NISER: Normalized Item and Session Representations to Handle Popularity Bias

TL;DR

This paper addresses popularity bias in session-based graph neural network recommendation models by proposing a normalization technique for item and session representations, leading to improved recommendations for long-tail and new items.

Contribution

The authors introduce a normalization-based training method that reduces popularity bias in GNN-based session recommendation models, enhancing long-tail item recommendation performance.

Findings

Normalized representations improve long-tail item recommendations

The approach outperforms state-of-the-art on benchmark datasets

Significant gains in online and offline settings for less popular items

Abstract

The goal of session-based recommendation (SR) models is to utilize the information from past actions (e.g. item/product clicks) in a session to recommend items that a user is likely to click next. Recently it has been shown that the sequence of item interactions in a session can be modeled as graph-structured data to better account for complex item transitions. Graph neural networks (GNNs) can learn useful representations for such session-graphs, and have been shown to improve over sequential models such as recurrent neural networks [14]. However, we note that these GNN-based recommendation models suffer from popularity bias: the models are biased towards recommending popular items, and fail to recommend relevant long-tail items (less popular or less frequent items). Therefore, these models perform poorly for the less popular new items arriving daily in a practical online setting. We demonstrate that this issue is, in part, related to the magnitude or norm of the learned item and session-graph representations (embedding vectors). We propose a training procedure that mitigates this issue by using normalized representations. The models using normalized item and session-graph representations perform significantly better: i. for the less popular long-tail items in the offline setting, and ii. for the less popular newly introduced items in the online setting. Furthermore, our approach significantly improves upon existing state-of-the-art on three benchmark datasets.