Topic models in contextual advertising

Follow the full discussion on Reddit.
I wrote a blog post about how my team and I at Schibsted, use probabilistic topic models, i.e. Latent Dirichlet Allocation (LDA), in the scope of contextual (cookie-less) ad targeting. We use the models to both suggest the most relevant keywords, i.e. keyword expansion, during ad segment creation, and also as a means of quickly getting the overview of our news media content, and its engagement. It's worked beautifully for us, in terms of helping our product specialists quickly construct highly relevant ad content. I hope there are enough details in the blog post to both introduce topic modelling to wider audiences, but also provide some techniques and lessons learned for those looking to deploy similar approaches into production. https://medium.com/p/2bc56d624c5f