Predicting innovative firms using web mining and deep learning

Evidence-based STI (science, technology, and innovation) policy making requires accurate indicators of innovation in order to promote economic growth. However, traditional from patents questionnaire-based surveys often lack coverage, granularity as well timeliness may involve high data collection costs, especially when conducted at a large scale. Consequently, they struggle provide makers scientists with the full picture current state system. In this paper, we propose first approach on generating web-based which have potential overcome some shortcomings indicators. Specifically, develop method identify product innovator firms scale very low costs. We use firm-level survey (German Community Innovation Survey) train an artificial neural network classification model labelled (product innovator/no innovator) web texts surveyed firms. Subsequently, apply hundreds thousands Germany predict whether are innovators or not. then compare these predictions patent statistics, extrapolation benchmark data, regional The results show that our produces reliable has be valuable highly cost-efficient addition existing set indicators, due its coverage granularity.