Algorithm Input: Partially labeled documents HIN based on semantic parsing Algorithm: Step 1: extract transition matrices of different meta-graphs Step 2: run personalized random walk based semi- supervised learning Step 3: Ensemble of different meta-graph guided random Walk Output Labels of all unlabeled dataAlgorithm • Input: – Partially labeled documents – HIN based on semantic parsing • Algorithm: – Step 1: extract transition matrices of different meta-graphs – Step 2: run personalized random walk based semi￾supervised learning – Step 3: Ensemble of different meta-graph guided random walk • Output: – Labels of all unlabeled data 25