One Person, One Model, One World: Learning Continual User Representation without Forgetting

By Fajie Yuan, Guoxiao Zhang, Alexandros Karatzoglou, Xiangnan He, Joemon M. Jose, Beibei Kong, Yudong Li. 2020

83 citations recorded by the source corpus.

Learning user representations is a vital technique toward effective user modeling and personalized recommender systems. Existing approaches often derive an individual set of model parameters for each task by training on separate data. However, the representation of the same user potentially has some commonalities, such as preference and personality, even in different tasks. As such, these separately trained representations could be suboptimal in performance as well as inefficient in terms of parameter sharing. In this paper, we delve on research to continually learn user representations task by task, whereby new tasks are learned while using partial parameters from old ones. A new problem arises since when new tasks are trained, previously learned parameters are very likely to be modified, and as a result, an artificial neural network (ANN)-based model may lose its capacity to serve for well-trained previous tasks forever, this issue is termed catastrophic forgetting. To address this issue, we present Conure the first continual, or lifelong, user representation learner --- i.e., learning new tasks over time without forgetting old ones. Specifically, we propose iteratively removing less important weights of old tasks in a deep user representation model, motivated by the fact that neural network models are usually over-parameterized. In this way, we could learn many tasks with a single model by reusing the important weights, and modifying the less important weights to adapt to new tasks. We conduct extensive experiments on two real-world datasets with nine tasks and show that Conure largely exceeds the standard model that does not purposely preserve such old "knowledge'', and performs competitively or sometimes better than models which are trained either individually for each task or simultaneously by merging all task data.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Information Systems

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. Wayne Xin Zhao
2. Yongfeng Zhang
3. Yupeng Hou
4. Hongyu Lu
5. Zuohui Fu
6. Ruobing Xie
7. Junjie Zhang
8. Shijie Geng

In Computer Vision and Pattern Recognition

1. Yongfeng Zhang
2. Wayne Xin Zhao
3. Zuohui Fu
4. Yupeng Hou
5. Fajie Yuan
6. Hongyu Lu
7. Zihan Lin
8. Shijie Geng

In Information Systems

1. Wayne Xin Zhao
2. Yongfeng Zhang
3. Yupeng Hou
4. Hongyu Lu
5. Zuohui Fu
6. Yingqiang Ge
7. Shijie Geng
8. Shuchang Liu

Citations in the corpus

• Continual Learning for Task-oriented Dialogue System with Iterative Network Pruning, Expanding and Masking (2021, background): “Continual learning (Parisi et al., 2019; Wu et al., 2018; Yuan et al., 2020, 2021) is hardly a new idea for machine learning, but remains as a non-trivial step for building empirically successful AI systems.”
• Chat-REC: Towards Interactive and Explainable LLMs-Augmented Recommender System (2023, background): “To address this issue, solutions have been proposed that either learn to model content features [16] or transfer representations from auxiliary domains [24,26].”
• Text Is All You Need: Learning Language Representations for Sequential Recommendation (2023, unknown): “Previous methods for knowledge transfer mainly rely on shared information between the source and target domains including common users [13, 31, 32, 35], items [26, 39] or attributes [29].”
• Iterative Network Pruning with Uncertainty Regularization for Lifelong Sentiment Classification (2021, background): “This is especially obvious when the pruning ratio is higher [41].”
• Large Language Models are Zero-Shot Rankers for Recommender Systems (2023, background): “We would like to take a more focused study on how they improve the recommendation performance of LLMs. Instruction tuning improves the ranking abilities of LLMs. Existing works show that instruction tuning significantly improves the generalization abilities of LLMs on unseen tasks [27, 32].”
• Recommendation as Language Processing (RLP): A Unified Pretrain, Personalized Prompt & Predict Paradigm (P5) (2022, background): “[72] Fajie Yuan, Guoxiao Zhang, Alexandros Karatzoglou, Joemon Jose, Beibei Kong, and Yudong Li.”