Show HN: Toxicity Analysis of Twitter Handles

twictator.herokuapp.com

3 points9notorp3 comments

I only started learning/implementing ml very recently. Gave up halfway through my previous attempt when i got to know about the compute intensive nature of ML implementations

About this app, (django,react,tfjs) It takes a twitter handle as input and lists the most recent tweets along with buttons for each to predict toxicity using tensorflowjs's toxicity model (pretrained) I plan to make a twitter bot with my learning so far. Also this project initially had a much larger scope and I had to cut down/scrap most of it as i'm using only freely available services to host this and running ml on such services did not seem fair to me.

Good job! although it would be really useful if you reduce the computational load and then simply analyse the last10 tweets. Itwould be more useful.

Thank you! Not sure if I can do that unless browsers go back to the old designs and make ipc possible among tabs/processes again, the compute happens on the browser so it entirely depends on the end user hardware config. And while making this I gave that approach a try as well but decided against it, since the (compute requirements of analysis) entirely depends on the size of input. On larger strings it takes considerably longer and has rendered the browser completely unresponsive on low end devices. Also, this is far from complete and the pre trained model is fairly inaccurate. I am mainly looking for collaborators to train this better.

Reply on news.ycombinator.com