Multi-intention semantic recognition method in intelligent dialogue

The invention discloses a multi-intention semantic recognition method in intelligent dialogues. The method comprises the steps of model initialization, model calling and model updating. In the model initialization, an independent intention model is trained for each type of dialogue scene, and model classification comprises an emotion classification model, an intention classification model and an intention model; the model calling intention processing service comprises the following steps of (1) text cleaning, (2) intention classification, (3) intention splitting, (4) intention screening, (5) intention sorting and (6) multi-intention merging. The input of model updating training is dialogue quality inspection and real dialogue learning. According to the invention, an intelligent recognition algorithm for complex intentions is realized, the intention recognition accuracy of intelligent dialogues is improved, and the dialogue effect is greatly improved. 本发明公开了一种智能对话中多意图语义识别方法，包括：模型初始化、模型调用、模型更新；所述模型初始化对于每一类对话场景，