Twitter as a potential goldmine : a data-driven product innovation approach in the wearable device domain based on machine learning

The recognition, assessment, and integration of customer feedback is an essential component for businesses to design their products and services appropriately. Although commonly applied methods provide the necessary information, they often require extensive manual labor, lack automation as well as scalability, and hence, are not suitable for ongoing use. As an alternative for identifying customer needs, this paper investigates the reliability of the BERT model, a machine learning algorithm that uses a trained neural network to classify a given dataset. Prior research from Kuehl et al. (2016) already adopted a similar approach within the e-mobility domain. This paper expands on the efforts of Kuehl et al. by applying the approach for the wearable device domain. In total, 10,000 tweets – short messages retrieved from the social media platform Twitter – were manually assessed, from which 130 ‘need tweets’ were identified. The needs were categorized into eight different groups and provided relevant product development ideas and inspirations. Furthermore, an analysis of the BERT model showed that the prediction and classification of the tweets are inconsistent and unreliable. Further research in this domain with adjustments in the dataset and expanding into other domains are suggested.