Alleviating Cold-start Problem in CTR Prediction with A Variational Embedding Learning Framework

TL;DR

This paper introduces VELF, a variational embedding framework that effectively mitigates the cold-start problem in CTR prediction by leveraging probabilistic embeddings and regularized priors, demonstrating superior generalization on benchmark datasets.

Contribution

The paper presents a novel variational embedding learning framework that combines probabilistic embeddings with trainable priors to address data sparsity in cold-start scenarios.

Findings

VELF outperforms existing methods on benchmark datasets.

Regularized priors enhance generalization over fixed priors.

Probabilistic embeddings reduce overfitting in cold-start cases.

Abstract

We propose a general Variational Embedding Learning Framework (VELF) for alleviating the severe cold-start problem in CTR prediction. VELF addresses the cold start problem via alleviating over-fits caused by data-sparsity in two ways: learning probabilistic embedding, and incorporating trainable and regularized priors which utilize the rich side information of cold start users and advertisements (Ads). The two techniques are naturally integrated into a variational inference framework, forming an end-to-end training process. Abundant empirical tests on benchmark datasets well demonstrate the advantages of our proposed VELF. Besides, extended experiments confirmed that our parameterized and regularized priors provide more generalization capability than traditional fixed priors.