Being a minnow on Steemit can be tough, your well-crafted posts may get unnoticed and buried in the noise. Accordingly, I made a proposition to develop a new Artificial Intelligence, codename TrufflePig, to help the Steemit curation community by identifying overlooked and undervalued content. The best part of this proposition? I already have some data to show! I have experimented a bit and developed a prototype. Here, I want to give an update on the current development.
First, I have to make a confession, I made some rookie mistake in the prototype, mea culpa. There was some data leakage from the training into the test set. Oh boy, I did forget about edited posts that are published more than once on the blockchain. So a few posts could be found duplicated in the training as well as the test set. Correcting this had, of course, a negative impact on the test set score. It dropped from ~0.2 to ~0.1. However, the truffle posts the bot is digging up are still quite nice. I'll get to them later.
Besides this rather embarrassing Machine Learning Fauxpas, the Github repository is filling up with new code regularly as you can see here. Especially the part about gathering posts and filtering has seen some major progress. I hope to churn out the first working version that will post daily truffle picks (!) on Steemit and promote undervalued content soon. Hopefully, it will be finished in less than in a month. By the way, the bot already owns an account on Steemit: @trufflepig.
Did you know that there is a website, called robohash, that creates nice robot avatars by hashing strings? Here is the new face of my bot, be pleased to meet trufflepig17:
What else do we have? Well, I experimented a bit with some more machine learning models like Doc2Vec. Yet, apart from increasing TrufflePig's memory hunger, it did not yield any meaningful improvements.
My experiments were not entirely fruitless, because I picked up some nice truffles on the way. For example, did you know that there is a young couple, @lily-da-vine, with an amazing life story about fleeing from the US to Mexico because of pressing charges? Found them thanks to (still offline) @trufflepig :-)
Finally, after a first TrufflePig MVP living on some private virtual cloud server, I'll be working on a more specific roadmap. It will cover a lot of the feedback I got from the Steemit community, such as including @cheetah's blacklist. So stay tuned!