Show HN: My Twitter brand monitoring side project | Hacker News Reader

Show HN: My Twitter brand monitoring side project

recon.io

39 pointsilldave27 comments

Hi HN - I've been working on a Twitter brand monitoring service that alerts you whenever anything interesting is tweeted about your brand. It analyses each tweet, so it can tell you when a user asks a question, when you get good & bad feedback and when a journalist tweets about you (amongst other things).

I'd love to hear your feedback.

I clicked on the demo and then on the positive tweets and the page that opened was just too huge, add pagination or date grouping or something. My firefox exploded. Also 2 scrollbars in windows/firefox14.01 http://i.imgur.com/9QdZE.png . Also how are you filtering the tweets, do you get every single tweet from twitter or can search for certian tweets containing keyword?

Thanks for the feedback and the headsup on the CSS bug, I'll get that fixed. Pagination sounds like a good idea, or reworking how it fetches the data from Twitter.

When a campaign is created it queries Twitter's API for the brand name (e.g. "Hipmunk") every 10 minutes, and then runs filters on each tweet to categorise it.

I love the simplicity of it. Don't be compelled to succumb to feature creep. There is already plenty to make people pay five bucks or more a week for this. The worth to movie, music, and videogame producers alone.

I guess you could add some mobile notification mechanism, but I already love the product as it is right now.

As far as I am concerned, you might as well charge twenty bucks a month for it. It's a great product.

Thanks very much! I completely agree with keeping the simplicity of it - I'd much rather focus on having it do one thing really, really well rather than lots of things "just ok". Really appreciate the kind words, thanks again.

Signed up with one of my brands. First of all, I have to say great design! The sign up process was easy and getting started by adding a campaign was simple. I can't comment yet on how well the service works but overall it's been a good impression.

Thanks for the kind words - let me know how you get on with it (dave@recon.io)

How are you determining if a tweet is positive, negative, or neutral?

It assigns a sentiment score based on a large bag of words. I'm making improvements to it each day, and it's slowly getting better. Terrible at picking up sarcasm though, but I think most sentiment analysis tools are.

The Twitter API lets you search by mood, I would assume it's just passing those along.

AFAIK that is an emoticon based thing, not real sentiment analysis.

It does? Never heard of that before.

I recorded a beta test for recon.io: http://brinx.it/fqA

Hope you enjoy the feedback.

This is awesome, thanks very much - BetaPunch looks incredibly useful.

Looks really awesome.

Just a little note. In your demo one of the negatives is misclassified which shouldnt really happen in a demo :P

Aha, thanks - good catch. Fixed!

Very cool, and great design! Seems you filter out tweets that are not in english? Is that provided by the Twitter API as well?

Thanks! Yep, at the moment I've set it to only pick up English language tweets but I could potentially roll out other languages at a later date.

Interesting twist. Excellent execution. Great MVP.

What sentiment analysis API do you use?

I guess he's using NLP (Natural language processing) for that.

He asked what API, not what field of techniques. NLP is comprised of many different things, all different techniques and observations and not concrete code implementation of them. I for one am also interested if he used some ready API that implements some NLP techniques.

This is the same response that several business development people I know would give. "NLP" is treated as a black box that turns words into useful output, hooray!

It's kind of sad that the two comments asking about how this works under the hood are at the bottom. A bag-of-words model -- any n-gram model, by extension -- is an entry-level approach: it has decent performance but won't win any awards for how it handles complex language structures (and you'd be surprised at how complex 140 characters can be). Hence most commenters are praising the UI.

Good luck moving forward with this. If your model learns to handle negation intelligently or starts to use morpho-syntactic features, I'd be glad to chat with you.

How is this new? Similar services have been around for years already.

This doesn't just alert you anytime someone mentions your brand (like a Google alerts style service might) - this analyses tweets to identify mentions that you might want to act on, like tech support issues, feature requests and (y favourite) when a journalist mentions you.

But that too already exists... I wanted to built exactly this ~6 months ago, but I didn't because there were already so many competitors back then.

Just because a service exists does not mean that someone else shouldn't launch something better. I love the simplicity of the service (reminds me of buffer that does one thing really well).

Reply on news.ycombinator.com