Walking forward and backward: towards graph-based searching and mining

Graph;searching and mining;random walk;semi-supervised learning;smoothness

Fang, Yuan

关键词: Graph;    searching and mining;    random walk;    semi-supervised learning;    smoothness;   

Others  :  https://www.ideals.illinois.edu/bitstream/handle/2142/72873/Yuan_Fang.pdf?sequence=1&isAllowed=y

美国|英语

来源: The Illinois Digital Environment for Access to Learning and Scholarship

【 摘 要 】

Graphs are powerful vehicles to represent data objects that are interconnecting or interacting with each other. We explore random walks on various kinds of graph to address different searching and mining scenarios. This dissertation focuses on two symmetric forms of random walk called the forward walk and backward walk, which can be applied to enrich three key tasks in searching and mining, namely, extraction, ranking and classification, in novel ways. More specifically, we enhance extraction with the metrics of probabilistic precision and recall, ranking with the senses of importance and specificity, and classification with heterogeneous contexts in terms of relationship type and confidence level. We further study the underpinning principle of random walks on a graph, which is often known as the smoothness assumption. We argue that smoothness is a pointwise property and requires probabilistic modeling. Thus, we propose a new framework to define pointwise smoothness probabilistically on a graph, which unifies two different "modes" of smoothness corresponding to the forward and backward random walks, respectively. Finally, our graph-based random walk solutions have consistently demonstrated promising empirical results for a wide range of searching and mining applications.