boost structur inform function space applic graph classif hongliang fei jun huan boost success classif algorithm produc linear combin weak classifi base learner obtain high qualiti classif model paper propos boost algorithm base learner structur relationship function space relationship gener work motiv emerg topic pattern base classif semi structur data includ graph effici incorpor structur inform design gener model undirect graph captur relationship subgraph base base learner method combin norm laplacian base norm penalti logit loss function logit boost approach enforc model sparsiti smooth function space span basi function deriv effici optim algorithm base coordin decent boost formul theoret prove exhibit natur group effect nearbi spatial overlap featur comprehens experiment studi demonstr effect propos learn method 