Uncovering Thematic correlations across Transportation Research journal series: Pitting human expertise against machine intelligence

Transportation research plays a significant role in addressing complex societal challenges. The Transportation Research (TR) journal series, comprising six specialized parts A-F, mirrors the thematic breadth of transportation research but presents challenges for certain researchers in clearly delineating thematic boundaries between journals, which could lead to manuscript misalignment and unfortunate desk rejections. Despite the significance of these journals, a systematic analysis of thematic overlaps and distinctions across the TR series using text classification methods remains unexplored. To fill this gap, this study first applies the BERTopic model on 16,341 TR abstracts between 2010 and 2024 to derive the topic distribution of each journal. Three machine learning classifiers and one deep learning algorithm are then trained to classify abstracts accurately into the appropriate TR journal part. Additionally, the journal relationships are analyzed using novel quantitative metrics. A survey inviting 2400 active transportation researchers is conducted to understand the classification performance of human experts. The study finds significant thematic overlaps, especially between TR-B and TR-C, predominantly around driving safety and traffic control, whereas TR-F emerges with highly distinctive thematic clarity. The support vector classifier (SVC) achieves the highest accuracy. When evaluated with the same testing dataset, the SVC significantly outperforms human experts according to the survey results. We publish our machine learning-driven classification tool, which can be used in manuscript submission processes to enhance the accuracy of journal selection.