Collecting valid information from electronic sources to detect the potential outbreak of infectious disease is time-consuming and labor-intensive. The automated identification of relevant information using machine learning is necessary to respond to a potential disease outbreak. A total of 2864 documents were collected from various websites and subsequently manually categorized and labeled by two reviewers. Accurate labels for the training and test data were provided based on a reviewer consensus. Two machine learning algorithms—ConvNet and bidirectional long short-term memory (BiLSTM)—and two classification methods—DocClass and SenClass—were used for classifying the documents. The precision, recall, F1, accuracy, and area under the curve were measured to evaluate the performance of each model. ConvNet yielded higher average, min, and max accuracies (87.6%, 85.2%, and 91.1%, respectively) than BiLSTM with DocClass, while BiLSTM performed better than ConvNet with SenClass with average, min, and max accuracies of 92.8%, 92.6%, and 93.3%, respectively. The performance of BiLSTM with SenClass yielded an overall accuracy of 92.9% in classifying infectious disease occurrences. Machine learning had a compatible performance with a human expert given a particular text extraction system. This study suggests that analyzing information from the website using machine learning can achieve significant accuracies in the presence of abundant articles/documents.

Named Entity Recognition (NER) plays a vital role in natural language processing (NLP). Currently, deep neural network models have achieved significant success in NER. Recent advances in NER systems have introduced various feature selections to identify appropriate representations and handle Out-Of-the-Vocabulary (OOV) words. After selecting the features, they are all concatenated at the embedding layer before being fed into a model to label the input sequences. However, when concatenating the features, information collisions may occur and this would cause the limitation or degradation of the performance. To overcome the information collisions, some works tried to directly connect some features to latter layers, which we call the delayed combination and show its effectiveness by comparing it to the early combination. As feature encodings for input, we selected the character-level Convolutional Neural Network (CNN) or Long Short-Term Memory (LSTM) word encoding, the pre-trained word embedding, and the contextual word embedding and additionally designed CNN-based sentence encoding using a dictionary. These feature encodings are combined at early or delayed position of the bidirectional LSTM Conditional Random Field (CRF) model according to each feature’s characteristics. We evaluated the performance of this model on the CoNLL 2003 and OntoNotes 5.0 datasets using the F1 score and compared the delayed combination model with our own implementation of the early combination as well as the previous works. This comparison convinces us that our delayed combination is more effective than the early one and also highly competitive.