New problem discovery method and device based on clustering

The invention provides a new problem discovery method and device based on clustering, and relates to the technical field of artificial intelligence, and the method comprises the steps: training a classifier according to a new problem with a pseudo tag, obtaining a current feature vector of the new problem based on the trained classifier, clustering the new problem according to the current feature vector, obtaining a current clustering result, and obtaining a new problem according to the current clustering result; and acquiring an alignment label aligned with the pseudo label in the current clustering result in the previous clustering result, taking the alignment label as a new pseudo label, iteratively executing the process until no new clustering cluster exists in the current clustering result, and outputting the alignment label of the currently obtained new problem. A classifier is trained through a new problem with a pseudo tag, clustering and classification are continuously iterated until a better feature