Research on Short Text Speech Understanding Methods for Factory Applications

In the factory's information processing system, there is a need to use voice input instead of buttons as input, which requires understanding the user's voice input. Unlike general natural language understanding, this type of input is usually short words without contextual connections, making it quite difficult to accurately understand. Firstly, based on the background of this paper, the text classification model and named entity recognition model have been improved and optimized: using TF _IDF model and LDA model have optimized the FastText text classification model. The Bert pre training model and Attention mechanism were used to optimize the BiLstm CRF named entity recognition model. Finally, the optimized text classification model and named entity recognition model were validated on the basis of the factory user voice query input statement data collected in this paper. The results verified the effectiveness of the optimized text classification model and named entity recognition model.