Knowledge graph construction commonly relies on multi-stage pipelines that combine extraction, matching, linking, and fusion tasks.  Each task may provide multiple implementations and parameter settings, yielding a large configuration space in which task-specific decisions can strongly affect end-to-end graph quality.  In practice, task and pipeline configurations are often selected manually, leaving many promising pipeline variants unexplored.

We formulate the configuration of KG integration pipelines as a budget-constrained optimization problem over task implementations and 
parameters.  We introduce a quality-aware search framework that iteratively generates valid configurations, executes the corresponding pipelines, evaluates the resulting knowledge graphs, and uses quality feedback to guide subsequent choices.  The framework enables a direct comparison of search strategies under a common evaluation protocol, including random search, Bayesian optimization, random neighborhood search, and a structure-aware heuristic refinement strategy.  The evaluation on structured RDF and text-based KG integration scenarios shows that high-quality configurations can be identified under limited evaluation budgets, while the relative effectiveness of the search strategies depends on the structure and quality landscape of the configuration space.