As everyone can enrich or rather impoverish crowd-sourcing contents, it is a crucial need to continuously improve automatic quality contents assessment tools. Structural-based analysis methods developed for such quality prediction purposes generally handle a limited or manually fixed number of families of nodes and relations. This lack of genericity prevents existing algorithms for being adaptable to platforms evolutions. In this work, we propose a generic and adaptable algorithm, called HSQ, generalising various state-of-the-art models and allowing the consideration of graphs defined by an arbitrary number of nodes semantics. Evaluations performed over the two representative crowd-sourcing platforms Wikipedia and Stack Exchange state that the consideration of additional nodes semantics and relations improve the performances of state-of-the-art approaches.