Intention recognition method and device based on multiple rounds of sessions

The invention discloses an intention recognition method and device based on multiple rounds of sessions. The method comprises the following steps: converting a word vector of each round of question information in M rounds of question information into a sentence vector to obtain M sentence vectors; performing attention coding operation on the M sentence vectors according to a predetermined label matrix to obtain a first session vector; performing GRU encoding on the M sentence vectors to obtain a second session vector; performing convolution operation on the M sentence vectors according to a plurality of preset convolution kernels to obtain a third session vector; and determining intention information of multiple rounds of sessions according to the first session vector, the second session vector and the third session vector. According to the invention, the technical problem of low user intention identification accuracy of multiple rounds of problems is solved. 本发明公开了一种基于多轮会话的意图识别方法及装置。其中，该方法包括：将M轮问题信息中的每轮问题信息的