Task Aligned Meta-learning based Augmented Graph for Cold-Start Recommendation

TL;DR

This paper proposes a Task aligned Meta-learning based Augmented Graph (TMAG), a fine-grained task aligned constructor is proposed to cluster similar users and divide tasks for meta-learning, enabling consistent optimization direction and validate the approach on three real-world datasets in various cold-start scenarios.

Abstract

The cold-start problem is a long-standing challenge in recommender systems due to the lack of user-item interactions, which significantly hurts the recommendation effect over new users and items. Recently, meta-learning based methods attempt to learn globally shared prior knowledge across all users, which can be rapidly adapted to new users and items with very few interactions. Though with significant performance improvement, the globally shared parameter may lead to local optimum. Besides, they are oblivious to the inherent information and feature interactions existing in the new users and items, which are critical in cold-start scenarios. In this paper, we propose a Task aligned Meta-learning based Augmented Graph (TMAG) to address cold-start recommendation. Specifically, a fine-grained task aligned constructor is proposed to cluster similar users and divide tasks for meta-learning, enabling consistent optimization direction. Besides, an augmented graph neural network with two graph enhanced approaches is designed to alleviate data sparsity and capture the high-order user-item interactions. We validate our approach on three real-world datasets in various cold-start scenarios, showing the superiority of TMAG over state-of-the-art methods for cold-start recommendation.

Figures (5)

• Figure 1: Visualization of model parameter $\boldsymbol{\theta}$ of the conventional framework with 4 tasks and the proposed TMAG with 2 tasks aligned by ages. The gradient descent direction in conventional meta-learning framework is biased towards $u_4$, while in TMAG users in the same group have consistent optimization direction thus avoiding such local optimum.
• Figure 2: The overview of the proposed TMAG framework. (a) In the pretrain phase, we use two attribute-oriented autoencoders to learn user and item attribute embeddings. (b) In the meta-training phase, we cluster users into different groups and formulate each group of users as a task. These tasks are leveraged to train the meta-training model, and we obtain the parameter $\boldsymbol{\theta}$. (c) In the meta-testing phase, firstly, we employ the learned $\boldsymbol{\theta}$ as the initialization parameter to get user and item representations. Then, we augment the adjacency matrix of the graph by combining the graph structure information and attribute information. At last, we employ the updated $\boldsymbol{\theta}'$ for recommendation.
• Figure 3: Impacts of the size of support sets.
• Figure 4: Impacts of augmentation coefficient.
• Figure 5: Visualization of t-SNE projected representations derived from NGCF and TMAG. Users are randomly selected in the same task from Yelp and are represented by stars. The points with the same color are users' interacted items.