Machine learning classification tool for innovation projects
By generating a model that was able to extract theory-based attributes from the data and connecting this data to an interactive and robust dashboard, new opportunities on working on this data, like showing trend analyses for portfolio management, become available for further research on the promising possibilities of machine-based classification of innovation descriptions.
Abstract
Different innovation typologies are used for classifying innovation projects for portfolio management. Classifying and prioritizing these projects is a time-consuming process and an inconsistent use of innovation typologies makes it hard to compare company’s innovation portfolios and academic literature. This study researched the opportunities for machine-learning based classification of textual project descriptions. In order to train the machine-learning model, samples were manually labelled with five suitable innovation typologies; exploitation – exploration, product’s architectural newness, product’s component newness, market pull – technology push incentive and business-to-business and business-to-consumer market. Compared to a comparable previous study, this study did not achieve a result on high accuracy but found the variables that influence the performance significantly. Thereby, a model was generated/trained that was able to extract theory-based attributes from the data and by connecting this data to an interactive and robust dashboard, new opportunities on working on this data, like showing trend analyses for portfolio management, become available for further research on the promising possibilities of machine-based classification of innovation descriptions.