The Analytics Team did a lot of good work this semester for the project. They used various algorithms to collect data on jokes and attempted to create their own using machine learning computation. They sent out multiple class surveys to the entire class in an effort to recommend similar jokes to users. To do this, the team used Item-Item collaborative filtering and User-User collaborative filtering. One of these looked at the similarities between different jokes and the other looked at the similarities between different users’ interests.
The analytics team used a neural network on a collection of Reddit jokes, referenced at http://karpathy.github.io/2015/05/21/rnn-effectiveness/. It trained for about forty hours and the results that it gave were humorous. Here’s the joke:
“My wife picks up his own finger. One of them is spending the church. “Go immediately pointing,” and every morning “I have a honeymoon. “How can I hear?”.”
As you can see this joke makes very little sense, but it was interesting to see how a neural net algorithm could work to generate new material. In upcoming semester, the team plans to tune the algorithm with different identifiers to yield better results.