Bichitrananda Behera

An Automated Biomedical Text Classification Process

In general, biomedical text documents constitute unstructured text documents from different biomedical repositories like PubMed and MEDLINE, web blogs, e-newspapers, medical reports, and social media. Figure 1 shows the overall architecture of an automatic biomedical text document classification process.

Figure 1. Biomedical Text Document Classification Process

Automatic biomedical text document classification consists of three major modules, namely, text preprocessing, building a classifier and model evaluation. The input to the text preprocessing module is the raw biomedical text documents from which it extracts valuable words or features and represents the documents in a suitable format through well-defined data preprocessing methods, feature extraction and feature reduction techniques. Methods such as tokenization, stop word removal, lemmatization, and stemming are used in data preprocessing to remove noisy and unwanted words from the documents to improve the performance of the classification model. Bag-of-Words (BOW), Term Frequency-Inverse Document Frequency (TF-IDF), Word Embedding, Word2Vec, N-gram, Global Vectors for Word Representation (Glove) and FastText are the feature extraction methods used to extract the useful features from the documents for the matured document representation. Finally, feature reduction method includes feature selection/feature transformation techniques to minimize the text features without changing its indent meaning of the content.

The original idea behind the classification process is to construct a classification model, i.e., a classifier from the training dataset by relating the features within the text documents to one of the target class labels. Once the classifier is trained, it must own the predicting knowledge of assigning category labels to the test dataset. The model evaluation module includes the phenomenon of verifying the proof of correctness of the built classifier using the state-of-the-art measurements like accuracy, error rate, precision, recall, and F-1 scores.