Fig. 1

Schematic workflow in the operation of the deep neural network classifier. The first step is the data acquisition, performed using PaperBot, a tool to search peer-reviewed publications from multiple sources using keywords. These acquired publications are each manually annotated as relevant or irrelevant depending on whether they describe new neural reconstructions. The second step includes filtering away the publications extracted from pdf files (due to lower text quality), balancing relevant and irrelevant data sets, extracting relevant paragraphs from the text to reduce noise and improve performance, and tokenizing to vectorize the text. The final step divides the data into train, validation, and test sets to optimize the deep learning classifier hyper-parameters and evaluate performance