Bilkent University
Department of Computer Engineering


Tensor Factorization and Its Applications


Zahit Saygın Doğu
MS Student
Computer Engineering Department
Bilkent University

Tensors are multi dimensional arrays, which is a generalization of vectors and matrices. Matrix factorization methods such as SVD captures the latent topics that are hidden in the relationships between 2 dimensions of the matrix. Tensor factorization methods such as Tucker and PARAFAC are the generalization of matrix factorization for higher dimensional data and they capture the multiway relationships between all of the dimensions of the tensor. Tensor factorization has a broad range of application areas such as chemometrics, neuroscience, text mining, social media analysis etc. In this study, we explore the application areas of the tensor factorization, and focus on the topic detection and tracking problem in news articles dataset. By framing the problem as an anomaly detection problem, which is proven to be solvable effectively using tensor factorization, we propose a novel approach to solve topic detection and tracking problem. Future work includes remodelling the problem for better effectiveness and utilize a parallel algorithm in order to improve the efficiency.


DATE: 12 March, 2018, Monday, CS590 & CS690 presentations begin at @ 15:40