Clustering and Topic Modelling of Business Research Trends during COVID-19

During COVID-19, a substantial amount of research literature was published in the field of business research. The new challenges posed by the pandemic tested the resilience of businesses as they tried to find ways to adapt to new business practices during COVID-19. To aid business stakeholders and researchers in understanding the pandemic’s impact on businesses, this study used a variety of methods to analyze a corpus of business-related articles published during COVID-19. To begin, an unsupervised clustering algorithm was used to group research papers similar to each other. We combined the t-distributed stochastic neighbour embedding (t-SNE) and k-means clustering for our research. We clustered the articles into 20 clusters using the elbow method to define the optimum number of clusters. Following the clustering of the literature, we used latent Dirichlet allocation to perform topic modelling on each cluster. We identified 457 topics from the 20 clusters obtained from t-SNE k-means clustering. To these 20 clusters, we manually assigned research themes such as technology, healthcare, economy, finance, supply chain management, education, travel and tourism, trade, mental health, emerging technology, energy sector, policy, workplace and employee, and strategic decision-making. Additionally, co-occurrence analysis was used to determine the relationships between keywords. The abstract text from all included studies was used to create a network map of co-occurrences. We concentrated on the word “business” to highlight the relationships and the direction of keyword connections. Finally, we provide a word cloud of the abstract text and a world cloud derived from the topic modelling keywords. At the end of the chapter, we check the clustering efficiency using the stochastic gradient descent classifier, and we also discuss the applicability and utility of each of these methods.