled Exemplar Replay (ADER) by periodically replaying previous training samples (i.e., exemplars) to the current model with an adaptive distillation loss. Experiments are conducted based on the state-of-the-art SASRec model using two widely used datasets to benchmark ADER with several well-known continual learning techniques. We empirically demonstrate that ADER consistently outperforms other baselines, and it even outperforms the method using all historical data at every update cycle. This result reveals that ADER is a promising solution to mitigate the catastrophic forgetting issue towards building more realistic and scalable session-based recommenders. CCS CONCEPTS• Computing methodologies → Neural networks; Online learning settings.

Learning an Adaptive Meta Model-Generator for Incrementally Updating Recommender Systems

“…Though these methods have shown promising results in CV and NLP, how to adapt them to incrementally update RS models is non-trivial, as they lack mechanism to explicitly optimize for the future performance. To overcome the specific forgetting problem in RSs incremental update, two lines of research have been developed in recent years, which are closely related to some of the methods in continual learning [13].…”

Continual Learning for CTR Prediction: A Hybrid Approach

“…There are very few work on CTR prediction addressing the non-stationary and drifting data problem. A recent work studied session-based recommendation task in continual learning setting by utilizing memorybased method to mitigate catastrophic forgetting (Mi, Lin, and Faltings 2020). However, it gave no formal treatment to the non-stationary and drifting pattern of real world data and relied on heuristics to populate memory, with no emphasis on positive knowledge transfer, which is opposite to our method.…”

“…ADER ADER is a continual learning algorithm targeting at session-based recommendation task (Mi, Lin, and Faltings 2020). It updates memory according to items' historical frequency only, thus may suffer from insufficient positive transfer.…”

One Person, One Model, One World

“…In fact, it is largely unknown whether the learning paradigms, frameworks, and methodologies for other domains are useful or not to address our problem. Meanwhile, there are also some recent work in [23,29,30] claiming that RS models should have the so-called 'lifelong' learning capacity. However, their methodologies are designed only to model long-term user behaviors or new training data from the…”

“…Note recent work in[23,30] also introduced a 'lifelong' learning solution for RS, the main dierence between our paper and them is described in Section 2.3.…”