Research on information extraction and sentiment data mining method based on neural network

In this work, an automatic information extraction and sentiment mining based on the neural network will be proposed, aiming at enhancing the efficiency further with deep learning technology for text processing tasks. On the experimental results, comparisons between BERT and more 'usual' Bi-LSTM+CRF models show that BERT is much stronger in both named entity recognition and sentiment analysis. In information extraction tasks, BERT can grasp good context in text and improve the recognition accuracy of named entities significantly. In the sentiment analysis task, BERT's performance is good on multitasking in sentiment classification; it does especially well in predicting positive, negative, and neutral sentiments. Due to its high precision, it also has a very high recall rate, which proves to be very robust. These results proved that BERT did the best in various scenarios, such as on each task, compared with a Bi-LSTM+CRF model, based on comprehensive performance indices and confusion matrix analysis, to further illustrate the wide applicability and efficiency of a neural network-based text-processing method in information extraction and sentiment mining.