How can we analyze real-world tensors where additional information is coupled with certain modes of tensors? The problem is effectively solved by coupled matrix-tensor factorization. There are many applications of coupled matrix-tensor factorization such as collaborative filtering, multi-way clustering, and link prediction. SCouT is a large-scale coupled matrix-tensor factorization algorithm running on the distributed Hadoop platform. By reusing intermediate data, carefully ordering computation, and transforming input matrix, SCouT significantly decreases the intermediate data and floating point operations.
The binary code of SCouT is available here.
| Name | Structure | Dimensionality | Nonzero | Download | Description |
|---|---|---|---|---|---|
| Microsoft Academic Graph | Paper-Author-Affiliation Tensor Paper-Field of Study Matrix |
122M*123M*2.7M 122M*47K |
325M 176M |
DOWN | Papers and their metadata |
| MovieLens | User-Movie-YearMonth Tensor Movie-Genre Matrix Movie-Year Matrix |
71K*10K*157 10K*20 10K*94 |
10M 21K 10K |
DOWN | Movie rating data from MovieLens |
| YELP | User-Business-YearMonth Tensor Business-City Matrix Business-Category Matrix User-User matrix |
70K*15K*108 15K*68 15K*590 70K*70K |
334K 15K 590 303K |
DOWN |
Business rating data from YELP |