Computational Intelligence Strategies for Effective Collaborative Decisions

Emelia Opoku Aboagye,Jianbin Gao
DOI: https://doi.org/10.1109/iemcon.2018.8615069
2018-01-01
Abstract:Matrix Factorization models have proven a remarkable achievements in data and image processing. It however has a wholesome computational complexity. In view of that, a hybrid fusion strategy that maximizes feature engineering and improves generalization thereby solving cold start problem of collaborative recommendation is proposed in this paper. Present e-commerce sites mainly recommend pertinent items or products to a lot of users through personalized recommendation. Such personalization depends on large extent on scalable systems which strategically responds promptly to the request of the numerous users accessing the site (new users). Tensor Factorization (TF) provides scalable and accurate model for collaborative filtering (CF) modelling. And feed forward neural networks enhances feature engineering. In this paper, we propose a fusion-based recommendation system to address both cold start and scalability. We propose to use a multi-modal approach which represent a multiview data from users, according to their purchasing and rating history. We use a Deep Learning approach to map item and user inter-relationship to a dimensional feature space where item-user resemblance and their preferred items is maximized. From our experiments, our novel deep learning multitask tensor factorization (NeuralFil) analysis is computationally less expensive, scalable and addresses the cold-start problem, for optimal recommendation decision making (Abstract)
What problem does this paper attempt to address?