Image Annotation via Reconstitution Graph Learning Model

Algorithm 2.

Random dot product method for simple graphs.

Input: the weight matrix of the image data graph.

Output: the weight matrix of random point product.

1: Take an all-zero matrix .

2: Find spectral decomposition of .

3: is a matrix of largest eigenvectors, . is a diagonal matrix composed of largest eigenvalues, where each negative eigenvalue is changed to 0.

4:

5: Return 2 until converges.

6: Calculate , return 1 until converges. is the edge probability matrix after random reconstruction, where .

Algorithm 2.