Classifier of human behavioral activities

Based on the ernie3.0 framework, the ernie3.0 pre-training model and tokenizer are loaded through PaddleNLP, the text is divided into non-activities and activities, and then the non-activities and activities are divided into non-necessary activities and necessary activities respectively; optional activities and social activities, and then define the optimizer, loss function, evaluation index, etc. required for training, and pre-model fine-tuning, through pre-training, let the machine learn to automatically analyze these semantic tendencies. Secondly, Python is used to crawl the Weibo tweet data, and the data is cleaned and preprocessed to provide basic data for classification research. Finally, based on Ernie 3.0 training and learning, the microblog text is predicted, and it is divided into spontaneous activities, social activities and other activities according to the recognition results, and the prediction results are saved.