A Deep Learning-based Recognition Model for Chinese Book Subject Words

In order to effectively identify subject words in Chinese books and documents, this paper proposes an automatic recognition model for subject words based on deep learning. The model first builds word vectors with the word to vector (Word2vec) model to obtain the feature information at the semantic granularity level of the vocabulary, and then uses the deep neural network (DNN) model to train the feature weights of the vocabulary to predict the probability that the keywords belong to the subject words to achieve binary classification. Finally, experimental results on a library bibliographic data set show that the TopicDNN model has a prediction accuracy of 85.32%, which has better performance for subject words recognition than traditional machine learning methods.