Active link prediction method based on knowledge rule guidance

The invention relates to an active link prediction method based on knowledge rule guidance, belongs to the field of link prediction, and guides two links of sample selection and model training through logic rules. In a sample selection link, a rule-based uncertainty sampling method is adopted, and information entropy uncertainty is combined, so that high-quality training samples rich in rule information and semantic information are screened out. In a model training link, rules and semantic information in a training set are fully utilized by utilizing rule information, so that a link prediction effect is improved. Comprising the steps of obtaining and processing knowledge graph data, obtaining an initial training set and a rule set, training a link prediction model, calculating information entropy uncertainty and rule uncertainty of an unlabeled sample, selecting a high-uncertainty sample for manual labeling by combining the information entropy uncertainty and the rule uncertainty, and finally updating the tra