Method for generating difficulty-controllable problem based on soft template and anti-factual reasoning

The invention discloses a difficulty controllable problem generation method based on a soft template and anti-factual reasoning. The difficulty controllable problem generation method is mainly used for effectively generating a problem of a corresponding difficulty level according to a specified difficulty label. According to the method, for a given sub-graph, a heuristic template is firstly designed to convert triples in the sub-graph into natural language expressions, and the ontology information of entities and relationships of the sub-graph is filled into the template to strengthen input; then a difficulty assessment method is provided, and the problems in the data set are divided into a simple category and a complex category by using the difficulty assessment method; then, a pre-training language model is utilized to initialize a soft template on data sets with different difficulty levels for guiding problem generation and ensuring grammar correctness and difficulty matching of the problem; and a soft tem