Antons, D. and Breidbach, C. (2017). Big Data, Big Insights? Advancing Service Innovation and Design With Machine Learning. Journal of Service Research, 21(1), pp.17-39.

From Digital Culture & Society

This article is one example of how machine learning can be used in the work-force – the researchers integrated the automation into research that would have taken them years to do. This streamlined the process for them.

Service innovation and service design is a sector that is continuously evolving, making recording processes that work within the industries hard due to the constant adaption and increasingly complex processes. Antons and Breidbach identify 3 major contributions that they wanted their research to cover.

1. Identifying 69 research topics, assessing their relevance in conjunction with the rest of the research, and estimating the trajectory that each of those 69 topics would take in terms of research possibilities.
2. Integrating 4 different research directions, providing 28 research questions and 12 research mechanisms for future researchers to employ in their own research.
3. Demonstrate the applicability of topic modelling machine learning to service research.

In their analysis of the previous research, the authors made four key observations. The first is that the majority of the research wasn’t theoretical, but rather phenomenological. This means that the research had been done on actual events the researchers underwent; not thinking about where the actual technology could go beyond what is already within our means. Secondly, Antons and Breidbach discovered that while all research done on service innovation applied to service design, the service design topics were very scattered and not connected to each other. Third, the age of the industries accounts for the forms of research having been done on each topic. Service innovation is an older category of research, so it has more quantitative research in its ranks. The younger category, service design, relies on qualitative research and conceptual methods. Finally, the authors also discovered that the topic network between the service innovation research and service design research is very loosely connected – there are not many overlapping topics that build off of each other’s research. Finally, the authors called for more cohesion between the two industries. The service design field is more likely to submit their articles to books, rather than journals, leading to a discrepancy of information available in journal databases.

The ability to use machine learning to analyze an entire set of data is still in its infancy stages, but this article was still able to analyze all of the research in a specific set of fields, which would have been substantially more difficult to program into the machine if these hadn’t been newer, under-developed research fields. The authors were able to point out gaps in the information, where there was research missing and would be needed in order to continue on in other subjects or topics. They also showed the potential trajectories that the current research could continue on in, giving current and future researchers a clear direction to set their sights on. This research will provide an important foundation for the two industries, service innovation and service design, continue to expand and will eventually collide. Antons and Breidbach were thorough in their research, using a machine learning model to extract a data set of 641 articles for them to analyze, which they were able to cut down considerably.

In terms of how they collected the articles to analyze in the first place, they set up a machine learning model to search for articles with the following keywords; service innovation*, innovation in service*, service design*. Where they differ from other articles on this topic is that they did not search for new service development* due to growing evidence that that idea has very little connection to the service innovation and design industries. They then inputted a series of algorithms and packages into the machine learning system that would reduce the documents down to their basic word text, sort through the words and consolidate the data for them to work through easier. From there, the work was manual to classify the data correctly. This is an instance where the technology was not available yet in order to fully do the task – also known in machine learning as supervised learning.

Megan McGuire