Tom Z Consulting

Here's the H1N1/Swine Flu map overlay(data from Google Maps) on top of the tweet sentiment map:

TweetSentiments.com now has thematic maps for Google Maps and Google Earth:
The color for each country represent its sentiment index.  Click on the country will display the sentiment details.

• Click on the above screen shots to go to the live urls.  And urls for Google Earth:
• http://tweetsentiments.com/sentiments_by_country.kml
• http://tweetsentiments.com/tweets_by_country.kml
• http://tweetsentiments.com/country_sentiment_ratings.kml

The sentiment index at http://tweetsentiments.com took two dives today, one with Farrah Fawcett and another with Michael Jackson. Below are some screen snapshots, notice the unusual amount of redness(negatives):

http://tweetsentiments.com has added OpenAmplify as the primary classifer for better sentiment prediction.  The OpenAmplify API performs NLP(natural languange processing) analysis on tweets and provides rich contextual results.  Right now we are only using the basic features of OpenAmplify and plan to take the full advantage of the API.

We have also added some new features:

1) Tweets on Google Maps -  you can now see where people are tweeting and in what sentiments.  And we will be adding Google Earth support shortly

2) Analyze This - give you the sentiments on a topic or for a user

3) Trending Topics - see the current popular topics and how people talk about them (color bars represent sentiments and degrees)

4) Top Countries - show the 10 most talkative countries (color bars represent sentiments and degrees)

And we've got a lot more features to come.

Tips: click on a photo thumbnail on the front or the analyze page will give you the sentiment analysis for that Twitter user.

Now the hourly sentiment index [0-50-100 most negative - neutral - most postive] is automatically updated at @tweetsentiment on Twitter.com.  Also the Top Twitting Country list and sentiments are broadcasted daily.  Below are the sample status updates, follow @tweetsentiments to get yours:

1. Current sentiment index: 59.72 Trending: up (Tweets positive: 49.41% negative: 29.97% neutral: 20.62%) about 1 hour ago from web
    
2. Top twitting country sentiments: United States 61.29 United Kingdom 61.05 Brazil 62.77 Canada 61.89 Japan 51.91 Germany 56.75 about 1 hour ago from web
    
3. Top twitting countries: United States 53.82% United Kingdom 8.15% Brazil 3.64% Canada 3.56% Japan 2.83% Germany 2.49% about 1 hour ago from web

If you use twitter, you might find http://tweetsentiments.com interesting.  It displays a scrolling list of recent tweets and for each tweeet you will find a sentiment icon/text.  The sentiments are predicated using a model built with the LIBLINEAR machine learning library, the model's precision ranges between 70-90%, and can be improved with better training data.

This is just a taste of a lot more things analytic to come at http://tweetsentiments.com(built with Ruby on Rails).

The Ruby interface to the LIBLINEAR machine learning C++ library is available at GitHub http://github.com/tomz/liblinear-ruby-swig/tree/master.

LIBLINEAR (www.csie.ntu.edu.tw/~cjlin/liblinear) can be used for very large scale linear problems like text mining, it is very fast - much faster than LIBSVM(Suport Vector Machines).

I used it to build a decent sized text classification model(with 10 classes and 10000 features) using 100,000 text records (through ActiveRecord in a Rails app) as training data, the recall is 99% and precision is around 90% on 2000 test records. The same model was tried with arbitrary and web page texts with pretty good results.

Here's a Rails app that displays recent tweets and later will display the sentiments for each tweet as well using machine learning techniques (LIBSVM): http://tweetsentiments.com

Here's a live installation of the Insoshi open source social networking application, you can switch to any one of the several dozen Google Translate supported languages(e.g. Chinese, Japanese, Korean, French, Spanish, Arabic, Hebrew, and etc) on the fly.  Obviously, machine translation is not suitable for production use, but it does allow you visualize and present your app in different locales for demonstration purposes

http://insoshi.tomzconsulting.com  (Rails 2.3.2/thin/I18n/Google Translate)

Here's a Ruby on Rails application that will automatically collect the latest newswire items from the New York Times using their newly published Newswire API:

  http://github.com/tomz/nytimes_newswire_api_app/tree/master

The Ruby Interface to LIBSVM (C++ implementation of Support Vector Machines for classification and regression) is now available at GitHub:

    http://github.com/tomz/libsvm-ruby-swig/tree/master

To install it as a gem for easy use within Rails:

  sudo gem sources -a http://gems.github.com   (you only need to do this once)
  sudo gem install tomz-libsvm-ruby-swig

Finally a complete Ruby inteface to the LIBSVM C++ library.  I've been frustrated by not having a good Ruby SVM interface/implementation so I decided to roll my own.  I studied libsvm-ruby which is a direct Ruby to CPP interface, the SWIG Python interface(part of LIBSVM distribution), and also the SWIG Ruby interface to Xapian.  I ended up picking up some Python and converted the SWIG Python interface to Ruby.  So far all the test code ran successfully and generated the same results as the Python version, once I get a chance to clean the code up a bit, I will upload it to GitHub.

Here's a modified version of the Rails i18n demo enhanced with Google Translate. The missing locales are automatically translated using Google Translate, this is nice to have in the development environment - no need to waste time to manually translate the pages into different languages.  The oringinal Rails i18n demo apps can be found here.

Just installed and tried Liferay 5.2, all I have to say is wow!  Very nicely done, see it here and here. The sample data and sites/pages demonstrated the social networking capabilities really well.  Can't wait to migrate my site from 5.1 to 5.2.

One nice feature I like about Liferay 5.1.x is that you can share portlets on other websites.  Here are some examples:

Blog

Message Boards

1. download the instant client files: basic.zip, jdbc.zip, sqlplus.zip, sdk.zip
2. unzip those into /opt/ -> /opt/instantclient_11_1
3. export LD_LIBRARY_PATH=/opt/instantclient_11_1 or 
   add /opt/instantclient_11_1 to /etc/ld.so.conf and run ldconfig /etc/ld.so.conf
4. export TNS_ADMIN=<directory where tnsnames.ora is in>
5. export PATH=$PATH:/opt/instantclient_11_1
6. cd to /opt/instantclient_11_1 and run ln -s libclntsh.so.11.11 libclntsh.so
7. sudo gem install ruby-oci8, and now your Rails apps are Oracle enabled