International Conference on Computer and Knowledge Engineering

Home / 15th International Conference on Computer and Knowledge Engineering

Introducing Meta-Contrastive Adaptive Autoencoder to Tackle Cold-Start Challenges in Sparse Domains

Authors :

Hossein Rashid¹ Erfan Arzhmand² Fatemeh Hosseini³

1- Department of Computer Engineering Qazvin Islamic Azad University Qazvin, Iran 2- Department of Computer Engineering South Tehran Branch, Islamic Azad University Tehran, Iran 3- School of Electrical and Computer Engineering University of Tehran

Keywords :

Cold-start recommendation،Meta-learning،Contrastive autoencoder،Temporal drift modeling،Sparse interaction matrix

Abstract :

Cold-start recommendation remains a significant obstacle in sparse environments where user or item data is limited. This paper introduces the Meta-Contrastive Adaptive Autoencoder (MeCAA), a unified model combining contrastive representation learning, meta-adaptation, and temporal drift tracking to improve recommendation quality under data scarcity. MeCAA employs dual autoencoders with contrastive objectives to learn robust embeddings, a meta-learning engine for few-shot personalization, and a recurrent mechanism to capture latent preference evolution. We evaluate MeCAA on a Spotify-derived playlist dataset and benchmark its performance against Neural Collaborative Filtering (NCF) and MetaKG. Across metrics including Recall@10, NDCG@10, and ColdStartHitRate, MeCAA achieves consistent improvements, especially in cold-start scenarios. These findings position our proposed method as an extensible framework for dynamic, sparse recommender systems.