The media content curation workflow is central to a MediaMiser Enterprise user. Every day, our media analysts are required to select a handful of relevant media items buried in the enormous amount of online and traditional content. Productivity is key and content relevancy is king.
The current approach to efficient content curation mostly relies on keyword based-filtering. Basically, we punch keywords in a full text search engine using boolean expressions such as “macaroni AND NOT cheese”.
But could media monitoring also leverage collaborative filtering?
I recently stumbled upon a blog post about how The New York Times built their recommendation engine. The analogy to content curation is very strong, in my opinion. The selections that a media monitoring professional makes on a daily basis can certainly be framed into a mixture of topic modeling and collaborative filtering.
The New York Times recommendation engine
So, you’ve been reading The New York Times for years, maybe more recently on your tablet, phone and computer. The New York Times is and always has been your favourite newspaper for the best articles and content. But don’t you find it frustrating to sift through articles that are irrelevant to you so that you can find the ones you are most interested in? Thankfully, The New York Times is making use of a recommendation engine based on Collaborative Topic Modeling so you see relevant and recommended articles first.
Collaborative Topic Modeling is a mixture of textual information retrieval and collaborative filtering. The New York Times has been hard at work creating a clever algorithm whereby it analyzes the articles you have read in the past 30 days. Based on article keyword tags, the algorithm will recommend articles which it believes will interest you. This is bound to succeed because if, for example, you read 15 articles with the tag “Trump”, you will likely want to read similar articles in the future.
This recommendation engine works well in most cases and can be a really great approach. However, keyword context changes and word frequency can quite often lead to an incorrect relevance score.
This is where collaborative filtering kicks in: if other people are actively reading the same news you are interested in, then you will likely be interested in other news they read.
A savant mix of the two approaches was the key to creating a powerful recommendation engine at The New York Times.
Collaborative media monitoring
At the end of the day, we are always looking for ways to lower the cost of media monitoring to our clients while increasing the quality and relevancy of their daily briefs.
How can we leverage and benefit from the content curation work performed by multiple analysts? Could analysts working in different accounts in the same industry benefit from collaborative filtering? Or could we look at media items shared by thought leaders and influencers in one client account and recommend them to other analysts?
The future of media content curation should not be limited to boolean expressions and keyword matching. It would benefit from a collaborative approach.