The New York Times, as reported on their blog have had several iterations, when creating their recommendation engine for the 'next best article'. The first iteration which was a 'content based filter' used a keyword tagging approach, which showed article recommendations using the associated topic it was classified under during the last 30 days from the date that article has been published. A drawback of this approach is that if multiple tags are used, then the potential recommendations can be a lot broader and potentially not what a user might be looking for. The second iteration was called as a 'collaborative filter' which used readers history between common users to determine the next best article. However, this system was unable to include new articles in its recommendations as a result. The New York Times ended up using a Collaborative Topic Modeling technique (developed by Princeton University Graduates), that (1) models content, (2) adjusts this model by viewing signals from readers, (3) models reader preference and (4) makes recommendations by similarity between preference and content. Their blog post provides pretty useful examples - https://open.blogs.nytimes.com/2015/08/11/building-the-next-new-york-times-recommendation-engine/. What are your thoughts on this approach? Have you seen any other examples?
Reply