| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 1022.21 KB | Adobe PDF |
Advisor(s)
Abstract(s)
Nowadays, with the widely usage of on-line stream video platforms, the number of media resources available and the volume
of crowd-sourced feedback volunteered by viewers is increasing
exponentially. In this scenario, the adoption of recommendation
systems allows platforms to match viewers with resources. However, due to the sheer size of the data and the pace of the arriving
data, there is the need to adopt stream mining algorithms to build
and maintain models of the viewer preferences as well as to make
timely personalised recommendations. In this paper, we propose
the adoption of optimal individual hyper-parameters to build more
accurate dynamic viewer models. First, we use a grid search algorithm to identify the optimal individual hyper-parameters (IHP)
and, then, use these hyper-parameters to update incrementally the
user model. This technique is based on an incremental learning
algorithm designed for stream data. The results show that our approach outperforms previous approaches, reducing substantially
the prediction errors and, thus, increasing the accuracy of the recommendations.
Description
Keywords
Stream Content Personalisation Collaborative Filtering HyperParameter Optimisation
Citation
Publisher
CEUR Workshop Proceedings