Identifying Research Topics in Human-Computer Interaction for Development: What Value Can Natural Language Processing Techniques Add?

Human Computer Interaction for Development (HCI4D) is an interdisciplinary field involving researchers, practitioners and funding organization from Human Computer Interaction (HCI), Information System (IS) and Development Studies. The fast-growing literature based on HCI4D means researchers need support in identifying the core research areas and current research patterns effectively and efficiently. This study investigates the value added by natural language processing (NLP) techniques when identifying research topics (core research areas) from a set of research publications in HCI4D by comparing the results with the result from traditional, manual literature searches. A generative statistical method called Encoder Representations from Transformers (BERT) and t-distributed Stochastic Neighbor Embedding (tSNE) is used as NLP techniques on a dataset of HCI4D publications. The top 10-word clusters generated were considered for semantic mapping to associated topics. Our findings confirm that NLP techniques are effective in identifying research topics and add value in terms of adding new topics and confirming existing topics. BERT, rather than tSNE, was found useful for supporting literature searches especially in identifying new topics. The inclusion of human experts in labelling the topics improved transparency and added value in terms of structuring to the topics provided from automated searches.