Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Rate my startup: KeyTweet, a Twitter client that learns what you like

keytweet.com

Post reply on HN
Print view View original

21–30 of 31 posts

Re: Rate my startup: KeyTweet, a Twitter client that learns what you like

#21

Post by jamesk2 »

Timothee wrote: post #11

Quick copy comment: I don't understand what this means on the frontpage: "We find the keywords distinct to who you follow" On the feed page, "Why am I seeing this?" shows this: "How keytweet works: When you click like or retweet. We try display tweets with similar content. If you don't rate or retweet. We build a list of popular topics in your feed. We then try to match tweets from your feed to those topics. The # on…

Thanks for the suggestions, we'll definitely have more people review the copy.

And you're right, people who don't follow very many people don't need this app and people who are on all day long don't need this app. But we suspect there are many people who do need it.

Re: Rate my startup: KeyTweet, a Twitter client that learns what you like

#23

Post by jamesk2 »

pedalpete wrote: post #22

Show us the magic! Before logging in, I'd like to see this work on a live twitter stream. I suggest taking somebody with a huge number of followers and running your algorithm on their stream and then showing how your tech works.

Sure, try this out:

http://keytweet.com/celebrity/categories/ Let's you see top tweets for other people. We've got one for "The top tweeters from hacker news"

Re: Rate my startup: KeyTweet, a Twitter client that learns what you like

#25

Post by jamesk2 »

kingkilr wrote: post #12

So I signed in, and it basically showed me a bunch of tweets I've already seen, since they're all from people I follow.

We sort the tweets you'd normally get by relevance rather than chronologically so if you are actively looking at every tweet already then you'd have already seen the tweets we recommend.

It's more useful if you follow a lot of people and don't have time to read every single one.

Re: Rate my startup: KeyTweet, a Twitter client that learns what you like

#27

Post by jamesk2 »

arnorhs wrote: post #10

This actually seems like a usable concept. Too often similar software tries to magically predict stuff in a not-so-transparent way. Your decision to focus on links is, I believe, right on. Because you're not trying to replace my twitter stream with your useless predictions, but you're trying to help me notice the most important links, I think this could work. However, you're doing something wrong with your encoding.…

Yes, we need to pay attention to our international users. Unicode is too much fun to ignore. :-)

Re: Rate my startup: KeyTweet, a Twitter client that learns what you like

#28

Post by keytweetlouie »

protomyth wrote: post #18

so I got the following keywords: "comments, prop, wave, link, google, democrats, photoshop" - Today is a poor day for this or I am really having a hard time with how I would use this. I do think it's a great idea, I just think the topic algorithm needs to look at multiple words and ignore some others.

Thanks for sharing your top words. We are looking into using bi-grams for the top words. When you say "look at multiple words" are you referring to how google and wave are not together?

Re: Rate my startup: KeyTweet, a Twitter client that learns what you like

#29

Post by protomyth »

keytweetlouie wrote: post #28
protomyth wrote: post #18

so I got the following keywords: "comments, prop, wave, link, google, democrats, photoshop" - Today is a poor day for this or I am really having a hard time with how I would use this. I do think it's a great idea, I just think the topic algorithm needs to look at multiple words and ignore some others.

Thanks for sharing your top words. We are looking into using bi-grams for the top words. When you say "look at multiple words" are you referring to how google and wave are not together?

Sorta. I was more concerned about "prop". I would assume it is about "prop 8", since there seems to be a lot of traffic on that. "Comments" and "link" are too generic words and probably should be removed.

I guess I would say you should be looking for common word sequences.

Re: Rate my startup: KeyTweet, a Twitter client that learns what you like

#30

Post by ABrandt »

IIRC, paulsingh created something similar (Philtro?) a while back. The focus was a little bit different--it acted as a spam filter--but I think you could definitely benefit from his knowledge. Good luck!

EDIT: It appears philtro was acquired back in 2009. He may not be able to speak freely about the service because of it, but that does certainly help validate your market.

Post reply on HN