| Abstract |
The technological advances in networking, mobile computing, and systems, have triggered
a dramatic increase in content delivery services. This massive availability of multimedia content and the tight time constraints during searching for the appropriate content, impose various requirements towards maintaining the user engagement. The recommendation systems address this problem by recommending appropriate personalized content to users, exploiting information about their preferences. A plethora of recommendation algorithms has been introduced. However the selection of the best recommendation algorithm in the context of a specific service is challenging. Depending on the input, the performance of the recommendation algorithms varies.
To address this issue, hybrid recommendation systems have been proposed, aiming to improve the accuracy by efficiently combining several recommendation algorithms.
This thesis proposes the εnabler, a hybrid recommendation system which employs various
machine-learning (ML) algorithms for learning an efficient combination of a diverse set of recommendation algorithms and selects the best blending for a given input. Specifically, it integrates three main layers, namely, the trainer which trains the underlying recommenders, the blender which determines the most efficient combination of the recommenders, and the tester for assessing the system's performance. The enabler incorporates a variety of recommendation algorithms that span from collaborative filtering and content-based techniques to recently-introduced neural based ones. The εnabler uses the nested cross-validation for automatically selecting the best ML algorithm along with its hyper-parameter values for the given input, according to a specific metric, avoiding optimistic estimation. Due to its modularity, it can be easily extended to include other recommenders and blenders. The εnabler has been extensively evaluated in the context of video-streaming services.
It outperforms various other algorithms, when tested on the "Movielens 1M" benchmark dataset. For example, it achieves an RMSE of 0.8206, compared to the state-of-the-art performance of the AutoRec [1] and SVD [2], 0.827 and 0.845, respectively. A pilot recommendation system was developed and tested in the production environment of a large telecom operator in Greece. Volunteer customers of the video-streaming service provided by the telecom operator employed the system in the context of a field study. The outcome of this field study was encouraging. Moreover an offline post-analysis of the enabler, using the collected ratings of the pilot study, demonstrated that it significantly outperforms several popular recommendation algorithms, such as the SVD, exhibiting an RMSE improvement higher than 16%.
|
Search
