Natural language processing techniques to reveal human-computer interaction for development research topics

Chiyangwa, Tawanda Blessing and van Biljon, Judy and Renaud, Karen (2021) Natural language processing techniques to reveal human-computer interaction for development research topics. In: 2021 International Conference on Artificial Intelligence and its Applications, 2021-12-09 - 2021-12-10.

[thumbnail of Chiyangwa-etal-ICARTI2021-Natural-language-processing-techniques-to-reveal-human-computer]
Text. Filename: Chiyangwa_etal_ICARTI2021_Natural_language_processing_techniques_to_reveal_human_computer.pdf
Accepted Author Manuscript

Download (1MB)| Preview


Designing systems for/with marginalized populations requires innovation and the integration of sophisticated domain knowledge with emergent technologies and trends. Researchers need to be cognizant of existing research trends when aspiring to design interventions to build on current and emergent needs. Traditional manual mechanisms for revealing developments in a field, such as systematic literature reviews (SLRs), cannot meet this challenge because they are time and effort intensive and the domain itself is dynamic and ever expanding. This compromises the efficacy of SLRs in keeping up with the growing academic literature. A number of emergent technologies and modern methods exist that could be harnessed to make it possible to monitor the field more effectively and efficiently. In this paper, we propose the use of natural language processing (NLP), an AI-powered text analysis technique that operates efficiently and requires limited human intervention. To investigate the use and usefulness of NLP for identifying research themes, we applied Latent Dirichlet Allocation (LDA), a topic modelling technique that uses a probabilistic model to find the co-occurrence patterns of terms that correspond to semantic topics. We applied it to a collection of 176 articles published in the Human-Computer Interaction for Development (HCI4D) field. We demonstrate the usefulness of the LDA method by comparing the findings of the LDA analysis to those of a manual analysis carried out by researchers. While NLP techniques may not be able to replace SLRs at this stage, we share some insights on how NLP techniques can complement SLRs to offset investigator bias and save time and effort in revealing emerging domain-related themes.