According to the survey, the performance of the existing knowledge graph answering system in entity linking, relation linking and SPARQL query construction in terms of execution time and accuracy cannot meet the requirements of knowledge graph answering system. For this challenge, a new feedback mechanism based Knowledge-Driven query construction method is proposed. This method takes the entity set and predicate set in the problem as input, and constructs SPARQL query statements in a knowledge-driven way to solve simple and complex problems, and further proposes heuristic ideas to deal with implicit entity problems. At the same time, the method also proposes to feed back the query results of the knowledge graph to the entity link and relationship connection steps, so that the SPARQL query statement is optimized again. The evaluation results of the LC_QuAD data show that this method outperform the existing state of the art in precision and recall rate.