Most　real-word　data　can　be　modeled　as　heterogeneous　information　networks　(HINs),　which　are　composed　of　multiple　types　of　nodes　and　links.　Classification　for　objects　in　HINs　is　a　fundamental　problem　with　broad　applications.　However,　traditional　methods　cannot　involve　in　heterogeneous　information　networks.　These　approaches　could　not　involve　the　relatedness　between　objects　and　various　path　semantics.　In　this　paper,　we　proposed　a　novel　framework　called　CHIN　for　classification.　It　utilizes　the　relevance　measurement　on　objects　to　iteratively　label　objects　in　HINs.　As　different　meta-path　performs　different　accuracy　for　classification,　the　proposed　framework　incorporates　the　weights　of　meta-paths.　As　our　experiments　show,　CHIN　generates　more　accurate　classes　than　the　other　classification　algorithm,　but　also　provides　meaningful　weights　for　meta-paths　for　classification　task.　©　2018,　Springer　Nature　Switzerland　AG.