Page 33 - 2022

Broyden–Fletcher–Goldfarb–Shanno (L-BFGS) gradient method. In the third framework,
we have proposed Deep Neural NMF with the integration of Deep Neural Network (DNN)
and NNMF, which improves CF performance by considering multiple attribute ratings of
users rather than single attribute ratings. L-BFGS technique is employed to do parameter
settings for DNN and to decrease error values in the recommendation system. To lower
the overfitting problem in DNN, we have used NNMF.

Extensive experiments are performed for assessing the proposed recommendation
frameworks. Five real world datasets have been used in the experiments: Movie Lens 1M,
Movie Lens 25M, Epinion and Trip Advisor multi-criteria dataset. The experiments consist
of detailed evaluation of the proposed frameworks. Experimental results show that the
proposed approaches outperformed state-of-art approaches. By implementing the
proposed frameworks, we could solve the cold start and the data sparsity problems in the
recommender system.

Key words: Deep Neural Network, Non-Negative matrix factorization, Artificial Neural
Network, L-BFGS, Recommender System, Social Tagging System, Multi-Criteria
recommender System, Latent Neighborhood model, Tensor based methods, Multi Batch
Quasi Newton Method, dimensionality reduction Techniques, HOSVD

02