The Potentials and Pitfalls of Computer Visioning and Machine Learning Methods for Communication Researchers

This paper will discuss the potentials and pitfalls for incorporating computer visioning and machine learning in technical communication research. We share the results of a case study our research team engaged with in our attempt to demonstrate an affordable, easy-to-use machine learning method that can be formalized so that researchers who are interested in working with bigger datasets will have a road map for doing so. Using standard computing power and the insight of a graduate student with experience in computer visioning, we engaged in a process in which we tried to achieve a balance between accuracy and the time it would take for the system to process images. We achieved 99.2% test accuracy on correctly identifying visuals that were not graphs. Yet this optimization required a tradeoff in correct identification of what was a graph, with a success rate of under 4%. We reflect on the conclusions from our research, suggesting that successful implementation of computer visioning tools for technical communication research requires more technical and chronological planning than we had hoped, but it can still achieve things that are impossible to achieve through individual or even team-based human computation. We discuss what communication researchers should consider when processing massive datasets and what some of the limitations of these methods are.