Start Date
14-12-2012 12:00 AM
Description
Traditional recommender system research often explores customer demographics, product characteristics, and transactions in providing recommendations. This study investigates the recommendation problem based on social network information. In light of the social network theories on the formation of a social network and its impact on human behavior, we present a multi-theoretical framework to model multiple facets of social relations for recommendation. Taking a kernel-based framework, we design and select kernels describing individuals’ similarities projected by social network theories. Moreover, we employ a non-linear multiple kernel learning approach to combine the kernels to increase the dimension of models on assessing individuals’ opinions. We evaluate our proposed framework on a real-world movie review dataset. The experiments show that our framework provides more accurate recommendations than trust-based methods, the collaborative filtering approach, and individual kernels. Further analysis shows that kernels derived from contagion theory and homophily theory contribute a larger portion of the framework.
Recommended Citation
Li, Xin and Wang, Mengyue, "A Multi-theoretical Framework for Social Network-based Recommendation" (2012). ICIS 2012 Proceedings. 16.
https://aisel.aisnet.org/icis2012/proceedings/KnowledgeManagement/16
A Multi-theoretical Framework for Social Network-based Recommendation
Traditional recommender system research often explores customer demographics, product characteristics, and transactions in providing recommendations. This study investigates the recommendation problem based on social network information. In light of the social network theories on the formation of a social network and its impact on human behavior, we present a multi-theoretical framework to model multiple facets of social relations for recommendation. Taking a kernel-based framework, we design and select kernels describing individuals’ similarities projected by social network theories. Moreover, we employ a non-linear multiple kernel learning approach to combine the kernels to increase the dimension of models on assessing individuals’ opinions. We evaluate our proposed framework on a real-world movie review dataset. The experiments show that our framework provides more accurate recommendations than trust-based methods, the collaborative filtering approach, and individual kernels. Further analysis shows that kernels derived from contagion theory and homophily theory contribute a larger portion of the framework.