Hybrid Recommender System Leveraging Stacked Convolutional Networks

Recommendation Systems has emerged as an essential component in web-based systems, as their ability to analyze customers' behavior and generate recommendations seeking customers' satisfaction is successfully accomplished. However, the success of these systems depends on amount of customers' personal preference data and content (items') metadata available for harnessing. Therefore, data sparsity poses a major challenge here. To alleviate this problem, data and models from other domains can be leveraged to gain good insight about customers' preferences and content similarities. In specific, this paper proposes the idea of extracting knowledge for transfer learning leveraging pre-trained deep neural networks. Knowledge from pre-trained models is used to efficiently identify similarity and capture customers' preference among the contents. To attain the objective, this paper presented an approach, for generating efficient top-n recommendations using a hybrid recommender model. Performance analysis is performed on the proposed approach and results obtained are promising. Furthermore, extensions for this work are also discussed