The present invention provides a method for generating a semantic expression for a standard question in a knowledge base. The knowledge base includes multiple standard questions, each standard question has multiple associated similar questions, and the method includes: for each standard question, obtaining multiple similar question segmentation results corresponding to the multiple similar questions of the standard question, where each similar question segmentation result includes word classes to which respective words in a corresponding similar question belong; for each standard question, selecting a phrase from an intersection of multiple similar question segmentation results of the standard question based on phrase occurrence frequencies, to form at least one semantic expression of the standard question, where each phrase includes a predetermined quantity of word classes; for all similar questions of all standard questions, performing standard question matching processing; for each semantic expression created for each standard question, determining whether the semantic expression is matched to at least one similar question of the standard question; and if yes, marking the semantic expression as a first state; or otherwise, marking the semantic expression as a second state; and deleting all semantic expressions in the second state.