Authors: Sangeetha Kutty, Richi Nayak
Conference: 27th Symposium on Applied Science (SAC 2012)
Date: 26/3/2012 - 30/3/2012
Existing recommendation systems often recommend products to users by capturing the item-item similarity measures and/or use ratings data. However, with the people- to-people networks it is inecient to utilize these types of recommendations as these networks recommend people to people and it has a two way relationship unlike the user- to-item networks which has only an one-way relationship. Also, the user-to-item methods use traditional two dimensional models to nd inter relationships between alike users and products but it is not sucient and ecient enough to model the people-to-people network. In this paper, we propose a novel hybrid recommendation method using the tensor decomposition-based recommendation model to recommend people-to-people based on their proles and interactions, and then nding associations within such groups for making eective recommendations. The use of tensor decompositions enable to identify latent correlations which helps to improve the quality of recommendations. As the people-to-people network data is multi-dimensional data which when modeled using vector based methods tend to result in information loss. Hence, this paper utilizes tensors that have the ability to highly correlate and nd latent relationships between similar users that facilitate in generating recommendations. Empirical analysis is conducted on a real-life online dating dataset and the results show the potential of the proposed hybrid recommendation system.
Keywords: people-to-people network, recommendation, tensor, structure and content, decomposition