I tested my first Machine Learning Classifier.
I'm excited to create my own Machine Learning Classifier, it produce above 90 percent accuracy. The more training and test data, the better prediction (accuracy) of results.
Resources I'm using to study Machine Learning.
- Video tutorial hosted by Josh Gordon. I highly recommend that you watch all the videos from the beginning if you're interested to know more about Machine Learning
- Python for programming language.
- SciKit Learn and Tensor Flow for Machine Learning framework.
Here's the Classifier code.
and the remaining code to test the result.
I'm hoping to add Machine Learning to filter out bad content, spammers, abuse use of bid bot and malicious users in https://Steeming.com website. It's a long way to go but something will happen to make a better experience of browsing the steem blockchain.
Source: https://que.com/machine-learning-my-own-first-classifier/
I can also use Machine Learning to improve my upvoting selection of Authors with good content. The current algorithm is working just fine, an improvement to produce quality content is better for the community in the long run.
and by the way ...
If you like what I do helping our community. Please Vote @YEHEY as your Witness. Your Vote is highly appreciated.
Go to https://steemit.com/~witnesses URL address then scroll down, type "yehey" and Vote.
I created a short URL to make it easier to vote, using this link https://on.king.net/witness simply click and vote. This will redirect to Steem Connect for secure authentication.
Thank you,
@Yehey
Let's steeming for a better tomorrow. Please UPVOTE, RESTEEM and FOLLOW me to keep you posted about our progress.
This is a noble goal.
Unfortunately it comes with an inherent problem, one that is perhaps nonintuitive to people who have not been tinkering with machine learning for a while and it runs thus…
How do you expect to get the training data for any kind of machine learning system on Steemit content?
This is not a rhetorical question. I mean that literally. In order to train any kind of algorithmic learning system to recognize types of content, you have to be able to feed the algorithm with a training set. That means a rather large corpus which is already tagged and classified. Generating that large corpus in the first place is, not to put it to finally, "nontrivial."
If you have an extant algorithmic system (non-learning) which can classify that content, you don't need a learning system. If you don't have an extant algorithmic system which can classify that content, all of it has to be done by hand – and then you're back to square one.
Wondering where the training data comes from leads immediately to trying to figure out how you define what is acceptable in that training set. "Bad content?" Humans, many of which are active users, can't even agree on what that looks like, much less whether it has a discernible pattern. "Abuse of bid bots?" The same. "Malicious users?" If ever there was a gateway judgment to ending up with a content filtering system guaranteed to lead to an echo chamber, it's that.
Machine learning systems are really hip, trendy, and buzzword compliant – but they have a an extremely limited field of application for systems which are intended to ingest a lot of content and excrete something useful. More "traditional" self-refining systems like Bayesian classifiers and the like can be much more useful, and instead of needing gigantic corpora of training material can be deployed with iterative feedback loops from a user in order to help sift content based on their own judgment rather than someone else's.
As tools which would go a long way toward solving problems on Steemit (and in social media in general), that's probably a far more productive line of inquiry.
Good gosh! AI is ruing the world one day! just like in the film!!
Some people are really smart, how could some brains think in codes?!