Name
Abstract | ||
|---|---|---|
Recommendation system, as one of cost effective solutions dealing with the information-overwhelming problem, has been adopted by many internet e-commerce platforms, such as Amazon, eBay, Asos, etc. User-based or item-based collaborative filtering, as one of the classic recommendation algorithms, suffers greatly from high computational consumption and matrix complexity when big data is involved. While neural network architecture based deep learning technique, on the other hand, performs outstandingly as an alternative solution solving regression and classification problems, especially with sparse inputs. Furthermore, deep learning alleviates the cold start problem to a certain extent which is an unavoidable flaw in collaborative filtering (CF) approach based recommendation algorithms. This paper proposes a deep learning network structure which utilizes the Restricted Boltzmann Machine and the artificial neural network for Weibo Users. The proposed structure is trained and tested through the actual dataset provided by VComic, who is an online comic book provider and shares the information with China's biggest social network - Weibo.com. The offline experiment shows that the proposed system outperforms the user-based collaborative filtering algorithm in the metrics of precision and coverage. Specifically, the proposed mechanism demonstrates the ability to mine the long tail under the premise of accuracy guarantee, as well as to reduce the system's complexity dramatically. |
Year | DOI | Venue |
|---|---|---|
2018 | 10.1109/WTS.2018.8363939 | 2018 Wireless Telecommunications Symposium (WTS) |
Keywords | Field | DocType |
Recommendation System,Deep learning,Dimension Reduction,Restricted Boltzmann Machine,Coverage,Prediction accuracy | Recommender system,Restricted Boltzmann machine,Collaborative filtering,Cold start,Computer science,Real-time computing,Artificial intelligence,Deep learning,Artificial neural network,Big data,Machine learning,The Internet | Conference |
ISSN | ISBN | Citations |
1934-5070 | 978-1-5386-3396-0 | 0 |
PageRank | References | Authors |
0.34 | 2 | 6 |