Graph-based semi-supervised learning via improving the quality of the graph dynamically

Graph-based semi-supervised learning (GSSL) is an important paradigm among approaches and includes the two processes of graph construction label inference. In most traditional GSSL methods, are completed independently. Once constructed, result inference cannot be changed. Therefore, quality directly determines GSSL’s performance. Most methods make certain assumptions about data distribution, resulting in heavily depends on correctness these assumptions. it difficult to handle complex various distribution for methods. To overcome such issues, this paper proposes a framework named Semi-supervised Learning via Improving Quality Graph Dynamically. it, based weighted fusion multiple clustering results integrated into unified achieve their mutual guidance dynamic improvement. Moreover, proposed general framework, existing can embedded so as improve Finally, working mechanism, effectiveness improving performance advantage compared with other proposal verified through systematic experiments.