Abstract
AbstractMeasuring innovation accurately and efficiently is crucial for policymakers to encourage innovation activity. However, the established indicator landscape lacks timeliness and accuracy. In this study, we focus on the country of Mauritius that is transforming its economy towards the information and communication technology (ICT) sector. We seek to extend the knowledge base on innovation activity and the status quo of innovation in Mauritius by applying an unsupervised machine learning approach. Building on previous work on new experimental innovation indicators, we combine recent advances in web mining and topic modeling and address the following research questions: What are potential areas of innovation activity in the ICT sector of Mauritius? Furthermore, do web mining and topic modeling provide sufficient indicators to understand innovation activities in emerging countries? To answer these questions, we apply the natural language processing (NLP) technique of Latent Dirichlet Allocation (LDA) to ICT companies’ website text data. We then generate topic models from the scraped text data. As a result, we derive seven categories that describe the innovation activities of ICT firms in Mauritius. Albeit the model approach fulfills the requirements for innovation indicators as suggested in the Oslo Manual, it needs to be combined with additional metrics for innovation, for example, with traditional indicators such as patents, to unfold its potential. Furthermore, our approach carries methodological implications and is intended to be reproduced in similar contexts of scarce or unavailable data or where traditional metrics have demonstrated insufficient explanatory power.
Funder
Technische Universität Berlin
Publisher
Springer Science and Business Media LLC
Subject
Economics and Econometrics