I was putting together some ideas for an alternative algorithm last night. There would be a personalization layer that would make the bulk of it - say 70% but there would also be a serendipity layer that would try to identify notes that have higher interaction than normal while also not having so much interaction as to already be considered trending. Most of my professional work revolves around algorithms for outlier detection so it would be based on the same concepts. I thought it would be fascinating to try these algorithms in a completely different context than normally deployed