The disclosure provides a method of acquiring a speech word library in a self-adaptive manner on the basis of history data and machine learning. The method includes: step S1, carrying out sentence pattern classification of a semantic plane on a speech recognition result to find a verb core in a speech instruction and verb elements related thereto; step S2, extracting the verb elements in the speech instruction, and combining machine learning and the user history data to select a plurality of word libraries; step S3, using a method in natural-language processing to carry out word segmentation of a syntax plane in the selected word libraries, synthesizing results of multiple word library fields for evaluation, obtaining a field with a highest evaluation score by solving to use the same as anoptimal result, outputting the optimal result, and updating the user history data at the same time; and step S4, combining the optimal result with sentence type analysis of a pragmatic plane to determine a final word library field. Through combining use situations of the user history word libraries with machine learning, the corresponding field is acquired in the self-adaptive manner from the history data of a user, and thus flexibility and accuracy are greatly improved.