Performing Content: Can NLP and Deep Learning algorithms predict reader preferences?

User engagement is one of the most important factors to steer and optimize the content delivered to customers, e.g. readers of online newspapers. Being able to predict this KPI for new, not yet published articles is a key factor in process optimization and state-of-the-art tools to assist editors in their daily work flow.

The talk shows that standard onsite tracking KPIs can be used to reveal driving mechanisms to model and predict reading times for articles published online. It will be discussed how statistical (GAM) and NLP related deep learning algorithms (Transfer Learning using BERT) can be combined to distinguish different driving mechanisms. Further insight about text characteristics and content that increases reader engagement is derived by applying Explainable AI (XAI) techniques in combination with less transparent machine learning approaches (XBG).