Biomedical Named Entity Recognition through spaCy: A Visual Exploration

Named Entity Recognition (NER) is a natural language processing subtask that involves identifying and categorizing entities referenced in text into predetermined categories such as person, location, organization, and medical condition. Visualization of various parameters in a dataset can help comprehend the output of a NER system, as well as debug and improve it. In this paper, the spaCy tool is used to test its visualization and biomedical NER capability using the BioNLP 2013 Cancer Genetics dataset. Topic modelling using Latent Dirichlet allocation (LDA) and Non-negative matrix factorization (NMF), sentiment analysis using TextBlob, and effectiveness of spaCy NER performance with four other machine learning techniques is evaluated. It is found that spaCy as a visualization tool is simple and provides deep insights into large complex dataset. Analyzing the sentiments conveyed through medical text, it is found that not all cancer related texts convey negative sentiments and fewer text has strong positive sentiment than negative. In BioNER performance comparison, support vector machine (SVM) based pipeline architecture outperforms all other models including spaCy. Though spaCy is an excellent visualization and generic NER tool, it is not a good BioNER tagger in the biomedical domain.