Sequential Scenario-Specific Meta Learner for Online Recommendation

Abstract

Cold-start problems are long-standing challenges for practical recommendations. Most existing recommendation algorithms rely on extensive observed data and are brittle to recommendation scenarios with few interactions. This paper addresses such problems using few-shot learning and meta learning. Our approach is based on the insight that having a good generalization from a few examples relies on both a generic model initialization and an effective strategy for adapting this model to newly arising tasks. To accomplish this, we combine the scenario-specific learning with a model-agnostic sequential meta-learning and unify them into an integrated end-toend framework, namely Scenario-specific Sequential Meta learner (or s 2 Meta ). By doing so, our meta-learner produces a generic initial model through aggregating contextual information from a variety of prediction tasks while effectively adapting to specific tasks by leveraging learning-to-learn knowledge. Extensive experiments on various real-world datasets demonstrate that our proposed model can achieve significant gains over the state-of-the-arts for cold-start problems in online recommendation 1 . Deployment is at the Guess You Like session, the front page of the Mobile Taobao; and the illustration video can also be watched from the link 2 .

Meta Matrix Factorization for Federated Rating Predictions

“…MeLU can estimate new users' preferences with a few consumed items and determine distinguishing items for customized preference estimation by an evidence candidate selection strategy. Du et al [10] unify scenario-specific learning and model-agnostic sequential meta learning into an integrated end-to-end framework, namely Scenario-specific Sequential Meta learner (s 2 Meta). s 2 Meta can produce a generic initial model by aggregating contextual information from a variety of prediction tasks and effectively adapt to specific tasks by leveraging learning-to-learn knowledge.…”

Adaptive-Step Graph Meta-Learner for Few-Shot Graph Classification

“…We use MAML to implement a fast adaptation mechanism for meta-learner because of its model agnostic property. Du et al [10] proposed an RL based step controller to guide meta-learner for link prediction. We argue that classification Figure 2: Diagram of the AS-MAML framework's learning process in a single episode on the 2-way-1-shot graph classification task.…”

“…Graph-R52 5-way-5-shot 5-way-10-shot 2-way-5-shot 2-way- 10 Table 2: Accuracies with a standard deviation of baseline methods and our framework. We tested 200 and 500 N-way-K-shot tasks on COIL-DEL and Graph-R52, respectively.…”

“…There are many papers discussing few-shot learning with meta-learning [12], data augmentation [53] or regularization [50], but most of them don't consider the graph data. Furthermore, there have been several methods for few-shot node classification [20,38,56] and few-shot link prediction [10,23,32,43], but they only focus on node-level embedding. Recently, Chauhan et al [7] proposed few-shot graph classification based on graph spectral measures and got satisfactory performance.…”

Curriculum Meta-Learning for Next POI Recommendation

“…b) Cross-domain recommendation [3] approaches cannot solve challenge (1): they require a large number of overlapping users or items between source domain and target domain, but the shared data among base and target cities is extremely limited. c) Meta-learning [36] has achieved great success as transfer algorithms in many few-shot learning applications [6,8,9,47], providing potential solutions for our problem. Nevertheless, existing meta-learning methods still cannot solve challenge (2), since they do not explicitly take the diversity among various users in different cities into consideration.…”