Recommendations on websites could improve with new algorithm

Tensor completion algorithm proposed for use in making recommendations via Netflix and Amazon.

fatoumata sangog webFatoumata SanogoPredictions and recommendations on streaming services and online shopping may see a new algorithm soon. Fatoumata Sanogo, a doctoral student in the University of Alabama at Birmingham College of Arts and Sciences, will present a new algorithm for tensor completion at the 2018 Annual SIAM Meeting in Portland, Oregon.

“When you open your Netflix page, they recommend movies or shows that you might be interested in watching,” said Sanogo, a student in the UAB Department of Mathematics. “This is based on an algorithm that looks at the movies and shows you have watched in the past. While most tensor completion methods use the Tucker model, our new approach uses the canonical polyadic decomposition model to reconstruct the unknown tensor or matrix.”

The matrix, or tensor, completion problem is about filling in missing entries from the partially observed entries of the matrix. Sanogo is working to find the unknown tensor from a given a tensor with partially observed data. 

Recommendation systems are supported by a matrix or tensor completion algorithm or filling in missing entries from partially observed entries of the matrix. The new model reconstructs the unknown matrix by finding the optimal factors through linear least squares and the singular vectors through a proximal algorithm of soft thresholding.

Fatoumata was selected to attend the 2018 Statistical and Applied Mathematical Sciences Institute workshop in July at North Caroline State University. Sanogo also won a Society for Industrial and Applied Mathematics travel award to present her work on tensor completion at the 2018 Annual SIAM Meeting in Portland, Oregon.

The two-week SAMSI workshop is highly competitive selecting only 35 to 40 students nationwide each year. Attendees work in groups of five to seven with faculty mentors to solve a real-world industrial problem.