"Sentilyze", Twitter Sentiment Classification - UPDATE

[chargil]

"Sentilyze" Twitter Sentiment Classfication

Sentilyze on Github

Changes:

• (New-9/6/12) Fix Classifier Issues, Add Mutual Information Keyword Generation Function
• (6/26/12) Added Naive Bayes Sentiment Classifier
• Most things related to television shows were removed. Now any tweet can be sentimentally classified!
• Changed the name from "See,Save,Skip" to "Sentilyze".

What's included in Sentilyze?

• Tweet Sentiment Classifer (Keyword Counting and Naive Bayes)
• Tweet Language Classifer
• Keyword List Generation using TF-IDF
• Keyword List Generation using Mutual Information
• Keyword List Expansion using WordNet

Hope more people find this useful with the new changes.

Also I am working on an addition Sentiment Classifer using Naive Bayes. So if you're interested in that, be sure to watch this post for more updates.

[Durai]

Can you please post this code into GitHub? So whomever wants to build the further enhancements can officially pull and extend it.

[chargil]

I will get it into the hpccsystems/ecl-ml repository as soon as possible.

[Durai]

Thanks for the reply. I will look forward to get it from github.

[chargil]

You can now find Sentilyze on github in the 'Examples' folder.

[chargil]

I've added a Naive Bayes classifier to both github and the "Contributions" section on hpccsystems.com.

[chargil]

I've just uploaded a new version of Sentilyze to contributions. For those of you who are tracking Sentilyze on github, the updates should show up shortly.

• Sentilyze.KeywordCount.Generate Added Keywords_MI function (generates a list of words sorted by descending Mutual Information weights)
• Created "File String Defintions" sections which should ease the setup of Sentilyze
• Sentilyze.NaiveBayes: Fixed an issue with the wordbag vector
• Sentilyze.Language: Fixed ngram issue with Language Classifier (unnecessary bi and trigrams were being generated)
• Sentilyze.*: Standardized the input/output formats of all functions. All preprocessing functions and classification functions take and return Raw (ML.Docs.Types.Raw) record sets.