Rate my startup: KeyTweet, a Twitter client that learns what you like | Hacker News Reader

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

keytweet.com

27 pointskeytweetlouie27 comments

We developed KeyTweet to deal with the overwhelming number of tweets in the average user’s feed What KeyTweet does: We find keywords distinct to who you follow then sort your timeline based on those keywords. We drop tweets over 24 hours old and ones without links. You're left with what we think you'll like best.

My business partner and I have been following hackernews since we’ve started on this project. We’ve both seen some great advice given to new startups here. Please feel free to say anything that comes to your mind, small or large. Thank you for your help.

Open articleView on HN

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 the left side is how relevent we feel that tweet is to you. Enjoy!"

It's confusing.

So my advice is for you to have somebody review all your copy.

edit: my comment might sound very dry, so let me add this: I think it's a good idea if done well, though I follow a small enough number of people that I might not need it yet.

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.

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. You seem to strip out the Icelandic characters from the stream. (A very common pitfall amongst english-speaking developers.)

ps. You need a designer.

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

If it works, you've solved the inbox problem. Pass go, collect $200. Suggest to wrangle a web designer to give your page a once over.

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"

I like it. It worked well for me and solved a problem I had last night. It might be that I don't know how to use twitter well. I have too many people I follow and there's too much activity now. I just want to quickly scan to see what's 'new and interesting' from the people I follow. It seems like this does that.

Request: Would it also make sense so that I can see the last tweet from everyone on my list or does something like that already exist? It's not really 'key tweet' at that point though.

ps. I don't mind and appreciate the simplicity of the design. I didn't read the copy though.

I see no screen shots, just promise of magic. You need to put forward something that will give the potential user something to go with before signing up onto your system.

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.

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.

Very nice! I played around with something similar. http://www.tweetspeedreader.com A technical explanation is on my blog post: http://www.carlosjustiniano.com/experiments-in-natural-langu...

It doesn't seem to handle Unicode very well, or your normalization routines need love. I have several German words in my topic list, and Umlauts get chopped out, for example the infamous "Blumenkübel" becomes "Blumenkbel", "für" becomes "fr", and so on.

Definitely agree with rjurney that you need to eventually update the homepage, but sweet idea for an app. I've signed up, but I'm still waiting to see how it actually works out my tweetstream

try hitting this url again http://keytweet.com/search/<your twitter name here>

I suggest pitching it towards people who'll be prepared to pay money for this: PR, and marketing folk who want to keep a tab on a lot of twitter feeds for example.

Call to action should be more obvious--maybe a big red Try It button below the copy. I finished reading and had to look around for how to actually try it out.

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.

Love the relevance so far! keep it up!

Thanks! We'd love to hear about any ideas to make it more relevant.

"Pandora Radio for Twitter"?

fyi, blog link isnt working for me right now. no page returned.

It's back up

KeyTweet's blog is on Posterous which had a DoS attack.

Reply on news.ycombinator.com