SELF-LEARNING METHOD FOR SEMANTIC FEATURE WITH MAXIMUM GAP, AND COMPUTER DEVICE AND STORAGE MEDIUM

Disclosed in the present invention is a self-learning method for semantic features with the maximum gap. The method comprises: constructing an encoder, wherein a constraint condition for the encoder is that the distance between codes of two pieces of relevant data is less than the distance between codes of two pieces of irrelevant data; and driving, by using an unlabeled training set, the encoder to train by means of self-learning. Further disclosed in the present invention are a computer device and a computer-readable storage medium. By using the present invention, there is no need to label training data, and an encoder can be trained by directly using original data. In addition, a gap-based constraint condition is also introduced, such that the effect of a training model in practical applications can be effectively guaranteed.