An Active Noise Correction Graph Embedding Method Based on Active Learning for Graph Noisy Data

© 2019, ICST Institute for Computer Sciences, Social Informatics and Telecommunications Engineering. In various scenarios of the real world, there are various graph data. Most graph structures are confronted with the problems of complex structure and large consumption of memory space. Graph embedding is an effective method to overcome such challenges, which converts graph structure into a low-dimensional dense vector space. In the real world, label acquisition is expensive, and there may be noise in the data. Therefore, it is important to find valuable noise nodes as much as possible to improve the performance of downstream task. In this paper, we propose a novel active sampling strategy for graph noisy data named Active Noise Correction Graph Embedding method (ANCGE). Given the label budget, the proposed method aims to use semi-supervised graph embedding algorithm to find valuable mislabeled nodes. ANCGE measures the value of noise nodes according to their representativeness and influence on the graph. The experimental results on three open datasets demonstrate the effectiveness of our method and its stability under different noise rates