You are viewing a single comment's thread from:
RE: Step 2 on developing UA: Calculating UA scores from a MongoDB subset
An algorithmic deterrent/solution to spam is a great idea and I hope it goes forward but @lextenebris raised the point that we need to create a snapshot and ACTUALLY see/test how the algorithm performs. First, can we identify the bots, and secondly, how will it impact users? Mathematical models are by definition cleaner than the real world in which they must perform. (Just ask any physicist ;-)
I would like to see a public tool that would let any member check to see the REAL impact of the algorithm ON THEM. And then maybe a vote!
What are the odds of getting some such tool?
As I've said multiple times before, UA is not intended as a stand-alone tool. It is a derived metric from the already existing follower data, it could be used combined with other metrics to detect spam (very small UA && very high post frequency && very high post rewards: accounts having that profile, could very well be "reward-grabbing self-upvoters"). Therefore note that UA is not just intended as an anti-spam tool, it can be used for all kinds of things. But regarding post curation, look at it as a "mathematical assistant and extension of human curation": it's not just algorithmical, it's an extension of human activity. Like a Word processor is to manual handwriting, and a CMS is for publishing website content, and a car is for transportation.
@yulem, we are still developing code to implement UserAuthority (UA). This is post #2 about that development, we're not done yet. As this article shows, I created a GitHub repo. Snapshot all you want with it. Because of the large Steem blockchain datasize, it's impossible to begin development on the entire dataset, because how should we identify if the results are correct or if the code contains a calculation bug? Therefore, we're first developing code using my test / example matrix and graph. Once that works completely, we will optimize the algorithm for faster / more efficient performance on the real-life Steem dataset.
The development roadmap looks something like this, from here on:
Nota bene: Another great "public tool" would be a live interactive followerGraph where an account's UA-score is visualized by node size.
Feel free to collab with for example @lextenebris to develop such a public tool. I would resteem development posts on that with much enthusiasm!!
Thanks for your thoughtful reply. I do get where you're coming from and fully support you in that approach with its associated goals.
Maybe "tool" is a bad word choice.
Steemit is a select microcosm of humanity and as such we are embarking on a very noble experiment in its governance. Consider each hard-fork as an amended constitution (of which UA should probably play a role) which governs us all. Having witnessed other such experiments and seeing them fail miserably for a variety of reasons I'm no doubt overly sensitive to the unanticipated ramifications of even the most seemingly insignificant acts. Hence, my tendency to simulate before deployment.
I'm afraid my development days are behind me but I get your point. I fully trust the future into your younger and capable hands.
I might be available for beta testing or QA ;-)