The tripartite tensor decomposition (TTD) model reveals the latent relationship among items, tags and users in social tagging systems in terms of a low order tensor obtained from the high-index sparse data space with the tensor dimensionality reduction technique. The Tripartite decomposition recommendation algorithms can produce high quality recommendations, but have to undergo expensive tensor decomposition steps when new users, new tags, or new items come in, which is significant in light of the tremendous growth in numbers of users, tags and items. In this paper, we present fold-in algorithms for Tripartite tensor decomposition to deal with the new users problem. We evaluate the fold-in algorithms experimentally on several datasets and the results demonstrate the effectiveness of the algorithm.
- recommender system
- tripartite tensor decomposition
- social network