Joint analysis of data from different sources can potentially improve one's ability to reveal latent structure in heterogeneous datasets. For instance, social network activities and user demographic information can be leveraged to improve recommendations. However, the incompleteness and heterogeneity of the data challenge joint factorization of multiple datasets. Aspiring to address these challenges, the coupled graph tensor factorization model accounts for side information available in the form of correlation matrices or graphs. Here, a novel ADMM-based approach is put forth to impute missing entries and unveil hidden structure in the data. The iterative solver enjoys closed-form updates that result in reduced computational complexity. Numerical tests with synthetic and real data corroborate the merits of the proposed method relative to competing alternatives.
|Original language||English (US)|
|Title of host publication||2018 IEEE Global Conference on Signal and Information Processing, GlobalSIP 2018 - Proceedings|
|Publisher||Institute of Electrical and Electronics Engineers Inc.|
|Number of pages||5|
|State||Published - Feb 20 2019|
|Event||2018 IEEE Global Conference on Signal and Information Processing, GlobalSIP 2018 - Anaheim, United States|
Duration: Nov 26 2018 → Nov 29 2018
|Name||2018 IEEE Global Conference on Signal and Information Processing, GlobalSIP 2018 - Proceedings|
|Conference||2018 IEEE Global Conference on Signal and Information Processing, GlobalSIP 2018|
|Period||11/26/18 → 11/29/18|
Bibliographical noteFunding Information:
The work in this paper was supported by NSF grants 171141, 1500713, and 1442686.
- Non-negative factorization
- Parallel factor (PARAFAC)/ canonical polyadic decomposition (CPD) model
- Recommender systems