A training sample validity detection method, a computer device, and a computer non-volatile storage medium, relating to the technical field of artificial intelligence. The method comprises: acquiring multiple extended questions, wherein each extended question is associated with a corresponding preset standard question (S101); randomly dividing the multiple expansion questions into preset copies of sample sets and dividing the preset copies of sample sets into a training set and a cross-validation set according to a preset ratio (S102); training a classification model by using the training set (S103); using the classification model to label the multiple extended questions in the cross-validation set by adopting a cross-validation method until all the extended questions are labeled (S104); acquiring the labeling results of all the extended questions output by the classification model (S105); and obtaining abnormal extended questions according to the labeling results, the labeling results of the abnormal extended questions being different from the associated preset standard questions (S106). The present application can solve the problem of low efficiency of training sample validity detection in the prior art.
展开▼