Interact2Vec: Neural Item and User Embedding for Collaborative Filtering

Autor(es)

Pires, Pedro R. ; Almeidn, Tiago A. ; Xavier-Junior, JC ; Rios, RA

• Tipo de documento: Artigo Científico
• Fonte: INTELLIGENT SYSTEMS, PT II; v. 13654, p. 16-pg., 2022-01-01.

Resumo

Recommender systems gained great popularity in the last decade. However, despite the significant advances, they still have open problems, such as high data dimensionality and sparseness. Among several alternatives proposed to address these problems, the state-of-the-art solutions aim to represent items and users as dense vectors in a reduced dimensionality space. In this context, one of the most contemporary techniques is neural embeddings-based models, i.e., distributed vector representations generated through artificial neural networks. Many of the latest advances in this area have shown promising results compared to established approaches. However, most existing proposals demand complex neural architectures or content data, often unavailable. This paper presents the Interact2Vec, a new neural network-based model for concomitantly generating a distributed representation of users and items. It has the main advantage of being computationally efficient and only requires implicit user feedback. The results indicate a high performance comparable to other neural embeddings models that demand more significant computational power. (AU)

• Processo FAPESP: 21/14591-7 - Sistemas de recomendação top-N conscientes de tempo baseados em aprendizado por reforço
• Beneficiário: Pedro Reis Pires
• Modalidade de apoio: Bolsas no Brasil - Doutorado