BTBAP: a model to predict job evolution using patent classification

Understanding technological trends in related fields can increase the accuracy of judgments about job changes. To find representatives for technological development, we study the matching relationship between jobs and patents, and analyse the number of patent applications over time to predict the future changes in jobs. To resolve this problem, we propose a matching model called BERT-TextCNN-BiLSTM-Attention-Prediction (BTBAP) to predict job evolution. It uses BERT pre-training for text embedding, TextCNN, BiLSTM and attention mechanism for feature extraction, achieving a match between patent classification and jobs. Experimental results proved that the accuracy, recall and F1 values of this model in patent classification reached 0.945, 0.944 and 0.944, respectively, which is better compared with other classification models. Through the model, we can obtain the matching degree of each job with the patents, and finally the trend of jobs will be analysed in terms of patent changes.