Project: AI in the Newsroom – Machine Learning for the Paywall of the F.A.Z.
- The aim of F.A.Z. was to obtain a data-based decision-making basis whether an article should rather be used for the paywall or as a free article for online advertising.
- jambit met the challenge with sophisticated natural language processing (NLP).
- jambit had to face a situation in which historical data was only available in a limited and unedited form. For further processing, jambit structured the data.
- Based on this historical data, jambit trained models using machine learning (ML).
- These models now predict different quality metrics, such as the number of subscriptions for new articles.
- An editor can now see from the recommendation in the CMS if the paywall for this article should be activated. It is calculated from the different quality metrics.
- Used AI methods: transfer learning, feature engineering and gradient boosting
- Used software technologies: BERT, PyTorch, scikit-learn, LightGBM, MLflow, Azure AutoML, Docker, Azure Functions
- The F.A.Z. won an AI solution that provides the editors with a data-based decision-making basis on paywall or free articles.
- The project achieved time and cost savings as well as quality due to an AI framework developed in-house by jambit https://github.com/jambit/sensAI. F.A.Z. benefited from know-how transfer from other jambit projects, for example in the automotive and Industry 4.0 sectors.
- AI experts, CMS developers and administrators from jambit worked directly together with an interdisciplinary team consisting of editorial staff, F.A.Z. Data Department and F.A.Z. Technical Project Management.
- Cost savings were also achieved through jambit's own GPU computing infrastructure and optimal embedding in the existing Azure Cloud ecosystem.